Veeam Software Group GmbH used VeeamON 2026 in New York City this week to punctuate its shift from “the backup company” to a data and artificial intelligence trust platform for the agentic era.

With a new architectural layer and an aggressive product roadmap, Chief Executive Anand Eswaran (pictured) and President of Products and Technology Rehan Jalil are betting that the next decade of enterprise infrastructure will be defined less by how quickly you can restore a virtual machine or data set and more by how confidently you can let AI act on your data.

From backup vendor to trust layer

For most of its 20-year history, Veeam has been synonymous with backups and fast recovery, to the point that “instant recovery” became part of the company’s identity. Eswaran reminded the VeeamON audience that Veeam earned its leadership by reducing customers’ RTOs from hours to about two minutes and by building “the broadest workload coverage on the planet across virtual machines, physical, hybrid multicloud and SaaS.” That foundation now protects more than 550,000 customers in more than 150 countries, including 82 of the Fortune 500, and drives more than $2 billion in annual recurring revenue.

In Eswaran’s view, what has changed is not the importance of recovery but the nature of the threats and the actors who access enterprise data. He framed Veeam’s history as three eras: traditional backup and recovery (“assume restore”), cyber resilience (“assume breach”), and now the agentic era of AI (“assume autonomy”), in which nonhuman identities operate at a scale and speed that existing tools were never designed to govern.

Defining the agentic AI problem

At the core of Veeam’s pivot is a view of how AI is deployed across large enterprises. Veeam’s research and telemetry indicate that autonomous AI agents already outnumber human employees by 82 to 1 on average, representing more than 250,000 non-human identities per organization. Even more concerning, Veeam reports that 97% of those agents have excessive privileges, dramatically expanding the blast radius of a single compromised or misconfigured agent.

Eswaran argued that legacy security architectures implicitly assumed “the actor was human,” and that assumption “just fundamentally broke instantly” when agents began accessing ERP, CRM, warehouses, email, files and SaaS systems in parallel. In that world, a new failure is “not just a breach, it’s a wrong decision executed at machine speed before anyone notices” — a point underscored by examples such as an AI agent deleting a production database and its backups in nine seconds, or autonomously recreating a cloud environment and triggering a 13-hour outage and millions of lost orders.

The missing layer in the AI stack

Veeam claims there is now a “missing layer” in the AI stack, positioned between data platforms and models. The solution is a unified data and AI trust layer that treats data, identities, access, regulatory posture and resilience as a single system. “The infrastructure to deploy AI exists,” Eswaran told attendees. “The infrastructure to trust it doesn’t.”

That thesis is why the company built its new Veeam DataAI Command Platform, described as “the industry’s first unified data and AI trust infrastructure for the agentic era.” The platform is the result of Veeam’s December 2025 acquisition of Securiti, a leading data and AI security posture management vendor, combined with two decades of Veeam’s recovery and data protection capabilities. By design, it spans both production and backup systems, enabling visibility into what AI agents can access, what they did and how to undo it with precision.

Inside the DataAI Command Platform

Architecturally, the DataAI Command Platform is anchored by the DataAI Command Graph, which Veeam calls a unified intelligence layer with more than 300 connectors spanning public cloud services, SaaS applications, on-premises systems, and now backup environments. Jalil described it as a “social graph for data,” continuously mapping data assets, users, permissions, AI agents, activity and protection status across billions of files and millions of tables.

On top of that graph, Veeam has defined six integrated capabilities as the core of its trust layer:

• DataAI Security combines data security posture management, identity intelligence and detection of toxic combinations, such as sensitive data exposed to overprivileged agents or external users.
• DataAI Governance enforces controls at the data source rather than at the agent, so both sanctioned and shadow agents “hit a brick wall” when they try to access governed sensitive data.
• DataAI Compliance maps to more than 100 regulatory frameworks, including the EU AI Act, DORA, GDPR, HIPAA, NIST and AI RMF, generating “auditable and defensible proof” on demand for boards and regulators.
• DataAI Privacy, powered by a “People Data Graph” that unifies structured and unstructured personal data, automates consent, data subject rights, data minimization and cross-border transfer controls in real time.
• DataAI Precision Resilience uses the graph’s deep context to “undo exactly what went wrong without rewinding the entire system,” down to a specific data element or five seconds of agent activity.

An agentic layer of built-in AI assistants can answer natural-language questions, such as “Is workload X protected?” and automate tasks like log triage, ticket management, and policy-driven recovery. Jalil explained the reality of the situation: “If you don’t understand what data you have, who’s touching it, and what changed, there is no automation, no precision and no compliance.”

Product proof points for the pivot

The 2025 edition of VeeamON marked the moment when the company outlined its vision, and this year, it focused on releasing products that align with the trust narrative. The headline announcement was the previously mentioned Veeam DataAI Command Platform, positioned as the missing AI trust layer and immediately available with the DataAI Command Graph and five core domains live. Existing Veeam Data Platform customers can connect to it via a new DataAI Resilience Module, gaining centralized visibility and agentic capabilities “with no re-migration required.”

On the resilience front, Veeam previewed Veeam Data Platform v13.1, introducing more than 70 features, including expanded hypervisor coverage (targeting 95%+ of the market), portability across hypervisors, stronger Active Directory Forest recovery, post-quantum cryptography enhancements and smarter NAS archiving for lower-cost long-term retention.

The company also introduced Veeam Intelligent ResOps, the first resilience offering built natively on the DataAI Command Platform, with Microsoft 365 as the initial workload. Intelligent ResOps uses the graph to unify data, context and recovery across SharePoint, OneDrive, Teams and Exchange. When an AI assistant or human makes a bad change, teams can see exactly what changed and whether the data is sensitive or regulated, then “restore only what’s needed instead of broad, disruptive restores.”

Finally, Veeam launched a Data and AI Trust Maturity Model, developed with McKinsey and informed by input from more than 300 chief information officers and chief information security officers, to provide enterprises with a structured way to benchmark and plan their path from AI experimentation to demonstrable, auditable trust. Organized around the four pillars of Understood, Secured, Resilient and Unleashed, the model includes 12 dimensions and 49 subdimensions across five maturity levels. It is delivered as a consultative assessment featuring scored profiles, peer benchmarking and a prioritized roadmap.

Eswaran and Jalil talk AI challenges

Eswaran consistently returned to the theme that “recovery is the ultimate currency” in a world where AI, identity, and data are “completely connected, always under attack.” At the same time, he acknowledged that traditional notions of recovery are no longer sufficient: “You cannot roll back the enterprise” every time an agent goes wrong; instead, “remediation needs to be precise. You must be able to undo just those five seconds of agent action, just that one element in a file which got changed.”

He also emphasized that Veeam’s rebranding as “the Data and AI Trust Company” is more than label-swapping. “For us, DataAI is not just a branding exercise; it is where data, access and controls, identity and AI come together in one connected platform, because in the agentic era, you cannot solve these as individual problems.”

Jalil, whose Securiti team now forms the core of Veeam’s data security and governance stack, framed the opportunity and responsibility for resilience teams in the AI era. “If you’re trying to bring AI into the enterprise, you’re not going to put your intellectual property and your information behind it without guardrails and without knowing that you can recover from anything,” he said. “That really is the opportunity for our community and for us to play a central role in enabling the safe transformation toward AI.”

Why this pivot matters

From an industry-structure standpoint, Veeam’s move expands the competitive field it operates in. The company is no longer just competing against legacy backup vendors; it’s now colliding with DSPM providers, identity-centric security platforms, privacy automation tools and a growing wave of AI governance startups, while also claiming a unique position as the only vendor that deeply understands both the live data plane and the backup plane.

For customers, the question is whether it’s better to assemble trust capabilities from multiple best-of-breed tools or to consolidate on a unified platform that unifies data, identities, AI agents, and resilience. Veeam’s bet is that in the agentic era, running security, governance, compliance, privacy, and recovery as separate disciplines with different vendors, budgets, and UIs “stops being an option,” because every gap in context becomes a gap in trust.

That’s the essence of Veeam’s big pivot: from promising to get you back up when “everything else breaks” to promising that your data, and the AI acting on it, will be understood, governed and recoverable by design. If the agentic era unfolds as Eswaran and Jalil describe it, the companies that can operationalize that promise at scale will define the next infrastructure category.

If Veeam had acquired a cyber company five years ago, the industry might have said, “Huh? Why?” and viewed Veeam as a company with a solution looking for a problem. However, operating AI at scale is fundamentally different from operating the information technology environment of a decade ago. Make no mistake, the agentic era is coming fast, and it’s going to create problems IT leaders can’t comprehend. In an analyst Q&A, Eswaran discussed Veeam’s competitive position: “Regarding that 80% of the Fortune 500 – our forward-looking thesis centers on the ability to leverage a ‘Contextual Intelligence Knowledge Graph.’ This provides essential context to your data and its surrounding ecosystem, bridging relationships across identity, AI, and agents. We believe this will be our primary differentiator.”

Final thoughts

For customers, the takeaway from Veeam’s big pivot is to treat data and AI trust as an architectural requirement, not an add-on feature. That means inventorying where AI agents already touch your critical data, consolidating visibility across production and backup, and insisting on controls that can both prevent bad actions and surgically unwind them when they occur.

Don’t wait for a regulatory mandate or a headline-making incident to force the issue; use this moment to pressure-test your identity, governance, and recovery assumptions against an “assume autonomy” world. The organizations that move now to unify security, compliance, privacy and resilience around a common data graph will be the ones that can adopt AI fastest — because they’ll be the only ones that can prove, to themselves and to others, that they can trust it.

RingCentral Inc.‘s latest quarter shows a company that has quietly turned artificial intelligence from a future story into its primary engine for product differentiation, operational leverage and, increasingly, growth. What started as a unified-communications-as-a-service provider is evolving into an AI-first customer engagement platform, with RingCentral AIR and related products at the front door of every conversation.

Steady top line, accelerating AI

In Q1 2026, RingCentral reported total revenue of 644 million, up about 5% year over year, with subscription revenue of 623 million, up 6% and still representing 97% of the mix. GAAP operating margin reached a record 7.8%, up from 1.7% a year ago, while non-GAAP operating margin expanded to 22.9%, up 110 basis points. Free cash flow totaled $141 million, or 21.8% of revenue, up 8% year over year. Management raised full-year guidance for revenue, margins and free cash flow to $590 million to $605 million.

The most important metric, though, wasn’t on the income statement: Annual recurring revenue from customers using at least one paid AI product is now over 10% of total ARR, has doubled year over year, and is growing double digits sequentially, with higher average revenue per unit and net retention than the rest of the base. As founder and Chief Executive Vlad Shmunis put it, “Our native AI products continue to gain traction, with ARR from customers using at least one paid AI product standing at over 10% of total ARR.”

Despite the strong numbers, the stock was trading slightly lower after-hours, although it is up 46% year to date and almost 60% over the past 12 months. After the call, I spoke with a handful of investors and equity analysts; many took profits, triggering a selloff, while others remain unsure of AI’s long-term impact on communications. Some believe the shift from people to AI agents could deflate the industry. I believe AI will drive usage of RingCentral and its peers, creating a rising tide for the industry.

This aligns with the thesis of several equity analysts. This morning, in her note to investor clients, Catharine Trebnick, an analyst at Rosenblatt Securities, wrote, “AI is now greater than 10% of ARR and has doubled year over year, with strong attach and retention; Customers using two or more AI products grew roughly 7x. Together, AIR, ACE, CEB, and RingCX are becoming a meaningful driver of revenue quality, even if they do not yet move the headline growth rate. We see this AI stack as a 2026/2027 upside driver rather than a near-term inflection.”

