By Ben Cohen

The entrepreneur looked around a Seattle restaurant for a booth where he could have a private conversation. As the co-founder of Annapurna Labs, a secretive Israeli chip-design startup, Nafea Bshara was used to operating in stealth mode. His business was so allergic to publicity that it barely even had a website.

But he was being especially discreet that night because his clandestine meeting was with an influential executive from one of the world's most valuable companies.

And it would result in one of the most consequential deals in tech history.

Their discussion of chips that began over beer and wine eventually led to Amazon buying the mysterious startup for about $350 million. Ten years later, it has become essential to the success of the whole company.

Amazon has long depended on Amazon Web Services -- and Amazon Web Services depends on Annapurna.

The company's entire AI strategy is now built on a foundation of chips designed by Annapurna, which is so crucial that analysts have described this custom silicon as the secret sauce of AWS.

The person who might just have the deepest understanding of Annapurna's value is Amazon's chief executive. Before he succeeded Jeff Bezos, Andy Jassy led the company's giant cloud-computing business and made this deal.

"If and when they go back and tell the story of AWS," Jassy told me in an email, "our acquisition of Annapurna was one of the most important moments."

It looks even more important now that the AI boom has sparked a trillion-dollar arms race with Microsoft and Google also investing gigantic sums of money in powerful chips -- the brains of artificial intelligence. All the cloud titans are building their own custom hardware, partly to chip away at their reliance on Nvidia. And no company is spending more than Amazon.

This year, Amazon is planning more than $100 billion of capital expenditures, mostly on the AWS infrastructure required for AI systems. It's even building a colossal supercomputer trained on an "ultracluster" of advanced chips designed by Annapurna. All this homemade silicon is the reason it can offer faster, cheaper and more efficient computing.

Basically, a company once known for books is now increasingly obsessed with chips.

And its strategy today was quietly made possible a decade ago with the prescient acquisition of a company that you've never heard of.

Annapurna was born in 2011 a long way from the research lab in Austin, Texas, where it now designs and tests the next generation of chips. It was founded in Israel by entrepreneurs who had decades of experience in the chip business when they decided to start a company together.

At the time, most people in tech were tantalized by consumer products -- phones, tablets and mobile gadgets. Bshara and his Annapurna co-founders were drawn more to infrastructure products -- chips, servers and behind-the-scenes hardware. After all, it's hard to build successful consumer products outside America, so Israeli companies tend to focus on this overlooked but lucrative niche of the tech industry. "We're infrastructure people," says Bshara, now an AWS vice president based in Silicon Valley.

And they were in the right place at what turned out to be precisely the right time. As cloud computing exploded over the next decade, so would the demand for those infrastructure products.

They named their startup Annapurna after the treacherous mountain in Nepal they were planning to climb until they became too busy doing something else that would take them into a cloud.

As it happens, Amazon had its own plans that sounded as daunting and crazy as scaling one of the world's most dangerous peaks.

You might think of Amazon as the everything store where you order stuff to arrive at your front door the next day. But these days, it prints money from other parts of the business. In fact, the AWS cloud division accounted for more than half of Amazon's profits last year and crossed $100 billion in revenue, which makes it bigger than some of America's biggest companies.

When Annapurna was founded, Amazon was just beginning to think about developing semiconductors to provide AWS customers with more options and lower prices -- and take greater control of its supply chain.

The man pushing innovation down to the silicon level was James Hamilton, an Amazon executive with the title of "distinguished engineer" and a shock of white hair worthy of that designation. He spends his days thinking about the sort of infrastructure the rest of us never think about. "For many people, it probably isn't that interesting," Hamilton told me. "For me, it's fundamentally captivating."

His bosses were intrigued when he wrote a memo for Bezos and Jassy in 2013 arguing it was time for Amazon to design chips.

At the time, a lot of people thought this was completely nuts.

The company might as well have decided to construct data centers out of Pringles.

Jassy knew it would be expensive and extraordinarily challenging. "Building a chip," he says, "is not for the faint of heart." He also knew that Amazon would need a partner to pull it off.

And then Nafea Bshara walked into a bar.

Not long after writing his fateful memo, Hamilton left work one afternoon and went to the Virginia Inn in Seattle's famous Pike Place Market for his meeting with Annapurna's co-founder.

Bshara didn't bring his computer because he figured it would be awkward to present from a laptop in a bar. Instead he visited a nearby UPS store and printed four slides about Annapurna that explained who they were, what they did and how they did it -- and why they should do it for AWS.

After working together, even the geekiest AWS employees were so impressed by Annapurna that Jassy soon picked up more than just a bar tab.

He struck a deal in 2015 to buy the whole company.

When a startup gets acquired by a tech giant, it's unusual for most of the team to stick around for a decade. But today, 68% of Annapurna's employees at the time of the acquisition are still there. It's also unusual for the startup's founder to compare his workplace to the happiest place on earth. "We're the Disneyland of computer architecture and silicon innovation," Bshara says.

Hamilton says Annapurna has been such an unusually seamless fit inside Amazon for several reasons, including some that apply outside Amazon.

One was that Annapurna engineers appreciate the importance of speed. "They know that months matter, weeks matter more and days matter even more," Hamilton said. But even as they move fast, they can't afford to break things. "If you make a mistake in software, it could cost you a week or two to fix it," he said. "With hardware, you can lose nine months to a year."

These days, the most exciting product coming out of Annapurna's labs is a chip for training AI models called Trainium.

There will be hundreds of thousands of them in the supercomputer Amazon is delivering to AI startup Anthropic this year, a hulking machine named Project Rainier after the massive peak closer to HQ than the Himalayas.

Some chips in AWS data centers are designed by Annapurna, like Trainium and Graviton, its popular central processor for general-purpose computing. Some are made by other companies, like Intel and chip kingpin Nvidia, whose enviable products handle the vast majority of AI workloads. The idea behind this buffet of chips is to provide a diversity of options at different prices depending on customer needs -- something for everybody, just like the Amazon store.

So a lot has changed since the first time Bshara and Hamilton met privately in Seattle.

But they recently met again to discuss Annapurna and Amazon. Except this time, they were in Israel. Jassy was there, too. So was Matt Garman, the CEO of AWS. And they were surrounded by employees who had a reason to celebrate.

It had been exactly 10 years since a deal that turned beer and wine into chips.

Write to Ben Cohen at ben.cohen@wsj.com

(END) Dow Jones Newswires

May 09, 2025 21:00 ET (01:00 GMT)

Copyright (c) 2025 Dow Jones & Company, Inc.