Amazon Web Services Inc.’s re:Invent is in the books, and if there was any question about whether the show is back, there shouldn’t be.
Two years ago, participation was capped as we emerged from the pandemic. Last year, some businesses still had not fully embraced travel and trade shows, and that kept the show from achieving the level of audience it has had in the past. AWS re:Invent was jam-packed this year, with an estimated 65,000 people spread across multiple venues. Trying to get into the Expo Hall for Happy Hour was like going through airport security in a non-PreCheck line during the holidays. It’s back.
Like all re:Invents, there was plenty of news ranging from compute to generative artificial intelligence to visionary stuff, such as Project Kuiper, AWS’ low Earth orbit satellites that promise to bring internet to everyone, finally closing the digital divide. I’ve found it important to look “beyond the news” for the underlying themes of the show to understand the impact of the cornucopia of the news is. Here are my top five thoughts from re:Invent 2023.
AWS is no longer only for builders
Since its inception, AWS has targeted builders – that is, information technology pros who build enterprise applications and services. Although the company does have a handful of software-as-a-service applications, such as the Amazon Connect contact center and Supply Chain, which are targeted to lines of business, these are the exception and not the norm.
Amazon Q, its new generative AI assistant, will change that as it makes it possible for nontechnical people to do technical things easily. At the event, I met with Tracy Daugherty, general manager of Amazon QuickSight, who shared how Amazon Q in QuickSight enables it through natural language. “Instead of taking a group of analysts days or even weeks to represent data graphically, our goal was that within 10 minutes, a nontechnical user could create a story about their data that they could share,” he said. Another example is CodeWhisperer, where a software power user can now write code and create an application without requiring a developer.
The developer will remain the primary audience for AWS in the near term, but it will be interesting to see how AWS transforms its go-to-market and the show. One final thought here: As AWS begins to move “up the stack,” how it manages partners will become more important. A couple of years ago, it hired Ruba Borno away from Cisco Systems Inc. to run its partner organization. Cisco, in my opinion, has the best partner organization in technology, and the work she has been doing will be critical in the continued evolution of AWS.
Its approach to generative AI is set up for long-term success
The generative AI holy wars are underway, with Microsoft having a perceived advantage because of its investment in OpenAI. I certainly think the Sam Altman debacle and rehiring gave Microsoft a marketing advantage as far more people have now heard of OpenAI compared with, say, Amazon Bedrock.
But success via marketing has never been the AWS way. Instead, it relies on building good quality products that help customers solve problems, and that will ultimately enable AWS to win in this area.
Amazon Bedrock is built on the concept of choice. If a customer wants to use an Amazon-built LLM, they’re there. If the customer would rather use Anthropic or Meta, have at it. If the customer starts with an Amazon model and then decides to switch, that’s fine, too.
AWS is operating with the thesis that there will never be “one model to rule them all,” and Bedrock provides customers with choice. The caveat with this is the feedback was at an AWS event with AWS customers, but the thesis is still solid. Providing choice is better for the customer and that wins out long-term.
It’s innovating in areas considered a commodity
I always find it fascinating that the AWS keynotes typically start with the core products, as many consider things such as storage and computing a commodity. The reality is that since it first launched Amazon S3 in the mid-2000s, the company has continued to innovate in this area.
Amazon Glacier S3 enabled companies to archive data at a fraction of the cost of object storage. Later, intelligent tiering was introduced to optimize the use of the different storage tiers. Chief Executive Adam Selipsky (pictured) mentioned during his keynote that customers have saved more than $2 billion with this feature.
At re:Invent, AWS introduced a new service – Amazon S3 Express One Zone — optimized for an organization’s most frequently accessed data. Customers can also choose their own Availability Zone, enabling businesses to locate frequently accessed data next to their high-performance compute to minimize latency.
Selipsky mentioned this storage tier was 10 times faster than S3 standard storage. I’ve always believed in technology; very few markets are truly commodities. If the vendor is willing to invest in it and deliver new features, customers will buy them.
Building its own silicon gives AWS a significant price-performance edge
When Selipsky moved to the topic of silicon, he stated, “We realized almost 10 years ago that if we wanted to continue to push the envelope on price-performance, we had to reinvent general-purpose computing for the cloud era, down to the silicon.”
In 2018, AWS became the first cloud provider to develop its own silicon when it introduced Graviton. Since then, it has released Graviton 2 and 3. Selipsky quantified the success of these chips in his keynote: “Today we over 50,000 customers using Graviton based instances.” At re:Invent, he announced Graviton 4, which has 50% more cores and 75% more memory bandwidth than its predecessor.
AWS also has chips for AI: Trainium for AI training workloads and Inferentia for AI inferencing. At re:Invent, AWS announced Trainium 2, specifically designed for high-performance training of foundation models.
I expect this to be an area in which AWS continues to invest heavily, although the AWS mission has always been choice. Customers can get the price-performance benefits of AWS customer chips or use Intel, Arm and, of course, Nvidia.