AIR and the new front door of communications

During its earnings call, RingCentral announced an expanded release of RingCentral AIR, its AI receptionist, featuring new capabilities that bring AI directly to the places where business-to-consumer interactions begin. AIR is now one of RingCentral’s fastest-growing AI products, with more than 11,800 paying customers and over 40% quarter-over-quarter growth.

With the release, RingCentral is wiring AI into real workflows:

• AIR for shared SMS inbox lets the same AI agent handle calls and texts, instantly replying to questions like “Do you service my area?”, “Can someone come today?” or “What does this cost?” and booking appointments via SMS when appropriate.
• AIR for call queues allows AI to sit in front of or inside queues to absorb peak and after-hours calls, answering FAQs, scheduling and capturing urgent issues, instead of leaving customers on hold or in voicemail.
• Integrations with Shopify, Calendly and WhatsApp extend AIR into the e-commerce, scheduling and messaging ecosystems customers already use, effectively turning the phone number and messaging channels into an intelligent, integrated digital front door.

Language is no longer a barrier to adoption. AIR can now autodetect and respond in the caller’s language in real time, initially supporting 10 languages, including English, Spanish, French, Italian, German and Portuguese.

The customer outcomes are the proof points. Keller Interiors cut average wait times from 12 minutes to 90 seconds across 33 locations, boosted customer satisfaction or CSAT by three points in four months, and did so without adding headcount. Maple Federal Credit Union reports a 90% reduction in hold times, less staff strain, and more time for “conversations that matter.” These are exactly the kinds of metrics business leaders and boards care about: wait time, abandonment, CSAT and cost to serve. The most basic thing businesses have to do is answer the phone, and AIR lets them answer 100% of them.

AIR is also priced to remove friction: Standalone plans start at $49 per month with 100 minutes, while existing RingEX customers can add AIR starting at 39 per month with the same minute bundle. That makes the decision feel closer to “add another line” than “launch a new contact center project.”

How AI is fundamentally changing communications

Underneath all of this is a broader transformation in how communications platforms are architected and monetized.

First, AI is dissolving the traditional boundaries among UCaaS, CCaaS, and CPaaS. RingCentral’s RCAI portfolio, which includes AIR/AIR Pro (front-end automation), AVA (real-time agent assist) and ACE (post-call analytics and coaching), is integrated across RingEX, RingCX and RingCentral WEM, so AI touches every phase of the conversation: before, during and after human involvement. That means the same platform that powers a sales rep’s softphone also automates an after-hours receptionist and scores calls in a regulated contact center.

Second, the point of differentiation is shifting from dial tone to AI orchestration. In the early UCaaS era, the battle was over uptime SLAs, global coverage and mobility; today, the strategic real estate is where a customer first expresses intent. RingCentral’s argument is that processing tens of billions of minutes and billions of calls and messages per year gives it a unique vantage point into that intent stream, and that AI agents like AIR and AIR Pro can now sit at that front door to triage, resolve, or route interactions in ways that weren’t economically feasible with human staff alone.

Third, AI is making communications data actionable. With ACE, more than 5,200 customers are turning raw call recordings into structured signals: sentiment, next-best actions, compliance gaps, and coaching opportunities. On the earnings call, one example cited was Cartelligent deploying AIR, AVA, and ACE together, which reduced lead abandonment to zero, connected 100% of live leads during business hours, achieved an 85% lead-to-sign-up rate, and delivered a CSAT score of 9.85/10. That kind of closed-loop pipeline, from inbound lead to conversation intelligence, was historically the domain of large, bespoke customer relationship management and analytics projects.

Importantly, RingCentral is leaning into a hybrid AI-plus-human model rather than a purely automated narrative. Shmunis was emphatic that AI will take over more and more interactions, but legal, regulatory and complexity constraints (think healthcare diagnoses or regulated financial advice) ensure a continued role for human agents. The opportunity, as he frames it, is AI before the human gets involved, AI assisting while the human is involved, and AI after the interaction to learn and improve the next one.

How AI is transforming RingCentral’s business

For RingCentral, AI is doing three things simultaneously: creating new revenue streams, increasing stickiness, and supporting margin expansion. On the revenue side, AI ARR is now material, over 10% of total ARR, and more than double what it was a year ago. According to management, customers using at least one AI product buy more, stay longer and achieve net retention above 100%. Products such as AIR, AIR Pro, ACE and the Customer Engagement Bundle (CEB) give RingCentral a clearer upsell path beyond “more seats” and into “more intelligence per interaction.”

AI is also deepening the moat. All of RingCentral’s RCAI and customer engagement products are natively built and owned rather than resold third-party tools, which matters for roadmap control, iteration speed and economics. When you combine that with a global network, omnichannel capabilities and scaled distribution through service providers like Cox Business and Spectrum Business, you get a platform that’s harder for point AI startups to displace.

Investors commonly fear that AI workloads will erode gross margins due to model and infrastructure costs. RingCentral has taken the opposite stance. The company says it is maintaining roughly the same gross margins on RCAI products by “using the right model for the right job” and by taking advantage of how quickly state-of-the-art models are commoditized or open-sourced. Combined with internal use of AI to drive its own efficiency, expanded offshoring and vendor consolidation, non-GAAP operating margins have roughly doubled over the last three to four years to the current 23% range, with a medium-term GAAP operating margin target of 20%.

Investor lens: A compounding AI cash flow story

For investors, RingCentral is no longer a pure “UCaaS growth at any cost” story (although it remains a Gartner MQ leader); it is a mid- to single-digit grower with strong cash-flow metrics and credible AI-driven expansion. The company is guiding to $590 million to $605 million in free cash flow this year. Management highlights free cash flow per share as a key metric and plans to use that cash to de-lever (targeting 1 billion in gross debt by the end of 2026) and return capital via buybacks and dividends.

The key debate will be whether AI-led products can eventually lift consolidated growth above today’s approximately 5% “same dance” level once COVID-era repricing and large-customer rationalization are fully behind them. With AI ARR already over 10% of the mix and growing much faster than the base, the setup is in place for AI to move the overall needle over time, especially as partners begin to scale these offerings into their own customer bases in 2027 and beyond.

In the meantime, RingCentral is an example of what an AI transition can look like in a mature software-as-a-service business: not a wholesale pivot, but a steady rewiring of the product portfolio, go-to-market motion and financial model around intelligence at every stage of the conversation.

Customer lens: Reimagining communications in the AI era

For customers, it’s important to shed conventional thinking around communications and reimagine how every call, text and message is handled, with AI working alongside your human teams rather than replacing them. The new RingCentral AIR capabilities, from handling voice and SMS in one place to integrating with Shopify, Calendly and WhatsApp, let organizations start small (front-desk and after-hours automation, cutting wait times, reducing abandonment) and then layer on richer AI like AVA and ACE for agent assist and post-call analytics as use cases mature.

For information technology and business leaders, the practical takeaway is to pilot AI where pain is highest (missed calls, long holds, inconsistent follow-up), use measurable metrics like abandonment rate, CSAT and cost per interaction to track impact, and think in terms of a hybrid roadmap where AI is embedded before, during and after human interactions across the existing communications footprint.

At this year’s IBM Think, Chief Executive Arvind Krishna’s keynote focused on artificial intelligence’s impact on the modern enterprise. Instead of the usual tour through features and roadmaps, his talk challenged information technology leaders.

The real divide in the next decade won’t be between those who do and don’t use AI, but between those who rebuild their operating models around AI and those who stay stuck in pilots and proofs of concept. Viewed through the lens of the old wedding rhyme — something old, something new, something borrowed, something blue — the themes of AI, first operations, hybrid cloud, quantum and sovereignty that Krishna (pictured) emphasized become a practical checklist for chief information officers deciding where to place their next big bets.

Something old: Operating models, not pilots

Krishna’s core message is that the gap between “who’s winning and who’s falling behind” is widening, not because of budgets or team size, but because some organizations are using AI to “fundamentally rethink their business” while others are stuck in “little pilots and little projects.” He argued that the real question for IT leaders is “How deeply is AI embedded in your business processes? Is it part of the enterprise, or is it something on the side?”

He also quantified the stakes. Estimates point to roughly 40% productivity gains by 2030. AI infrastructure investment is up by roughly 150%, and IBM has realized “four and a half billion dollars in productivity gains” from AI and automation. He explained, “That’s not a projection. Those are reported numbers in our filings.” The lesson for IT is to treat AI as a new operating model, not a tool. Krishna went so far as to say, “AI is not helping your business. It is your business model. You’ve got to be AI-first, not AI-enabled.”

A great example came from Aramco, which Krishna called “a great example of AI-first thinking.” IBM first installed computers there in 1947 and is now helping “transform the Kingdom of Saudi Arabia into a global AI and digital hub,” including a new collaboration on using AI to address complex industrial challenges. That arc, from mainframes to AI-first operations, is the “something old”: replatforming the business around each computing paradigm shift, not treating technology as a bolt-on.

Something new: The quantum frontier

The explicit “something new” in Krishna’s framework is what he called “the quantum frontier.” He grouped it with AI-first and hybrid cloud as one of “three vectors that are important for you all to consider,” and positioned quantum not as science fiction but as an inevitable extension of current infrastructure and AI investments.

Krishna described this moment as “day zero of the AI revolution,” emphasizing that the biggest value is still ahead because most enterprises run AI “at the margin,” improving a workflow here and a use case there, while leaving core end-to-end processes largely untouched. Quantum sits just beyond that horizon: once enterprises have modernized their data, embedded AI into decision flows, and built hybrid-cloud control planes, they create the substrate for quantum to matter in areas such as optimization, materials, and risk analytics.

For IT leaders, the quantum takeaway is architectural rather than purely scientific. If you accept that quantum will become a specialized accelerator for hard problems, architect today’s AI and data platforms so they can eventually plug into quantum services without a massive rewrite. In practice, that means modular workflows, open orchestration and the assumption that tomorrow’s most valuable compute will not live in a single cloud or a single machine.

Something borrowed: Hybrid cloud and openness

The “something borrowed” in this keynote is the hybrid cloud model that enterprises have been dealing with for a decade, now recast as the backbone of AI and sovereignty. Krishna repeatedly stressed that value comes from “mixing” different models and deployments, “leveraging really big models and massive infrastructure where appropriate, and leveraging on-premise models and models at the edge where appropriate.”

He framed this as the way to avoid lock-in and unlock enterprise value. The companies best positioned are those that “really work hard on making the models work inside a real enterprise,” with “technical depth” and “decades of trust, especially for regulated industries,” plus “flexibility across providers, so you avoid lock-in.” In other words, IT should borrow the best from hyperscalers, on-premises systems and edge environments, then bind them into a single operating fabric instead of betting on any one stack.

Krishna underscored that this is where the real return on investment shows up. Most organizations, he noted, are stuck in stages one and two of the AI journeys, where manual workflows with some task augmentation and workflow optimization, while leaders move to stages three and four, where they redefine entire businesses and see “150% ROI compared to those who remain stuck in the first two.” Hybrid cloud is the “borrowed” pattern that lets IT leaders experiment broadly without sacrificing control, compliance, or portability.

Something Blue: Sovereignty as IBM’s differentiator

The “something Blue” is classic IBM territory — trust, control and sovereignty, all wrapped around advanced infrastructure. Krishna argued that “technology is as important to growth and national competitiveness as either finance or defense,” meaning that every nation and every enterprise “needs AI and cloud infrastructure that they control.”

On control, he was explicit: “Nobody else can turn it off, nobody else can tamper with it, nobody else can make it go dark when geopolitics or cable cuts under the ocean come in the way.” He warned that these are “not theoretical considerations” but “real and urgent business requirements of now,” elevating sovereignty from a compliance checkbox to a board-level resilience topic.

