Trainium is Amazon's in-house AI chip, first launched in 2020, which the company has touted as the most "cost-effective" chip for training AI models

"Matt Garman, the new chief executive of Amazon's cloud unit, says artificial intelligence is a race without a finish line.

"It is such a foundational technology," Garman said. "It's just a thing that's going to happen forever."

Garman, who became CEO of Amazon Web Services in June, is about to headline his first re:Invent, the company's annual conference, which regularly attracts tens of thousands of attendees to Las Vegas.

Each year, it is tradition for the world's largest cloud provider to invite its chief executive to rattle off a slew of product updates in a keynote address. On Tuesday, Garman will be sharing new features and products in areas including AI and computing, which he described as "real, needle-moving changes."

The changes that Amazon unveils do need to be revolutionary, some analysts and customers say. The cloud giant has been perceived as falling behind its tech rivals in the AI race, though it has continued to introduce new AI capabilities targeting business customers.

Even in overall cloud-computing, long Amazon's domain, its rivals are catching up. Amazon's share of the global public cloud-computing market fell to 39% in 2023 from 39.9% in 2022, according to research and consulting firm Gartner. Meanwhile, its rival Microsoft gained market share, rising to 23% from 21.5% in those years, while Google rose to 8.2% from 7.5%.

Andy Jassy, Amazon's chief executive and former CEO of AWS, has said he expects AI to drive tens of billions of dollars of revenue over the next several years for the company, yet Amazon has also tamped down expectations by repeatedly saying it is the "early days" of AI.

Garman echoed that sentiment, saying it is "still incredibly early" in how far generative AI will go. Still, he said Amazon's AI strategy has remained the same since the AI boom kicked off about two years ago.

"We felt very strongly that there was not just going to be one model, but there's going to be a lot of models that people wanted to lean into," Garman said, referring to Amazon's tactic of acting as a neutral platform that offers businesses options for AI models, a sort of Switzerland among the cloud giants. Like Amazon has done in retail, AWS is aiming to be a bit of everything for everyone in AI models.

In late November, the company announced that it invested an additional $4 billion in Anthropic -- doubling its investment in the AI startup to $8 billion. Google last year agreed to invest up to $2 billion in Anthropic. Microsoft has invested $13.75 billion into OpenAI, with which News Corp, owner of The Wall Street Journal, has a content-licensing partnership.

"We thought it was a good investment, and wanted to keep investing in that," Garman said about Anthropic. "And then from a product point of view, they're also really leaning in with us on Trainium."

Trainium is Amazon's in-house AI chip, first launched in 2020, which the company has touted as the most "cost-effective" chip for training AI models. As part of its new investment, Anthropic will use Amazon's chips to train and run its future AI models, the companies said.

Amazon is also planning to spend more than $100 billion over the next decade on data centers -- devoting more investment money to its cloud computing and AI infrastructure than to its sprawling network of e-commerce warehouses.

AWS's custom chips are part of how the company is lowering the cost of AI for clients, Garman said, along with making its services and hardware more capable.

The question of return on investment for AI initiatives has dogged chief information officers and other corporate technology leaders since at least this year, and analysts expect costs and value of AI to be a big part of the conversation at re:Invent this year.

There are also complaints that the cost of cloud has grown as companies use more of it for AI. Gartner forecasts that each area of cloud spending is expected to grow by a double-digit percentage in 2025 as technology leaders prepare their information-technology infrastructure for AI.

"Almost every CEO or CIO that I talk to, they're basically saying, 'Look, my organization did 100, 200 proof of concepts,'" Garman said, referring to companies' AI experiments. "Then they say, 'How do I go find the one, two, five of those proof of concepts that are valuable, and that are delivering real ROI?"" [1]

1. Amazon Web Services CEO Promises 'Needle-Moving' AI-Product Features. Lin, Belle.  Wall Street Journal, Eastern edition; New York, N.Y.. 03 Dec 2024: B.4.