AWS reinvents its GenAI strategy

Welcome to Runtime! Today: How all sides of a long-running product-strategy discussion inside AWS emerged at re:invent 2023, the fallout from ten days of chaos at OpenAI, and the latest funding rounds in enterprise tech.

(Was this email forwarded to you? Sign up here to get Runtime each week.)

The AI storm is coming

LAS VEGAS — AWS began life as a one-stop shop for the basic computing resources needed to build a business on the internet, and over the years — amid some internal debate — gradually started to manage those resources on behalf of late-arriving cloud customers who needed help before deciding to build a few full-blown SaaS applications for those customers. At re:Invent 2023, it drew a little bit from all of those strategies in hopes of jumpstarting its AI business.

In his third re:Invent keynote since taking the top job in 2021, CEO Adam Selipsky emphasized why he thinks AWS's investments in internal chip design set it apart from competitors, unveiled new managed services to help companies work with data, and introduced a comprehensive enterprise-focused chatbot called Amazon Q. He also announced that AWS will join Microsoft, Google, and others in offering indemnification to customers in the event of generative AI lawsuits, after ducking the question earlier this year at a press event in Seattle.

Just about everything in tech starts at the chip level, and AWS introduced new silicon for both general-purpose computing and AI.

• Graviton4 is the latest generation of AWS's cost-effective Arm-based server chips, and it promises 30% better performance than Graviton3 on most workloads.
• Trainum2 is the newest version of its homegrown AI training chip, and Selipsky said it would allow customers to train their AI models four times faster than the original version.
• But Nvidia remains the performance king when it comes to AI workloads, and CEO Jensen Huang joined Selipsky on stage to announce an expanded partnership that will include a commitment from AWS to offer Nvidia's Grace Hopper CPU/GPU combo chip as well as its DGX Cloud service to AWS customers.
• "DGX cloud is Nvidia's AI factory," Huang said. "We are incredibly excited to build the largest AI factory Nvidia has ever built," he said, revealing plans to construct a supercomputing GPU cluster on AWS called Project Ceiba, named after the largest tree in the Amazon.

The generative AI boom is entirely dependent on data, and as companies start to experiment with building their own AI models on their own data, they need new tools to manage that process.

• AWS unveiled a new version of its S3 storage service called Express One Zone, which is generally available today and was designed to provide fast access to corporate data that is sensitive to latency, such as the data used to power real-time applications.
• It introduced new services that help customers avoid the complicated ETL (extract, transform and load) data pipeline process when using its Aurora, DynamoDB and Redshift databases, and added vector search capabilities to Amazon OpenSearch Service.
• And during Peter DeSantis' keynote last evening, AWS announced new serverless services for Aurora, ElasticCache, and Redshift that automatically handle database sharding.

The most interesting announcement of the day was AWS's answer to Microsoft's Copilot strategy; Q, either named after James Bonds' gadget wizard or the probably-fake anonymous leader of a cult of deranged lunatics, depending on your worldview.

• Selipsky emphasized that Q — launched as a public preview — is strictly business, highlighting examples of how it could be used to help automate IT management tasks like migrating applications or generating business intelligence reports.
• One example: "With Amazon Q Code Transformation, a very small team of Amazon developers successfully upgraded 1,000 applications from Java 8 to Java 17 in just two days," Selipsky said. "This is months, if not years, of developer time saved."
• Rather than adding a generative AI assistant to basically every one of its services, which is how Microsoft is rolling out its AI tech, Q is the latest example of AWS's relatively focus on building complete end-user applications for its customers that span a variety of business needs.

It's been almost a year since OpenAI released the preview of ChatGPT right in the middle of re:Invent 2022, and AWS barely mentioned the concept of generative AI during that week. Obviously, this year has played out very differently, and this week's relentless focus on AI — with an entire AI-focused keynote from Swami Sivasubramanian still to come on Wednesday — shows how much the world changed.

• AWS has always poked fun at its enterprise tech rivals during re:Invent, but Selipsky took a sharper edge this year, taking shots at both Microsoft and Google in references to the OpenAI debacle (more on that in a bit) and Google Cloud's massive outage in France earlier this year.
• The company is used to setting the pace of enterprise tech, and while it is still the primary engine of that market it's clear that it might no longer enjoy that luxury.
• So while AWS has been locked into this post-Thanksgiving yearly product launch schedule for a very long time now, it might be time to reconsider the cadence of its product roadmap.

Open and shut?

One of the wildest weeks in recent tech history concluded over the holiday weekend with the return of OpenAI co-founder and CEO Sam Altman to his old job and the appointment of two new board members. While that seems to have been enough to quiet the employee revolt — and probably welcome by Microsoft executives who would have had to quickly spin up a separate AI research arm staffed by hundreds of expensive engineers — it's not clear that the dust has settled.

For one thing, Altman and comrade Greg Brockman, who will remain president of OpenAI, no longer have a seat on the board of the directors. Neither does Microsoft, although it's probably safe to assume it prefers dealing with Bret Taylor and Larry Summers rather than the group that put its $13-billion investment in OpenAI in jeopardy.

Microsoft retains access to OpenAI's technology, but still doesn't have the direct input that a major investor in almost any other startup would have over the direction of the company. And the new board plans to conduct some sort of investigation into whatever the hell just happened, which even if conducted with the lightest of touches could hang over the company for months to come.

Enterprise funding

Oxford Quantum Circuits raised $100 million in Series B funding and unveiled what it called "the world’s first enterprise-ready quantum platform," which feels like a bit of a stretch.

Pika landed $55 million in seed funding and launched its video editing and creation tool based on generative AI technology.

The Runtime roundup

Alibaba reassigned several current executives to lead its cloud computing unit and shut down its quantum computing research after it canceled a planned IPO for the cloud division last week.

Police arrested five members of a ransomware gang operating out of Ukraine after an international investigation into attacks that resulted in "losses exceeding several hundred million euros."

Google's geothermal project in the Nevada desert is up and running, helping to produce power for two data centers in the region.

Vertiv now offers customers the option of using mass timber when building prefab data centers, which could have less of an impact on the environment than producing steel for those buildings.

Thanks for reading — see you Thursday!