For IT leaders, this is where IBM’s “Blue” brand equity is being reasserted. AI-first operating models, hybrid-cloud architectures, and the quantum frontier all fall within a sovereignty envelope. If generative AI and modern infrastructure are the new engines of growth, Krishna’s challenge was stated matter-of-factly. Business and IT leaders need an architecture in which those engines cannot be shut down by anyone else, for commercial, technical, or geopolitical reasons.

Final thoughts: Day zeros and the missing middle

Krishna’s “day zero of the AI revolution” framing is compelling because it makes clear that most enterprises are still running AI at the margins. That is, they are tuning tasks and workflows while core, revenue-generating processes remain largely untouched. His statistics on potential 40% productivity gains by 2030 and IBM’s 4.5 billion dollars in annualized productivity from AI and automation underscore that the value is real, not theoretical.

What the keynote left largely implicit, however, is the organizational playbook. Who owns the AI operating model at the C-level? How do business and IT co-fund end-to-end reinvention? And how do boards govern AI as “the business model” rather than just another technology program?

Several critical execution themes were also underplayed. The grind of data readiness and governance needed to safely automate end-to-end processes; the reskilling and cultural change required to turn domain experts in tax, legal, supply chain and customer service into AI-literate co-designers; and the ecosystem strategy needed to make hybrid cloud and sovereignty real, not aspirational.

Krishna did a great job setting the strategic stakes around AI-first, hybrid cloud, sovereignty and the quantum frontier and spotlighting flagship customers like Aramco. But the “missing middle” now falls to IT leaders: turning that agenda into data foundations, skills programs, partner choices, and governance structures that move AI from pilots to the heart of the business model.

Nvidia Corp.‘s latest networking innovations meet the needs of a new kind of network that supports the unique demands of artificial intelligence factories.

Ethernet is no longer a generic plumbing choice but an enabler of high-performance AI. With today’s unveiling of Multipath Reliable Connection, or MRC, on Spectrum-X Ethernet, Nvidia is pushing Ethernet even deeper into AI-native territory — and doing so in partnership with OpenAI Group PBC and Microsoft Corp.

On the surface, MRC is a new remote direct memory access or RDMA transport protocol, now open-sourced via the Open Compute Project. In reality, it’s a production-proven way to keep tens or hundreds of thousands of graphics processing units fed and synchronized by using a single RDMA connection to stripe traffic across multiple paths and dynamically steer around congestion and failures. OpenAI has already used MRC on Spectrum-X to train recent frontier large language models powering ChatGPT and Codex, and Microsoft is deploying it in some of its largest AI factories built on GB200 systems. The important point is that MRC isn’t a lab experiment but a set of algorithms that has already earned its place in some of the most demanding AI environments on the planet.

What Nvidia announced

There are three intertwined elements to the announcement:

• MRC as a transport: MRC is an RDMA transport that lets a single connection fan out across multiple network paths, using multipath awareness, congestion hints and fast retransmission to keep bandwidth utilization high and failure recovery fast.
• SpectrumX as the platform: MRC runs natively on Nvidia SuperNICs and SpectrumX switches, riding on the same scaleout Ethernet fabric that already underpins large GPU clusters.
• An open spec via OCP: The protocol specification is being published through the Open Compute Project, with Nvidia, OpenAI, Microsoft, Advanced Micro Devics Inc., Broadcom Inc. and Intel Corp. all participating in its development.

That openness is important to scaling MRC. Nvidia has been adamant that everything in Spectrum X is built on standard protocols, with no proprietary wire formats and no lock-in at the packet level. The “secret sauce” is in how they partition control logic among NICs, switches and host software, not in a closed protocol. MRC follows that pattern: Anyone can implement the spec, but Nvidia believes its execution on SpectrumX hardware, with deep telemetry and fabric control, will be hard to match.

Why MRC matters for gigascale AI

When a frontier model is being trained across tens or hundreds of thousands of GPUs, the network is effectively part of the compute pipeline. If a link flaps for a few milliseconds or a path gets congested, it’s a stall in a multimillion-dollar training run and can cost big money.

MRC addresses that problem in several ways.

• Multipath load balancing: Instead of pinning a flow to a single path, MRC can distribute traffic for an RDMA connection across many paths, smoothing out hot spots and using all available fabric capacity.
• Congestion-aware routing: The protocol uses real-time signals from the fabric — congestion events and path health — to steer around overloaded links, sustaining high bandwidth even under stress.
• Fast, precise retransmission: When data is lost, MRC retransmits quickly and precisely, minimizing the impact of short-lived faults on long-running jobs.
• Microsecond failure bypass: SpectrumX can detect a path failure and reroute in hardware in microseconds, which is crucial when thousands of GPUs must stay in lockstep.

During a call, Nvidia Senior Vice President Gilad Shainer described MRC as extending the routing “brain” all the way to the host. The network interface card and the host-side management stack (in OpenAI’s case, its own software) can actively participate in routing decisions, thereby overriding or influencing what the switches do. That’s a major shift from classical Ethernet designs, where a hosted tenant has little or no control over the fabric.

In more traditional cloud models, a hosted customer has visibility and control at the virtual machine or server level, but the network fabric remains opaque. OpenAI wanted to change that, acting as a “smart tenant” with the ability to govern routing policy, congestion responses and failure behavior from the server edge. MRC is the mechanism that reconciles that desire with the realities of a shared, hyperscale fabric.

The role of multiplane architectures

Another key piece is SpectrumX multiplane support. Large AI factories are increasingly built as multiplane networks. That is a separate, independent network plane that provides a full path between GPUs. Think of it as having multiple disjointed fabrics in parallel, each serving as an alternative route for the same east-west traffic.

SpectrumX solves this. Hardware-accelerated load balancing across planes keeps latency predictable while scaling to hundreds of thousands of GPUs. Failures or maintenance events can be absorbed by shifting traffic between planes without disrupting training jobs.

MRC sits on top of this, using multiplane awareness to exploit those parallel fabrics more intelligently. The result is a kind of AI-native Ethernet fabric where redundancy, performance and control are baked into the transport, not bolted on via box-by-box tinkering.

MRC vs. adaptive RDMA vs. UEC

Nvidia is careful to present MRC as “another protocol” on SpectrumX, not a replacement for everything else. Today, SpectrumX supports at least two main Ethernet transports for AI. Spectrum-X plus adaptive RDMA is a general-purpose AI Ethernet with adaptive routing in the switches and NIC-level optimization. Spectrum-X with MRC is an RDMA transport emphasizing multipath, host-driven routing and governance.

There is also the Ultra Ethernet Consortium, which is a multivendor effort to define a new Ethernet RDMA-based fabric. I asked Shainer about the long-term implications of these Ethernet variants and he gave a very pragmatic answer. He does not see the world collapsing onto a single “winner” like UEC. Instead, he expects more variety: Different hyperscalers and AI providers will tune their transport protocols to their own workloads and operational models.

In that context, MRC is a great example of a “custom Ethernet for AI” that’s already running in production, while UEC is another evolving effort. Technically, MRC builds on RoCEv2 as defined by the InfiniBand Trade Association, then extends it with multipath, host-governed routing and the multiplane integration.

Some concepts that surfaced in UEC discussions — such as enhanced congestion control — also show up in MRC, but wired into Nvidia’s hardware and host stack. From a user point of view, the important bit is that SpectrumX gives you a choice: you can run Adaptive RDMA, you can run MRC, and there are other undisclosed variants SpectrumX can support that are specific to other large customers.

Hosted users vs. owners: A new control boundary

One of the more interesting subtexts in my conversation with Shainer is the distinction between “hosted users” and “infrastructure owners.” If you own the AI factory, you can program switches, NICs and hosts end-to-end; you can roll your own routing algorithms and congestion-control tweaks anywhere in the stack. If you’re a hosted customer — OpenAI on top of Microsoft, for example — you typically only control the host. The network underneath is someone else’s problem.

MRC exists largely to bridge that gap. By embedding new logic in the SuperNIC and exposing it to host-side management, a tenant can make meaningful routing decisions that the fabric will honor, without direct switch access. That allows OpenAI, or others with similar models, to optimize for their specific training jobs — changing routing strategies, reacting to congestion patterns, or tuning behavior per workload — without owning the whole data center.

That’s an important pattern to watch as AI ecosystems get more layered and multiparty. We’ll see more cases where a model provider wants near-owner-level control over routing and telemetry, even when they’re running on someone else’s iron. MRC is an early pattern for how that could be done safely over Ethernet.

Why this matters for the industry

From an industry perspective, MRC and SpectrumX underscore three trends.

First, AI is forcing Ethernet to specialize. Ten years ago, you could plausibly talk about “one Ethernet” dominating the data center. Today, we have a spectrum: shallow-buffer vs. deep-buffer switches, DCB vs. ECN-driven fabrics, a variety of RDMA variants, and now AI-specific transports such as MRC. Shainer’s line that “there is Ethernet, and there is Ethernet, and there is another Ethernet” isn’t just a joke — it’s the reality of the role the network plays in AI.

Second, open specifications with proprietary implementations are becoming the norm. By pushing MRC into OCP, alongside contributions from AMD, Broadcom and Intel, Nvidia gains ecosystem credibility while still betting that its Spectrum-X implementation will perform best. It’s the same playbook Nvidia has used in InfiniBand: standards on the wire, differentiation in silicon, and software.

Third, UEC is now one of several options, not the ordained future. With MRC in production on GB200-based clusters at Microsoft and in OpenAI environments, Nvidia can point to a working, large-scale, open Ethernet transport that doesn’t depend on the UEC kitchen to finish its meal. That doesn’t kill UEC, but it does make the future feel more pluralistic — one where hyperscalers, silicon vendors and model providers define and adopt the flavors that best match their economics and risk tolerance.

For enterprise buyers and service providers, the practical takeaway is this: When evaluating “AI networking,” don’t stop at port speeds and buffer sizes. Ask which transport protocols the fabric supports, how they’re implemented in NICs and switches, what telemetry and host-side control you get, and how quickly the system can respond to failure and congestion. In other words, treat the network as part of the AI architecture, not just a line item.

Nvidia’s MRC announcement, backed by OpenAI and Microsoft, is a strong reminder that in gigascale AI, Ethernet must mature and function as an AI-native fabric. With Spectrum-X, Nvidia is betting that the winning networks won’t just be fast — they’ll be intelligent, programmable and tailored to the unique demands of AI factories.

Extreme Networks Inc. used its Extreme Connect 2026 user conference this week to make a strong case that artificial intelligence-driven networking has finally arrived.

Building on Platform ONE, the company rolled out a full-stack vision spanning new Wi-Fi 7 access points, enhanced fabric-ready switching and a second-generation artificial intelligence layer called Agent ONE, designed to act less like a chatbot and more like an operational co-worker for NetOps teams. Framed by the insistence of Nabil Bukhari (pictured), chief technology officer and president of AI platforms, that Extreme is now an “outcome company” — and backed by several consecutive quarters of double-digit revenue growth — the announcements position Extreme well in the race to bring agentic AI to production networks.

What Extreme announced at Connect

The positioning of Extreme as an outcome company is enabled by Extreme’s “stack,” which starts at the hardware layer and extends up to AI to help customers meet business goals and cut operational costs. The evolution of that stack showed up in three buckets:

Extreme Platform ONE, expanded

Platform ONE, generally available since July 2025, is now the anchoring control plane for wired, Wi-Fi, fabric and software-defined wide-area network, with a strong focus on visualization and unified operations. This includes the following new capabilities:

• Single, “living” topology for physical, Wi-Fi and fabric layers, plus alerts/events, clients and inventory in one dashboard.
• Deep fabric visibility, so every node, link and service exposed, replacing the “fabric is a black box” complaint that kept coming up with customers and partners.
• Zero-touch provisioning and intent profiles for fabric, so teams can define once and push everywhere; Platform ONE monitors configuration drift and lets you snap devices back into compliance with a click.
• Integrated guest, location and wireless intrusion prevention system or WIPS services directly in the Platform ONE UI, eliminating separate portals and duplicate maps.
• “Edge Services” so third-party switches and access points (Cisco, Aruba, HPE, Juniper and the like) can be discovered and managed via a local on-premises service but visualized and orchestrated from Platform ONE.

