AWS relies on generative AI in a direct duel with Microsoft

Unlock Editor’s Digest for free

Roula Khalaf, editor of the FT, picks her favorite stories in this weekly newsletter.

Through its close alliance with OpenAI, Microsoft has been able to outpace the rest of the technology industry in the booming field of generative artificial intelligence.

But judging by the number of announcements at Amazon Web Service’s annual tech showcase event in Las Vegas this week, the AI ​​gap in the cloud computing world has narrowed. Microsoft’s biggest cloud competitor has now deeply integrated generative AI into its computing platform and integrated the technology into many of its services, analysts said.

Like Microsoft and others, AWS is pushing deeper into a new era of AI-powered automation. The question now is whether customers are willing to entrust some of their most important functions to generative AI or invest serious money in the technology.

Amazon said in its latest earnings release that AI is already a “multibillion-dollar” business and growing at more than 100 percent. According to AWS CEO Matt Garman, AI is on its way to becoming a core function in every enterprise application, potentially sparking a new wave of demand for the company’s core computing infrastructure.

“It will need computing power, it will need storage, it will need databases and it will need inference as a key part of that application,” he said in an interview with the Financial Times.

Still, Garman acknowledged that the high level of interest in generative AI in the two years since ChatGPT’s launch has not yet translated into serious use of the technology in business. “A lot of customers have experimented a lot,” he said. Most are now trying to identify the few uses of AI that might justify a major investment.

AWS’s announcements at its re:Invent conference on Tuesday in Las Vegas included a third generation of its Trainium processors, which will be used to train large language models. Apple said it used AWS chips to run some of its services and had good results testing Trainium to train its internal AI models.

Like other cloud companies, AWS still relies heavily on Nvidia’s general purpose computing units. But the latest developments show it is now on par with Google, which trains its Gemini models on the in-house chips it has used for nearly a decade, said Patrick Moorhead, principal analyst at Moor Insights & Strategy. Combined with its other chips, including those for operating and training AI models, Amazon now has “the largest selection of homegrown chips” of any major cloud company, he added.

AWS has also narrowed the gap with Microsoft and OpenAI when it comes to the models that serve as the foundation for its AI services, analysts said. Garman said AWS has spoken to OpenAI about trying to offer their models to its own cloud customers. Although OpenAI’s exclusive alliance with Microsoft stands in the way, he said he believes the two will find a way to work together in the longer term.

Even Microsoft CEO Satya Nadella no longer talks about hosting OpenAI as a crucial competitive weapon. Instead, he recently claimed that LLMs are becoming a commodity and that most of the value comes from the tools and services on top, suggesting that competition with its key cloud rivals AWS and Google is moving to another level.

Meanwhile, at the event in Las Vegas, AWS demonstrated how it has integrated AI into more of its own technologies and services to encourage customers to bring more of their applications and services to the cloud.

“The biggest challenge with AWS is how complex it is and how many services it provides,” said Steven Dickens, principal analyst at HyperFRAME Research. Adding AI to simplify things has led to “200 small successes,” he added. “None of this is very sexy, but it is absolutely important in everyday life.”

One sign of how AI could change the competitive balance between cloud companies is AWS’ attempt to use the technology to poach customers from Microsoft. It unveiled new AI-powered tools to automate the tedious and time-consuming task of rewriting applications that run on Microsoft’s Windows operating system to run in the AWS cloud.

AWS’s direct targeting of Windows customers follows news first reported in the FT that the US Federal Trade Commission has opened an investigation into whether Microsoft used licensing restrictions to prevent customers from joining competing cloud providers change.

Echoing the complaints of others in the industry, Garman said Microsoft had offered its customers preferential rates if they kept their applications on its own cloud rather than choosing one of its major competitors, claiming it had made “false concessions” to prevent an EU investigation into the issue last year. But he also claimed that Microsoft’s actions could backfire: “In some ways it’s a tailwind for us because it doesn’t deserve trust.”

Meanwhile, in the latest attempt to push generative AI deeper into business, AWS also announced new tools for organizing and coordinating groups of AI agents to perform more complex functions. Many of the tasks companies want to automate involve multiple steps, meaning they need a set of specialized agents to carry them out, said Vasi Philomin, vice president of generative AI at AWS.

But even Garman admits that most customers have just started thinking about how AI tools like these might be integrated into their processes or how they might manage the risks.

“We’re still in the early stages of a lot of these things,” he said. “Users must decide when agents can be fully autonomous, when agents still need people in the loop, and when agents need guardrails around them.”