How Amazon went from an AI also-ran to a real contender

SEATTLE—Not long ago, Amazon was seen as an also-ran in the great AI race.

Microsoft, in particular—with its partnership with OpenAI and cloud business outpacing Amazon’s AWS—seemed to have reclaimed its place as the Puget Sound’s reigning tech giant.

Things are looking much different for AWS today.

That’s thanks to some savvy deals of its own with OpenAI and Anthropic, plans for $200 billion of infrastructure spending and Amazon’s long-term bet on custom chips. It’s also not involved in the ugly tech trial of the year sucking in the biggest names in AI, from Elon Musk and Sam Altman to Mira Murati and Microsoft CEO Satya Nadella.

“People thought we were behind,” Matt Garman, head of Amazon’s cloud-computing business, AWS, told me in an interview here on the corporate campus. “As we’ve progressed, they’ve seen our strategies start to evolve, and they’ve started to see other people realize that that strategy has a lot of merit.”

Along the way, as the cloud business progressed from a side project to a major part of Amazon’s business, companies stopped asking why they should turn their computing needs over to a bookseller.



The current moment was helped by a decision made more than a decade ago to acquire Annapurna Labs to begin designing chips for AWS’s cloud-computing needs—the goal being lower cost and using less energy.

That effort developed a chip called Graviton that handles traditional CPU needs, and Trainium for AI. Both are useful for inference, the next phase of AI compute needs. (The same demand for inference computing is fueling a resurgence in Intel shares.)

If the chips business was a stand-alone unit, Amazon would be generating an annual run rate of $50 billion from the sale of chips to AWS and others. Trainium chips are at the heart of huge investment commitments from both OpenAI and Anthropic, deals that involved Amazon investing billions of dollars into them.

Even Nvidia’s Jensen Huang has recently said he was too slow to see the importance of such investing. “Without Anthropic, why would there be Trainium growth at all? It’s 100% Anthropic,” he said last month on the Dwarkesh Podcast. “I didn’t deeply internalize how difficult it would be to build a foundation AI lab like OpenAI and Anthropic and the fact that they needed huge investments from the supplier themselves.”

OpenAI has also announced it restructured its relationship with Microsoft, clearing the way for a relationship with Amazon. On the day of the opening arguments in the Musk-Altman trial, Amazon and OpenAI executives were celebrating an announcement that detailed how AWS customers would be able to use ChatGPT.

“We’re quite excited about what that unlocks for customers,” Garman told me later.

There’s a lot of road left ahead in the AI race.

Write to Tim Higgins at tim.higgins@wsj.com

Source

Leave a Reply

Your email address will not be published. Required fields are marked *

2 − one =