Full Wi‑Fi 7 and switching refresh

Scott Calzia, vice president of product management, highlighted that Extreme was out of the gate early with Wi‑Fi 7 and is using Connect 2026 to showcase a broadening of its lineup for both high‑density and cost‑sensitive deployments.

• New WiFi 7 APs: 5022 and 5060 (indoor/outdoor 4×4 tri-band with dedicated sensor radio), plus 3020/3060 2×2 tri-band for entry-level and the 3020W wallplate for hospitality/dorm and retail use cases.
• On the wired side, new 5420M mixed-media switch (48port fiber/copper) and new 100G/400G options for the 7830, plus a universal ruggedized 4600 series to push fabric into operational technology/industrial spaces.

Agent ONE: Second-generation AI stack

When talking about AI, Bukhari was explicit that Extreme isn’t just “bolting a chatbot onto a frontier model.” He walked through a layered AI architecture:

• Frontier models and AI infrastructure at the base (sourced from hyperscalers, not built by Extreme).
• An Extreme AI Core is a networking-specific knowledge graph that encodes how MACs, clients, policies, sites and services relate across the Extreme universe.
• A skills layer where connectors, data pipelines and workflows live, so AI agents follow your standard operating procedures, change controls and tie into security, storage and compute, not just networking.
• An agentic layer, where Extreme Agent ONE operates in modes like Agent ONE Coworker (interactive copilot) and, later, Agent ONE Operator (autonomous operator).

The first mode, Agent ONE Coworker, is a proactive “warm coworker” that continuously monitors the network, investigates anomalies, and can execute changes on your behalf once you approve a plan. Extreme’s roadmap calls for a second mode, Agent ONE Operator, in Q4 CY26, to run workflows autonomously within guardrails.

I thought Bukhari summed up Extreme’s mission nicely when he stated, “Everything that we have thought of, everything that we have built, everything that we announce, is for one purpose, and that purpose is to make your life easier.” Too often, information technology vendors create new solutions but forget about the practitioner. Network operations have grown increasingly complex, and one of the first orders of business for AI should be to let engineers do what they need to do, but much more easily and quickly.

Customer traction on display

Extreme sprinkled real-world stories throughout the keynotes to demonstrate the value of Platform ONE and the new hardware. Chet Patel, director of innovation and technology for Caribe Royale Orlando, which hosted Extreme Connect, discussed how Extreme helped reduce Wi-Fi-related complaints. “We had hundreds of tickets before we deployed Extreme, and then after, we now have zero tickets,” he said. “It’s now ‘set it and forget it,’ which is not something we could have ever said about Wi-Fi before.”

In the “drinking your own champagne” portion of the keynote, Extreme’s own Chief Information Officer Anisha Vaswani explained the value of cross-portfolio integration. “We used to have one tool for managing on-prem, we had XIQ for wireless, and we had Ipanema (acquired by Extreme) for SD-WAN,” she said. “Today, we can manage all of that in one platform, and it’s made our lives easier and streamlined operations.”

Another customer, Richard Gingerich, systems engineer for Sight and Sound Theatres, talked about the value of Extreme’s fabric and the desire for an agentic style interface to assist the helpdesk. “I would love for the help desk to interact with the AI agents, he said. “Just ask the bot, ‘Bob has a problem, what’s going on?’ or ‘What VLAN is port 48 on?’ without having to come to the platform UI.”

Those quotes reinforce that Extreme’s AI story isn’t only front-of-house. They’re designing agents to serve as front ends for nonexpert tiers of the organization.

Financial performance underscores transformation success

It has been a long journey for Extreme. When Chief Executive Ed Meyercord (pictured below) and Norman Rice joined Extreme, the company’s viability was in question. A number of strategic acquisitions and a few years of engineering work have put the company in a strong position, with a simplified, strong enterprise portfolio.

Onstage, Chief Marketing Officer Monica Kumar and Bukhari both referred to “feel the momentum,” which is supported by recent numbers. Extreme has quietly put together a multiquarter run of solid growth while shifting the business mix to SaaS. The most recent quarter, the third of fiscal year 2026, saw revenues of $316.9 million, up 11% year-over-year, marking the fifth straight quarter of double-digit growth. Software-as-a-service annual recurring revenue is now $236.4 million, up 29%.

On the market side, Extreme’s stock is trading in the mid-20, up 42% this year and almost 70% in the past 12 months. Extreme Connect hosted several investors, who appear bullish on the company’s outlook.

Takeaways for network practitioners

Though Extreme Connect addresses many audiences, including mine, the most important audience is the network practitioner – the people who work day after day to keep networks up and running, ensuring businesses function. If you’re running a campus, distributed enterprise or OT-heavy environment, there are three practical angles from Extreme Connect that matter:

AI that understands networks, not just text

The Agent ONE design, which includes a knowledge graph, skills and an agentic layer, is an attempt to encode networking domain expertise so AI can do more than summarize PDFs. That shows up in tangible capabilities like real-time five-second packet streaming per client for Wi-Fi troubleshooting in Platform ONE, AI-driven control via conversation, and nudges that surface anomalies before users complain.

Operational convergence is becoming table stakes

Extreme is betting that “one pane of glass” is finally real, not just marketing: wired, Wi-Fi, SD-WAN, fabric, guest, WIPS and even multivendor assets hanging off “edge services” all land in Platform ONE. For practitioners, that means you can reasonably push vendors to show how their AI and observability work across the full topology, not just their own APs or switches.

Design for human-in-the-loop governance

Bukhari’s talk stressed that AI should optimize joint performance of human and machine,” and that humans must remain in and on the loop. As you evaluate AI-heavy networking offerings, including Extreme’s, the questions to ask are:

• Can I see every action the agent proposes and approve/deny it?
• Can I encode my change windows, rollback policies and site reliability engineering practices as skills?
• How easy is it for Tier1 to safely interact with the agent without blowing up the network?

Most IT leaders I speak with are bullish on AI as an IT tool, but many engineers remain cautious because you often don’t know what you don’t know until it’s too late. Looking ahead, here are a few pragmatic steps to move forward, while minimizing risk.

• Start piloting AI-assisted operations in low-risk domains (Wi-Fi troubleshooting, reporting, configuration validation) and capture hard metrics on MTTR and ticket volume.
• Push vendors to expose their “AI core” – including data models, guardrails and integration points – rather than accepting black-box copilots.
• Use these agentic platforms to bring networking closer to identity, security and ITSM. Extreme’s integrations with Entra ID, security information and event management and ServiceNow are a good benchmark to use.

Final thoughts

The broader story emerging from shows like Extreme Connect is that networking is finally entering the same AI-first transition that has already reshaped software development and security operations. As agentic systems and domain-specific copilots move from demos into everyday workflows, expectations for NetOps will shift. Platforms will be judged less on how many knobs they expose and more on how effectively they translate intent, policy and telemetry into closed-loop, measurable outcomes.

Customer experience used to be something tuned at the edges. Better call center scripts, nicer portals, lower prices or a new survey could make incremental improvements in consumers’ perceptions of a brand.

Today, it starts much earlier because every interaction across the channels customers use every day matters. Text, voice and messaging are no longer just “pipes” for getting a message from point A to point B. They’re now the primary mechanism through which trust is built or eroded in seconds.

The evolution of CX is long overdue. My research shows that 90% of companies now say they compete primarily on customer experience, up from just 28% five years ago. Many consumers will walk away after just one or two bad interactions. In that environment, the communications plumbing behind those moments isn’t back-office information technology anymore; it’s strategic infrastructure. RingCentral announced several innovations today to help businesses achieve step-function improvements in CX.

What RingCentral announced

The innovation payload introduced several capabilities to make customer interactions more trusted, interactive and real-time.

• Rich Communication Services (RCS) with Branded Messaging lets businesses send rich, verified messages — logo, brand name, tagline — directly into the native messaging app, evolving from basic SMS to interactive conversations.
• Enterprise Branded Calling displays a company’s name and logo on outbound calls to combat spam fatigue and improve answer rates.
• International SMS expansion adds SMS in the U.K. and Australia, plus notification support to 190 countries with intelligent routing and 98% average deliverability, targeting global CX consistency.
• AI Receptionist (AIR) extends from voice to shared SMS inboxes and call queues, handling inquiries and overflow with AI so fewer customer messages fall through the cracks.
• The Customer Engagement Bundle (CEB) for Microsoft Teams brings voice, SMS, call queues and analytics natively to Teams, effectively turning Teams into a lightweight contact center.
• Operator Connect for Teams adds global PSTN calling (46 countries) with RingCentral’s telephony, messaging and AI stack, managed from the Teams Admin Center and backed by 99.999% uptime.

Collectively, these can unify identity, trust, and reach across channels, making every interaction identifiable, contextual and measurable.

Why CX stakes have never been higher

As mentioned earlier, CX is how brands compete. Product availability is no longer an issue, making the experience different. This is why many business leaders consider CX improvement the top use case for AI solutions.

At the same time, consumer tolerance is collapsing; one or two poor interactions are now enough for many customers to churn to a competitor, especially in digital-first segments such as banking, retail and healthcare.

In that environment, three realities collide:

• Fragmented stacks mean customers bounce between phone, email, SMS and chat without continuity.
• Rising spam makes customers reluctant to answer calls or tap on texts from unfamiliar numbers.
• Labor constraints make “throwing more agents at the problem” unsustainable.

RingCentral’s announcement is interesting because it doesn’t add yet another standalone CX tool; instead, it focuses on enriching existing channels, making them more trusted and more automated, particularly for mobile messaging and Teams-based workflows.

Making every interaction trusted and recognizable

As businesses evolve their CX strategies, trust must be a core component. With RCS Branded Messaging, a text from a generic short code becomes a verified, visually branded experience that displays the name, logo, and tagline in the native messaging thread. Subsequent phases will add rich media, carousels, and one-tap replies, turning one-way alerts into conversational flows that support confirmations, rescheduling and promotions without requiring an app download.

That matters because recognition reduces the “is this spam?” hesitation that leads to messages being ignored and appointments being missed. Rich, app-like experiences within the messaging client can handle more of the journey, including search, selection, and confirmation, without bouncing the customer around.

On the voice side, Enterprise Branded Calling gives businesses a consistent, recognizable identity: name and logo on outbound calls. In sectors like healthcare, where organizations see higher callback and answer rates from clearly identified calls, that means fewer missed critical conversations and less staff time spent following up with patients.

For CX leaders, the implications are simple to understand. If customers don’t trust who is contacting them, you never get to show how good your service is. Branded identity across SMS and voice will become table stakes for modern engagement because it fosters trust.

AI that closes gaps, not just cuts costs

Most AI narratives in CX focus on cost reduction. More compelling is AIR’s role in eliminating “dark moments” in the journey. By extending AI Receptionist to shared SMS inboxes and call queues, RingCentral enables cross-channel automation.

Key capabilities include:

• Interpreting customer intent and using approved knowledge to generate accurate, real-time responses via voice and SMS.
• Handling overflow and after-hours calls so customers receive an immediate response rather than a busy tone or a long hold.
• Capturing structured context — who called, about what and what was promised — for faster, informed follow-up.

This directly addresses many of the hidden CX killers, including missed messages, abandoned calls, slow first responses, and agents repeatedly asking customers to restate the same issue. Existing engagement stacks often lack unified data and analytics. AIR, along with the broader RingCentral platform, targets those integration and insight gaps.

For a CX leader, KPIs should shift away from call deflection and focus more on reducing missed interactions, shortening time to first meaningful response, and improving consistency of answers across channels. That’s the lens through which this kind of embedded AI can materially move NPS and CSAT.

Turning Microsoft Teams into a CX hub

Many enterprises now live in Microsoft Teams for internal collaboration, but customer engagement often sits somewhere else, in a contact center, point solutions, or legacy telephony. RingCentral has been ahead of the market in using an embedded approach to fix that.

The embedded RingCentral app brings a softphone dialer, SMS, call queues, and presence directly into the Teams UI, allowing users to place and manage calls without leaving Teams. With CEB for Teams, RingCentral is layering customer engagement on top of that foundation:

• Voice and SMS are managed natively, with queue position, estimated wait time, and callbacks from within Teams.
• Shared SMS inboxes and calling workflows so frontline, backoffice, and specialists can coordinate around the same customer threads.
• Postcall summaries, realtime performance, and historical analytics embedded in Teams to make interactions measurable, not just transactional.

While many vendors bolt on calling to Teams via basic direct routing, RingCentral has set the standard for embedded telephony and engagement in Teams with its integrated softphone, presence sync, and now CEB features. Operator Connect for Teams extends that model globally, pairing RingCentral’s five-nines reliability and AI stack with Teams’ native admin experience.

How these capabilities change CX outcomes

These capabilities can improve a number of core CX metrics:

• Reachability and response. Branded identity plus international SMS with 98% average deliverability helps ensure critical messages actually reach customers and are recognized as legitimate.
• Speed to resolution. AI handling and triaging routine inquiries via voice and SMS compresses the time to first response, particularly during peak periods and after hours.
• Consistency across channels. A common platform for calls, SMS, and Teams interactions, with shared knowledge and analytics, reduces the “different answers on different channels” problem.
• Agent and employee productivity. Embedding engagement tools where employees already work (Teams, shared inboxes) removes swivelchair integrations and lets human staff focus on exceptions, empathy and complex problemsolving.

I asked Joe Rittenhouse, co-CEO of Converged Technology Professionals, one of RingCentral’s top resellers, about the impact on customers. “Customers are less interested in standalone tools and more focused on whether their existing channels actually work together for the customer,” he told me. “What we’re seeing with these RingCentral updates is a practical step toward that — trusted identity on calls and messages, AI that quietly handles the basics, and deeper use of Teams as a workspace rather than another silo.”

Recommendations for CX leaders

For CX leaders looking to modernize CX, there are several actionable steps I recommend:

• Make trust and identity a design principle. Treat branded messaging and calling as core CX elements, not add-ons; ensure every outbound interaction clearly identifies who you are and why you’re contacting the customer.

• Unify voice and messaging around journeys, not channels. Map your top customer journeys, onboarding, appointments, renewals and redesign them to use a mix of RCS, SMS and voice in a coordinated way. Leverage capabilities like shared SMS inboxes and call queues to ensure the same team can see all touch points around a given customer or case.

• Deploy AI where “silence” hurts most. Start with AI Receptionist-style automation in high-abandonment or after-hours scenarios, where missed interactions are common, and expectations for instant response are highest. Measure success on reduced missed calls and messages, improved first-response times and higher completion rates for key tasks (appointment confirmations, payments and the like).

• Prepare your operating model for “AI plus human” CX. Revisit roles, KPIs and training with the assumption that AI will handle more routine frontdoor tasks; shift human agents toward complex, emotionally nuanced, or high-value engagements. Build governance for AI knowledge bases and guardrails to ensure generative responses remain accurate, on-brand and compliant.

Final thoughts

As CX becomes the primary battleground and customer patience continues to erode, the winners will be the organizations that modernize the underlying communications fabric, making every interaction trusted, contextual and responsive, and meeting employees where they already work.

Major League Baseball opened the 2026 season with a technological first: the Automated Ball-Strike or ABS Challenge System, powered by T-Mobile’s private 5G network, is now deployed across all 29 U.S. ballparks.

While headlines focus on umpire challenges and strike zones, the real story for information technology professionals lies beneath the surface — in the network infrastructure that enables split-second decisions in environments where failure simply isn’t an option. In corporate IT, we have become accustomed to “best-effort” networking, but there are situations, increasingly in this AI-first world, where “good enough” is not nearly good enough.

The ABS system allows batters, pitchers and catchers to challenge ball-and-strike calls by tapping their helmet or cap, triggering an instant review via Hawk-Eye camera technology that tracks pitch location with millimeter precision. Each team receives two challenges per game, with additional challenges awarded in extra innings if the original allotment is exhausted. During spring training testing, the average challenge took just 13.8 seconds to resolve, with 4.1 challenges per game. That speed is possible only because of the network architecture T-Mobile built specifically for this use case.

Why private 5G was the only option

According to Scott Jacka, T-Mobile’s senior director of technology development strategy, the decision to deploy a dedicated private 5G network rather than rely on public cellular or Wi-Fi infrastructure was driven by three critical factors: local infrastructure integration, latency requirements and reliability guarantees.

“Major League Baseball owns the infrastructure on site at each ballpark,” Jacka explained during a recent analyst briefing. “They’ve got 12 Hawk-Eye cameras that go around the park. Those all connect back to a centralized room in the stadium. There’s processing happening locally, and there are operators running the game of baseball who are connected up in the press box.”

This local workflow architecture made a private network essential. T-Mobile deployed dedicated Ericsson EP 5G cores at each ballpark, connecting directly to MLB’s existing infrastructure via stadium cabling that MLB already owns and operates. This approach allowed T-Mobile to complete installations in about two days per stadium — a timeline that would have been impossible if crews had to run entirely new cabling through massive stadium environments.

The network uses 20 megahertz of T-Mobile’s N41 spectrum (2.5-gigahertz band) carved out specifically for MLB, with each stadium broadcasting a unique Public Land Mobile Network or PLMN identifier. This ensures that only authorized MLB devices can connect to the network, preventing interference from the tens of thousands of fans on T-Mobile’s public network in the same venue.

CBRS wasn’t an option

One of the most instructive decisions for IT professionals evaluating private network deployments is T-Mobile’s decision to avoid the Citizens Broadband Radio Service spectrum, despite its growing popularity for enterprise private networks.

Jacka cited multiple reasons for this decision. First, CBRS may pose reliability challenges in markets where military operations could cause interference — San Diego’s Petco Park, located near major Navy operations, was specifically cited as a risk factor. Second, CBRS typically operates at lower power levels than licensed spectrum. However, the decisive factor was ecosystem maturity. “There’s really not an ecosystem for 5G standalone in CBRS,” Jacka explained. “We would have been in something like NSA or just CBRS, and that didn’t necessarily make sense to us for the reliability objectives we were trying to work on with MLB.”

This is a critical lesson for enterprise IT practitioners. Emerging technologies and spectrum allocations may offer cost advantages, but proven, licensed spectrum with mature device ecosystems often delivers better outcomes for mission-critical applications.

Performance targets and device ecosystem

T-Mobile’s network targets approximately 100 megabits per second of downlink cell throughput — more than sufficient for the system’s primary use case, which is predominantly downlink traffic. The network supports iPads in dugouts and bullpens, where players study real-time performance data, and laptops in the press box operated by MLB staff who run the ABS system.

The device strategy also evolved during deployment. When the project began in early 2023, iPads accepted physical SIM cards. By the time of full deployment, Apple had transitioned to eSIM-only devices, requiring T-Mobile to adapt its provisioning approach. This highlights an often-overlooked aspect of enterprise network planning: Device ecosystems evolve rapidly, and infrastructure must be flexible enough to accommodate change without major rearchitecture.

Deployment lessons: From days to hours

T-Mobile’s deployment journey offers valuable insights into how enterprises can accelerate complex infrastructure projects through iterative learning. The first ballpark installation, at a college tournament at Houston’s stadium in March 2023, took several days and required significant on-site engineering. By the time of the final installations across 29 major league and two minor league ballparks, the team had refined the process to about two days per venue.

“The first time we took that out, it took us a little longer to set up,” Jacka acknowledged. “Through all the experience we’ve accumulated, the barriers are getting smaller and smaller. We’ve got a pretty good recipe for that.” This learning curve underscores a broader lesson for IT organizations: Pilot projects and controlled rollouts aren’t just risk-mitigation strategies; they’re opportunities to build deployment expertise that dramatically improves efficiency at scale.

Operational simplification through abstraction

One challenge Jacka identified is helping customers with traditional IT backgrounds adapt to wireless network operations. T-Mobile addresses this through its T Platform management system, which provides simplified views of network performance, device provisioning, alarms and system health, without requiring customers to master every detail of 5G architecture.

“Instead of trying to learn everything about a private network, they can get a more polished view that highlights the information that’s really critical to them,” Jacka said.

This abstraction layer is crucial for enterprise adoption of private 5G. IT teams shouldn’t need to become cellular network experts to deploy and manage private networks — they need intuitive tools that surface actionable intelligence while hiding unnecessary complexity.

The broader enterprise opportunity

Though MLB’s ABS system is a high-profile deployment, T-Mobile sees similar opportunities across multiple verticals. Jacka cited strong interest in government, oil and gas, manufacturing and broadcasting — especially as broadcasters face the loss of unlicensed spectrum they’ve traditionally relied on.

Each of these sectors shares requirements with MLB: local processing and data workflows, reliability requirements that exceed what public networks or Wi-Fi can provide, and latency-sensitive applications where milliseconds matter.

For enterprise IT pros, the MLB deployment shows that private 5G has matured from proof-of-concept to production-ready infrastructure for mission-critical applications. The key is aligning the technology with use cases that have clear requirements for local processing, deterministic performance and dedicated spectrum — exactly the criteria that made private 5G the right choice for calling balls and strikes in America’s pastime.

It has been just under a decade since Amazon Web Services Inc. launched Amazon Connect, taking its own internal contact center-as-a-service solution and commercialized it. At the time, there were many doubts about whether it could succeed in a mature market with several established vendors. The company loaded Connect up with artificial intelligence features long before AI was cool, and a unique utilization-based pricing model to disrupt, and it rapidly gained traction.

Today, it’s the primary customer experience platform for many major brands, including Capital One, Hilton Hotels, State Farm and Air Canada. AWS is repositioning the Connect brand as a family of agentic AI solutions that integrate into business workflows, not just the contact center.

The new portfolio comprises four products: Amazon Connect Decisions for supply chains, Amazon Connect Talent for high-volume hiring, Amazon Connect Customer for customer experience (the original Connect, rebranded), and Amazon Connect Health for healthcare delivery. As with Connect, these products are built on capabilities Amazon first used to run its own operations at massive scale, from optimizing a catalog of more than 400 million SKUs to hiring 250,000 seasonal workers in a single peak season.

Agentic AI as ‘teammate,’ not tool

A key design tenet across the suite is what AWS calls “humorphism” – the idea that AI should behave like a teammate, not a traditional application with menus and forms. Instead of adding AI features to existing software, the Connect products are built from the ground up for agents that can reason, remember, and act while collaborating with humans.

That shows up in patterns like agents proactively asking planners about upcoming promotions that could affect demand, or interviewing job candidates overnight and handing recruiters a curated brief in the morning. In our conversation, Pasquale DeMaio, vice president of Amazon Connect Customer and Talent, underscored that philosophy: The AI handles much of the screening and interviewing process, “but then moving forward after that, it hands off to a human recruiter,” so people still make the final call. He was crystal clear on our call that the Amazon Connect portfolio isn’t here to replace people but to let them work smarter and faster.

Why the pivot is working

Connect’s evolution has been underway internally for some time. DeMaio told me it has “been over two years” since he last referred to Connect as a contact center, arguing that the market continued to pigeonhole the service as “just about pipes” even as Amazon shipped more AI-centric capabilities. Renaming the CX product Amazon Connect Customer is intended to make it unmistakable that “Connect is really an AI service.”

It’s worth noting that this isn’t about Amazon building speculative AI products and hoping customers will find a use for them. The company is productizing systems that already power core Amazon businesses: supply chain optimization powered by its SCOT foundation models, high-volume hiring tuned to its seasonal workforce, and healthcare workflows honed through One Medical and Amazon Pharmacy. As DeMaio put it, Amazon often “learns so much” and “does some amazing science” for internal tools, then spends time making them “enterprise-grade for a broader set of customers,” which is the original Connect story.

Amazon’s internal-first advantage

Amazon’s history of taking internal platforms external is well-known — AWS itself began as infrastructure for Amazon.com — but the Connect family is a more targeted version of the same playbook. Internally, Amazon used its hiring science to quickly identify the right candidates, maintain a high performance bar and systematically reduce bias. The new Amazon Connect Talent essentially packages that operating model for customers.

Because the tools are born in production, they start with clear outcome metrics: time-to-fill, offer-in-a-day targets, retention and evaluation consistency for Talent; forecast accuracy, exception resolution time and working capital impact for Decisions; and containment, handle time and NPS-style measures for Customer. DeMaio said the same science that lets Amazon “move very quick, but keep a very high bar for the hiring process” is what they’re now offering to enterprises grappling with weeks-long hiring cycles.

There’s also a go-to-market benefit: each of these workloads can reach departments that historically would never buy a contact center, and then lead them back toward the broader Connect platform. DeMaio acknowledged this dynamic, telling me he believes Talent will be a “backdoor” into the customer experience side, even though it’s intentionally sold as a standalone product that does not require adopting other parts of Connect.

Inside Amazon Connect Talent

Among the new offerings, Amazon Connect Talent may be the clearest signal of how far Connect has moved from its contact center roots. Talent is aimed squarely at high-volume hiring, where organizations constantly trade off speed and quality.

The workflow starts with an existing job description. AI agents generate a complete interview plan, including competencies, structured questions, and evaluation criteria, which recruiters can review and adjust. Candidates then interview on their own schedule, 24/7, via voice. The agent asks job-related questions, adapts to responses, and assembles a package of anonymized competency scores, transcripts, and notes for recruiters. DeMaio emphasized that this can compress hiring cycles from “a week or weeks” to “a day,” driving “5x, 10x improvement” in time-to-offer for many high-volume roles.

Equally important is the objectivity story. Candidate names and other identifying information are removed from recruiter dashboards, eliminating obvious bias vectors such as name, address, or accent. “You can never provably remove 100% of bias,” DeMaio cautioned, but Amazon’s goal is to give customers “good guardrails,” visibility into where bias might be creeping in, and tools to “prevent it on the front end” while keeping hiring a “shared responsibility.”

Benefits for customers and recruiters

For customers, the immediate benefit of Amazon Connect Talent is speed and scalability without surrendering control. Recruiters tune the experience up front, using prompts, job descriptions, or other data, and can adjust thresholds as they observe how competency scores translate into retention or performance for their specific roles.

This also changes the recruiter’s day. Instead of starting a week buried under unread applications, they begin with a curated pipeline, consistent scores, and full transcripts when detail is needed. DeMaio described it as a “fundamental sea shift” in which recruiters no longer need to be involved in every touchpoint, freeing them to focus on relationships rather than repetitive administrative work. Over time, customers can build a library of interview templates for different job types while letting the agentic layer handle execution.

Connect Customer and Decisions stay core

Even as Amazon expands into talent and healthcare, the rebranded Amazon Connect Customer remains the flagship for customer engagement. It now offers configuration tooling that enables business teams, not just developers, to stand up conversational experiences in weeks rather than months, covering identity verification, payments, recommendations and issue resolution. Enterprise customers like United Airlines have been able to go from concept to production in about three months, compared with six months or more with legacy stacks.

Amazon Connect Decisions shows how Amazon is extending the same agentic pattern deeper into operations. Built on more than 25 specialized supply chain tools and SCOT-backed forecasting models, Decisions continuously generates and tunes demand forecasts, then triages thousands of alerts into a small set of prioritized exceptions, each with root-cause analysis and suggested resolutions. Customers such as Wells Vehicle Electronics and TVS Motors are already using it to shift from weeks-long planning cycles to adjustments measured in minutes.

The bigger picture for AWS

Taken together, the new Connect family is AWS’ answer to a key enterprise question: how do you operationalize AI beyond pilots in the contact center? By anchoring each product in a real business function and Amazon’s operational history, AWS is positioning Connect less as a channel-centric platform and more as a suite of AI teammates that sit wherever work actually happens.

DeMaio told me this rebrand is about “resetting the idea of what Connect is,” building on a decade-long brand that customers “trust and love,” and using it to “bring AI to enable AI as teammates to help people be better, not to remove connection.” If AWS executes, Connect may come to be known less as Amazon’s contact center and more as Amazon’s application layer for agentic AI across the enterprise.

Cisco Systems Inc.’s new Universal Quantum Switch introduced last week is a strong proof point regarding the network’s importance in scaling quantum.

For information technology leaders, the key takeaway is that quantum is shifting from isolated computing hardware to an interconnected fabric, and Cisco has been positioning itself as the core quantum interconnect for whatever qubit technologies ultimately prevail.

Why quantum needs a network

Quantum’s big promise, namely solving problems such as molecular simulation, materials discovery, portfolio optimization and large-scale scheduling, requires on the order of 10^5 to 10^6 logical qubits, far beyond what any single system will deliver this decade. Current roadmaps top out in the thousands, and possibly in the low tens of thousands, by 2030 at best.

That gap forces a fundamental architectural shift:

• Instead of betting on a single, gigantic quantum computer, the industry is converging on distributed quantum computing. That is, many smaller processors act as a single logical machine via a quantum network, much as classical computing scaled out over Ethernet and IP.
• To do that, you cannot just move classical “results” between machines; you must move the quantum state while preserving entanglement, so the processors behave as a single, aggregated system rather than independent islands.

That’s why a network built specifically for quantum is so pivotal. In classical networks, switching silicon turned point-to-point links into the Internet. In quantum, a switch that can route entangled photons without destroying their quantum properties is the missing ingredient to turn isolated experiments into a quantum network.

What Cisco announced

Cisco’s Universal Quantum Switch is a research-grade quantum fabric element designed to route entangled photons at room temperature over standard telecom fiber while preserving quantum information across multiple encoding modalities.

While there was a bucket list of attributes to this, the most notable are:

• Quantum property-preserving switching. Conventional optical switches destroy quantum information. Cisco’s design uses an internal “quantum state converter” to take whatever encoding comes in, convert to an internal format, then reconvert on exit, without collapsing the quantum state.
• Modality universality. It supports major quantum encodings (polarization, time-bin, frequency and path) and can translate among them, so a neutral-atom system could communicate with a superconducting or photonic system through the same fabric.
• Network-native characteristics. It operates at telecom wavelengths compatible with DWDM fiber, targets nano-second reconfiguration, and is designed to share expensive elements, such as entanglement sources and detectors, across many endpoints.

Cisco couples this with its earlier entanglement source chip, which can generate roughly 200 million entangled photon pairs per second at telecom wavelengths and at room temperature, plus a stack of entanglement distribution, swapping, and teleportation protocols. In fieldwork with partner Qunnect over New York metro fiber, Cisco has already demonstrated multi-kilometer entanglement swapping at rates orders of magnitude above prior lab-only experiments.

Cisco now has the “transmitters” (entanglement source), the “fabric” (quantum switch) and the early “control plane” (compiler and orchestration software) needed to turn quantum boxes into a networked platform.

Why the network is central to quantum’s future

Quantum hardware grabs headlines, but the economic value will emerge when enterprises can treat quantum capacity as another pooled resource — much as GPUs and CPUs are consumed today via cloud and high-performance networks.

The network is the enabler in three ways:

• Scaling out in addition to scaling up. By teleporting qubits over entangled links, many modest-sized machines can function as a virtual large-scale quantum computer, accelerating the arrival of “useful” quantum computing for chemistry, finance and logistics.
• Heterogeneous quantum data centers. Different modalities excel at different algorithms (for example, trapped ions, neutral atoms and superconducting qubits), so future quantum data centers are likely to combine them. A modality-agnostic switch lets you architect for a heterogeneous future now, rather than betting on a single winner.
• Quantum-enhanced classical applications. Even before million-qubit systems exist, quantum networks enable new classical services, such as coordinated decision-making across distant trading engines (“Quantum Sync”) and fiber intrusion detection via entanglement-based sensing (“Quantum Alert”). Both rely on sharing entanglement across many endpoints, something only a scalable quantum fabric can provide.

As classical infrastructure hits physical and economic limits, the ability to add “quantum links” for specific high-value functions becomes strategically important. This is precisely where a player who understands routing, synchronization and operations at scale can differentiate.

Why Cisco is well-positioned

Quantum networking is greenfield because it involves entanglement distribution rather than conventional store-and-forward. Cisco’s approach is to build quantum networks by leveraging the existing optical fabric as much as possible, hence its focus on optical telco frequencies. A classical IP network is still required for signaling and reconfiguration. The deep knowledge required in both domains plays to Cisco’s strengths:

• End-to-end quantum networking stack. Through its incubator group, Outshift, Cisco is building hardware (entanglement chip, universal switch), software (quantum compiler, orchestration, distributed error correction) and integration with post-quantum cryptography—all anchored in an architecture that assumes heterogeneous processors from multiple vendors.
• Compatibility with existing infrastructure. Room-temperature operation at telecom wavelengths means the quantum fabric can ride on existing fiber, amplifiers and much of the optical ecosystem, rather than requiring exotic cryogenic links. That dramatically lowers deployment friction for carriers and cloud providers.
• Ecosystem and field experience. Cisco is already partnering with major modality providers, including IBM Quantum (superconducting) and Atom Computing (neutral atoms), and working with operators on metro-scale testbeds. This gives it both a voice in emerging interfaces (for example, quantum NICs and compilers) and practical experience integrating quantum gear into noisy, real-world environments.

Strategically, Cisco is playing to its strengths and approaching quantum much as it did with artificial intelligence. Rather than trying to own the entire stack, it’s becoming the fabric that brings together different vendors across modalities and locations. If quantum follows the same trajectory as classical and AI, where value concentrates around platforms that pool and route specialized resources, Cisco should be in a position to ride another rising tide.

What IT leaders should do now

Most CIOs and network leaders will not deploy a quantum switch next year, but the decisions they make over the next three to five years will determine how prepared they are when quantum moves from research to revenue.

Here are a few recommendations:

1. Treat quantum as a multivendor, networked service

Assume you will consume quantum computing from multiple providers — hyperscalers, specialized quantum clouds and possibly on-premises systems — and that those resources will need to interoperate. Architect your data center and wide-area network strategy with the expectation that quantum interconnects (for example, metro-scale entanglement links) will become another class of high-value connection, much like today’s private cloud onramps. Watch how vendors such as Cisco, IBM, and Atom define quantum NICs and APIs; those will become the “Ethernet ports” of the quantum era.

2. Start with quantum-adjacent pilots

You do not need a quantum computer to gain experience with quantum networking concepts. Explore early quantum-enhanced classical applications. For example, secure fiber monitoring, ultra-precise time synchronization or coordinated decision services in financial trading, through pilots with carriers and vendors active in this space. Use those projects to build internal expertise in entanglement-based security models, operating procedures and failure modes (including denial-of-service on quantum links) without betting on a specific qubit technology.

3. Align security and networking roadmaps

Quantum cuts both ways with security. Quantum computers threaten current cryptography, but quantum networks also enable intrinsically secure communication models. Accelerate post-quantum cryptography programs for classical control and management planes; the classical signaling around a quantum network must be hardened long before large-scale quantum adversaries exist. Track how networking vendors integrate quantum-safe algorithms into routers, switches and controllers to avoid a bifurcated “quantum-secure island” bolted onto an insecure core.

4. Build a quantum-literate architecture team

Quantum networking spans physics, optics, distributed systems and security; it will not fit neatly into any current silo. Designate a small cross-functional team (network, security, cloud and data science) to own your quantum roadmap, including vendor relationships with Cisco, IBM, hyperscalers and specialized startups. It’s important to give them a mandate to develop reference architectures for “quantum-ready” data centers and metro networks, with clear assumptions about timelines.

Quantum will not replace classical infrastructure; it will augment it where the economics justify it. Cisco’s universal quantum switch signals should simplify scaling the technology and make it less of a physics experiment and more of a roadmap IT can plan against.

Nvidia Corp. and Google LLC used the search giant’s annual Cloud Next event to deepen their long-running partnership, creating a full-stack “artificial intelligence factory” that integrates Google’s AI Hypercomputer infrastructure with Nvidia’s latest solutions, including Blackwell, open models and agentic and physical AI tooling.

With this announcement, Google expands its distribution of Nvidia’s accelerated computing stack, while customers gain a faster, lower-risk path from AI experimentation to large-scale deployment.

What was announced at Next

• Google Cloud is extending its AI Hypercomputer architecture with new Nvidia-powered instances (including Grace Blackwell systems and the upcoming A5X instance based on the Nvidia Vera Rubin platform) for large-scale “AI factories” for training and inference.
• Virgo Networking is a data center network fabric designed for megascale AI. Introduced by Google, it serves as the backbone for Google’s AI Hypercomputer and will enable the Vera Rubin A5X instance to scale to 960,000 graphics processing units across multiple sites.
• Agentic AI and “physical AI” use cases are showcased: Nvidia Omniverse libraries and the open-source Nvidia Isaac Sim robotics simulation framework are available on Google Cloud Marketplace, enabling developers to build physically accurate digital twins and develop custom robotics simulation pipelines to train, simulate, and validate robots before real-world deployment. In addition, Nvidia NIM microservices for models such as Nvidia Cosmos Reason 2 can be deployed on the Google Enterprise Agent Platform and Google Kubernetes Engine.
• The partnership spans cloud (Google Enterprise Agent Platform, GKE, DGX Cloud), on-premises and edge via Google Distributed Cloud on Nvidia Blackwell, providing customers with a consistent platform from lab to production across environments.

A decade-long full‑stack collaboration

Nvidia and Google Cloud have been co-developing the accelerated cloud stack for about a decade, starting with early K80/P100 GPU instances and evolving into the AI Hypercomputer architecture.

That collaboration has been expanded to address the entire AI stack:

• Infrastructure: Nvidia GPUs (H100, RTX PRO 6000, GB300, GB200, B200, H200, H100, L4 and A100 GPUs today with Vera Rubin coming) power GCE, GKE, Vertex AI, Batch, DGX Cloud and Distributed Cloud, all tied into Google’s custom networking, storage and schedulers.
• Libraries and software: Nvidia CUDA, cuDNN, Dynamo, NeMo, Nemotron and optimized JAX/PyTorch are integrated with Google Cloud services and reference architectures.
• Managed services integrations: Vertex AI, GKE, Cloud Run and Google’s AI Hypercomputer all have Nvidia GPUs and autoscaling, and provide native observability, so customers consume Nvidia as an on-demand cloud primitive rather than a bespoke hardware project.
• Models and agents: Gemini models on Vertex and the Gemini Agent Platform are now cross-linked with Nvidia’s open Nemotron models and NeMo tools, giving customers a choice of model families optimized for Nvidia hardware.

For customers, co-engineering means there is no need to stitch together GPUs, schedulers and frameworks, as the combined stack is designed to be turnkey and is approaching “utility” status.

Google’s million‑plus-GPU footprint

Google has quietly built out one of the world’s largest accelerated infrastructure deployments, with well over a million Nvidia GPUs deployed across its global fleet for internal products and Google Cloud services.

There are two implications for this scale. The first is that it shortens deployment times. Because the backbone, supply chain, and data center footprint are already GPU‑centric, adding each new GPU generation (Hopper, Blackwell, Vera Rubin) can roll out faster, and those accelerators show up quickly in customer‑facing SKUs like A3/A5X and DGX Cloud.

The second point is that there should be plenty of capacity for AI factories. The technology footprint that underpins Google’s AI Hypercomputer concept — multitenant, massively scaled clusters where training, fine-tuning and inference share the same fabric — makes it realistic for enterprises to spin up large language model and agent workloads that run across tens of thousands of Nvidia GPUs without bespoke infrastructure engineering.

Information technology leaders no longer have to guess which region or instance type will still be available at scale in 18 months — Google is standardizing on Nvidia as the default accelerator fabric, alongside its tensor processing units.

Nvidia makes the move from general‑purpose to accelerated computing easier

Nvidia has rewritten the computing stack by shifting heavy compute workloads away from general-purpose central process units toward GPU-accelerated architectures optimized for parallel workloads.

Key aspects of that shift:

• From instruction-driven to parallel data-driven: Traditional CPUs are optimized for serial workloads, whereas GPUs deliver massive parallelism that AI, HPC, graphics and data analytics exploit; CUDA and its ecosystem make that parallelism programmable at scale.
• From components to platforms: Despite the media positioning Nvidia as a chip company, that’s only one part of its offerings. The company sells a full platform — GPUs, CPUs, interconnects (NVLink), networking, systems (DGX, GB300 NVL72),and extensive software stacks such as CUDA, cuDNN, Dynamo and NeMo.

That “accelerated computing” mindset is why Nvidia maps so cleanly onto Google’s AI Hypercomputer strategy: Both focus on building dense, software-defined supercomputers rather than generic cloud infastructure as a service.

Why Nvidia reach beats any single TPU/ASIC

Specialized accelerators like TPUs and other application-specific integrated circuits are powerful and often positioned as a threat to Nvidia, but they are narrow. Nvidia’s bet has always been horizontally broad programmability. This has the following benefits:

• Ecosystem gravity: Virtually every major AI framework (PyTorch, JAX,), along with a long tail of domain-specific frameworks and libraries, has first-class, production-hardened support for Nvidia GPUs because CUDA is the de facto standard for accelerated computing.
• Workload diversity: Nvidia accelerates not only LLMs but also recommendation systems, traditional ML, scientific HPC, data analytics, simulation and digital twins, media, gaming and graphics pipelines, all on a common platform.
• Portability across environments: The same CUDA binaries and container images can run on-prem, at the edge, or on public clouds such as Google Cloud, AWS, Azure and others, giving independent software developers and enterprises a broad distribution surface that no proprietary ASIC can match.

So although TPUs will remain strong within Google for specific workloads, Nvidia’s cross-industry, multicloud footprint makes it attractive to enterprises that need to ship software to any customer, anywhere.

For Google Cloud, aligning with Nvidia broadens the appeal of its AI infrastructure to customers who want a neutral, portable accelerated platform rather than a proprietary stack that locks them into a single cloud or architecture.

Why this matters to Google Cloud

• Differentiated yet open: It can lead with TPUs for internal products and select Vertex AI offerings, but partnering with Nvidia lets it claim the broadest possible ecosystem support for enterprise AI, spanning open-source to proprietary models.
• Faster innovation cadence: Google inherits Nvidia’s rapid GPU roadmap (Hopper to Blackwell to whatever is next) and combines it with its own networking, storage and AI orchestration fabric — meaning customers see new capabilities sooner, with less integration pain.

Why this matters to Nvidia

• Distribution and visibility: Google Cloud becomes one of the most visible, multitenant showcases for Nvidia’s latest platforms, spanning training, inference, agents and physical AI, strengthening Nvidia’s position as the default AI hardware choice.
• Deeper stack integration: Tight integration with Vertex, GKE, Cloud Run and Distributed Cloud provides Nvidia privileged access to enterprise workloads and telemetry, which can feed back into its software and hardware optimization loops.

Why customers should care – and how it accelerates AI adoption

For customers, this partnership is about reducing risk and shortening time-to-value. Specifically:

• Lower platform risk: Building on Nvidia via Google Cloud enables customers to follow two strong roadmaps — Nvidia’s for accelerated computing and Google’s for hyperscale AI infrastructure —r ather than betting on a single proprietary accelerator.
• Faster path from PoC to production: There is so much hype today about customers getting stuck in proofs of concept. With this partnership, customers can prototype with Gemini or Nemotron models on Vertex or GKE, then scale to DGX Cloud or massive AI Hypercomputer clusters without changing hardware architectures or rewriting for a different accelerator.
• Operational maturity: Google wraps Nvidia GPUs in managed services with autoscaling, observability and MLOps patterns, so teams can focus on models and applications instead of driver versions, firmware, and schedulers.

This combination lowers the organizational friction of adopting AI because infra teams, data scientists and app teams share a common, battle‑tested platform.

Nvidia-Google partnership can accelerate AI adoption

While customers want choice, too many variables in an equation can slow things down. The Google-Nvidia stack provides enterprises with a reference design for building AI factories, cloud-scale clusters for training, fine-tuning, inference and simulation — that they can consume as a service or emulate on-premises with similar building blocks.

• Support for agentic and physical AI: By integrating Nvidia’s NeMo, Nemotron and robotics and digital twin platforms, customers can move beyond chatbots to agents that plan, act and interact with the physical world, all on the same accelerated platform.
• Ecosystem leverage: Because “all kinds of frameworks and algorithms run on Nvidia,” enterprises can adopt best-of-breed open-source components, ISV solutions, and custom models without fighting the hardware; that flexibility encourages experimentation and shortens the iteration loop.

Google has spent a decade playing third fiddle to Amazon Web Services and Microsoft Azure, but its partnership with Nvidia gives it a first-fiddle story in AI: a co-designed AI Hypercomputer, tuned for agentic and physical AI, that turns Google’s Nvidia-powered supercomputers into a product enterprises and startups can actually buy. Google has a decade-long partnership with Nvidia and offers the widest range of Blackwell instances today. In the AI era, choice is important, and Google gives customers that.

The DP World Tour will become the first professional sports organization to use Amazon Leo as its official satellite connectivity partner, deploying low Earth orbit or LEO terminals at tournament venues starting in 2026.

The network uses more than 3,000 LEO satellites to deliver high-speed internet to locations underserved — or completely unserved — by terrestrial infrastructure.

At selected events, the Tour will deploy a mix of Leo Nano, Leo Pro and Leo Ultra antennas around the course, with the top-end Leo Ultra delivering up to 1 gigabit per second down and 400 megabits per second up — enough capacity to support not only broadcast but also dense onsite digital experiences. The golf course itself effectively becomes a temporary, high-performance edge network, spun up in days rather than months.

From a networking perspective, this marks a fundamental shift: Instead of hauling miles of temporary fiber and hoping local mobile operators can keep up, organizers can “show up, point antennas at the sky, and light up the entire tournament,” as Amazon Leo Vice President Chris Weber put it. That mindset — treating the network as a rapidly deployable, location-independent utility — extends far beyond golf.

How LEO can reshape the fan experience

Golf is uniquely punishing from a connectivity standpoint: The “venue” is an expansive outdoor environment that may span several square miles, with fans and operations scattered from remote tee boxes to the media compound and overflow parking. Traditional Wi-Fi and cellular are pushed to the limit under that combination of distance and density.

Also, no two courses are the same, and fans are constantly on the move. The density of the fans is constantly changing hole by hole. More people will likely follow Calum Hill or Rory McIlroy than a player ranked near the bottom of the tour. Providing consistent, high-quality access has long been a problem for golf.

By enabling high-bandwidth connectivity throughout the course, the DP World Tour can finally treat the fan experience as a digital canvas rather than a constrained resource. A few concrete possibilities include:

• Hyper-personalized mobile apps: With reliable coverage from “remote tee boxes to broadcast compounds, hospitality areas, and parking lots,” the Tour can leverage real-time shot tracking, player-location maps, and context-aware notifications for every attendee on site. Imagine walking with a favorite group and having the app automatically surface live stats, win probabilities, and historical performance for the shot you are about to see.

• Immersive, data-rich viewing: The DP World Tour already uses data-driven insights to provide new levels of intelligence to fans globally; pairing that with consistent on-course bandwidth makes it practical to deliver multi-angle video, advanced analytics and AI-generated storytelling to spectators’ phones without saturating local networks. Think of “Golf Zone” experiences where fans can toggle among player point of view, aerial drone feeds and augmented reality overlays showing shot trajectories in real time.

• Smarter navigation and crowd management: Chief Technology Officer Michael Cole has identified AI-guided navigation as a priority to help fans move around the venue more efficiently. This requires continuous device connectivity so the system can understand flows, bottlenecks and dwell times across the course. With LEO backhaul and distributed antennas, organizers can layer in dynamic routing, live wait times at concessions, and targeted safety alerts in ways that simply weren’t viable with patchy cellular.

• Frictionless commerce and operations: Reliable bandwidth for point-of-sale systems, ticket scanning, merchandise inventory and concessions is often invisible until it fails; most fans notice only when lines stall or payments time out. Satellites that provide consistent uplink and downlink capacity across the site enable smoother transactions, better inventory visibility and the ability to spin up pop-up experiences in previously unreachable corners of the course.

What matters here isn’t any single app or feature; it’s the ability to view the entire venue—from the first tee to the farthest parking lot—as a continuously connected, intelligent environment. That’s the leap LEO networks make possible: they remove the last-mile excuse for not investing in richer digital experiences.

Intelligent venues as a template for other industries

Although this announcement is set in professional golf, the underlying pattern applies to virtually any industry where operations and customer experiences take place in areas that aren’t well served by fixed connectivity. The DP World Tour stages tournaments in 25 countries across five continents, often in rural locations with varying levels of infrastructure. That’s not unlike an energy company managing remote fields, a logistics provider operating a depot at secondary ports, or a healthcare provider delivering pop-up clinics in underserved communities.

The common thread is a need for high-quality, rapidly deployable connectivity independent of local build-outs. Amazon Leo’s model — bring antennas, point at the sky and get enterprise-grade bandwidth — can be mirrored by other LEO constellations and next-generation satellite services as they mature. For chief information officers, that means “we can’t get a reliable network there” is increasingly a business decision rather than a technical inevitability.

In practice, intelligent venues built on LEO connectivity set a template in three ways:

• They show how to decouple digital strategy from local infrastructure constraints, enabling you to design experiences for what you want to deliver, not what the incumbent ISP can support.
• They reinforce that connectivity is a prerequisite for AI, not the other way around; you cannot do real-time guidance, analytics or automation at the edge if the network is the bottleneck.
• They highlight a new operating model in which network capacity is treated as just-in-time, event-scope infrastructure rather than fixed, sunk capital.

Why the network matters more than ever

The DP World Tour’s stated ambition is to create “truly connected and intelligent courses, wherever we are in the world.” That goal is only possible if the network is architected as a first-class platform layer, not as an afterthought to applications or devices.

Network performance is directly tied to the quality of digital experiences: if you want real-time scoring, interactive apps, and on-site AI, you need deterministic bandwidth, predictable latency, and resilient failover. LEO networks help address this by providing a controllable, high-performance backhaul that organizers own and operate, rather than relying entirely on best-effort cellular for mission-critical functions like scoring and payments.

More broadly, as enterprises lean into AI and automation at the edge, the network becomes the business’s control plane. Data generated in the field is only useful if it can be moved, processed, and acted on within the right time window. In that sense, LEO connectivity is not just “adding internet” to remote locations; it is extending the enterprise nervous system to where work and customers actually are.

Practical advice for IT leaders: Leveraging LEO and elevating the network

For technology leaders reading this news outside of sports, the takeaway is not “we should buy satellite.” The real message is that you need a strategy to make dense, high-quality connectivity available wherever your business needs to operate — and to treat that connectivity as a strategic differentiator. Here are some actionable steps:

• Map your “off-grid” experiences: Identify the physical environments where your customers and employees struggle with connectivity, such as remote sites, outdoor areas, temporary events, mobile operationsd and partner locations. For each, list the high-value digital experiences you’re not currently delivering because “the network can’t handle it.”

• Treat connectivity as a product, not a utility: Assign product ownership for connectivity as you would for a customer-facing app, with clear experience-level objectives (latency, availability, throughput). Build a roadmap that treats low Earth orbit, private 5G, fixed wireless and traditional WiFi as interchangeable tools to achieve those goals.

• Pilot in low Earth orbit to shift the business conversation: Start with constrained, high-visibility use cases: large outdoor events, remote operations centers, pop-up experiences and disaster recovery sites. Measure not only technical metrics but also business impact: higher sales per visitor, reduced downtime,and better safety or engagement scores.

• Close the loop between network and AI strategy: Ensure that every edge AI initiative — from real-time video analytics to dynamic routing — has a corresponding network architecture. Use the DP World Tour model as a reference: AI for fan guidance and insights was constrained by connectivity; LEO is the enabler, not the headline.

• Elevate the network in executive conversations: When presenting digital transformation roadmaps, place connectivity architecture on the same slide as applications and data, not in the appendix.

The DP World Tour’s embrace of Amazon Leo is a great example of what happens when you stop treating connectivity constraints as a given. As more enterprises follow suit in their own domains, the competitive line will be drawn less by who has the best app and more by who has built the most capable, flexible, and pervasive network to support it.

The 2026 edition of Adobe Summit this week marks a historic turning point for the software giant.

It not only showcased the next frontier of “agentic artificial intelligence” but also served as the swan song for Shantanu Narayen, who delivered his final keynote as chief executive. Narayen, who has steered Adobe through the transition to the cloud and the birth of the digital experience category, used his final stage to outline a future in which AI doesn’t just assist humans — it performs work on their behalf.

Though this idea has been heard before, what was unique was the sheer scale of the data foundation Narayen presented. By connecting agentic AI to the Adobe Experience Platform — which already processes 35 trillion segment evaluations a day — Adobe isn’t just giving agents a “brain” but a complete memory of every customer interaction. This allows the AI to perform work that is contextually aware of a customer’s entire history, not just the immediate task at hand.

Here are the five key takeaways from Narayen’s presentation:

1. The dawn of the “agentic enterprise.”

Unsurprisingly, the keynote’s central theme was the shift from generative AI to agentic AI. While generative AI focuses on creating content from prompts, agentic AI is designed to execute multistep business goals. Narayen introduced Adobe CX Enterprise, an end-to-end system that uses specialized “co-workers” to identify prospects, orchestrate journeys and optimize brand visibility.

As Narayen noted during the presentation: “Tools don’t create; people do. But winning isn’t just about producing the most content. It’s about producing the right content, on brand, at scale, and delivered in a way that feels personal.”

This has always been Adobe’s calling card, and AI lets its customers do what they have always done, but now much faster.

2. Digital twins and the marriage of physical and digital AI

A highlight of the keynote was the appearance of Nvidia CEO Jensen Huang, a longtime friend and collaborator of Narayen. The two visionaries discussed the “Omniverse” and the necessity of high-fidelity digital representations of physical products. Huang emphasized that for AI to truly transform industries such as manufacturing and logistics, it must understand the physical world through “precision digital twins.”

Huang explained the evolution of the computing model: “In the future, you’re going to come up with a seed of an original idea, and every time your customers enjoy it, your content is going to be processed by generative AI and presented, described and illustrated in a way that is contextually sensible.”

3. Creativity is the new productivity

Narayen and David Wadhwani, president of Adobe’s Creative Business, argued that in an era flooded with “AI slop” and a “sea of sameness,” creative differentiation is the only way for brands to survive. To address the “content crunch,” where demand is expected to grow fivefold over the next two years, Adobe unveiled the Adobe Creative Agent.

This agent serves as a human-led, agent-accelerated partner that can convert a strategic brief into a creative one, source assets, and even perform complex edits in Photoshop (such as the new “Rotate Object” feature) in seconds. By automating the mundane, Adobe aims to free humans to focus on the “human insight” that machines cannot replicate.

4. Brand intelligence as the new governance layer

As AI-generated content proliferates, the risk of “hallucinations” or off-brand messaging is high. To address this, Adobe introduced Adobe Brand Intelligence. This “living, breathing system” learns from a company’s past approvals, rejections, and brand guidelines to ensure every asset created, whether by a human or an agent, is compliant.

For chief information officers and security leaders, accountability and governance raise questions such as: How are prompts logged? How are training and inference data governed? How do agents respect consent and privacy policies when acting across channels? The keynote implicitly challenged customers to “hold Adobe accountable” on these topics, which, in turn, means enterprises need their own internal AI governance playbooks to fully leverage the new platform’s capabilities.

5. AI as a job multiplier, not a replacer

This was a critical point to raise because it addresses the elephant in the room — job displacement. Huang offered an excellent counter-narrative. He cited radiologists: When AI began reading scans with superhuman accuracy, demand for doctors actually increased because they could see more patients and focus on higher-level clinical outcomes.

Huang argued that the same is happening to creators and engineers: “I think that the fact that we’re now so productive, we can experiment and iterate so fast, we’re going to be busier than ever. In the final analysis, what you pay for is work done.”

Translating Huang’s “more doctors” point to the enterprise stack, Adobe’s agentic framework becomes the clinical workflow in which many “AI opinions” (models, agents, segmentations, optimizations) are coordinated and triaged. The organizations that win won’t be those that replace marketers, data scientists, or designers, but those that give these experts a richer panel of AI “second opinions” embedded directly in creative, analytics and journey-design tools.

Every technology transition we have had has an initial wave of job elimination, followed by significantly more jobs created on the back end, and AI will do the same. The key for employees is to understand where the new bottlenecks are and to build the skills to address them.

Narayen to his peers: Ambition must outrun your current roadmap

Narayen’s keynote framed us as still in “the early innings” of a gen AI- and agent-driven era, and said leaders will be judged on whether they replatform their businesses around data-driven experiences, not just pilot features. His broader philosophy came through in a line he has used publicly: “If you can connect all the dots between what you see today and where you want to go, then it’s probably not ambitious enough or aspirational enough.”

Amid the Summit announcements, the message was that CIOs and chief marketing officers cannot wait for a perfect playbook; they need to set bolder targets for real-time personalization, experimentation and AI-assisted workflows, even as the implementation path continues to evolve. For tech leaders, Narayen’s challenge translates into rethinking funding models, governance and talent so that AI-driven experience transformation is treated as a multiyear platform shift rather than a series of tool budgets.

A legacy of transformation: Summarizing Shantanu Narayen’s tenure

Shantanu Narayen’s 18-year tenure as CEO should be remembered as one of the most successful corporate pivots in history. He famously led Adobe’s transition from perpetual “boxed” software to a software-as-a-service subscription model, which initially sent stock prices lower but ultimately led to nearly a 20-fold increase in market capitalization.

By expanding Adobe’s reach from the “Creative Cloud” into the “Experience Cloud” through landmark acquisitions such as Omniture and Marketo, Narayen transformed Adobe from a toolmaker for designers into a mission-critical platform for the world’s largest enterprises. He leaves the company not only as a leader in digital media but also as the primary architect of the “Experience Economy.”