How Cerebras survived burning $8M a month to become a $60B AI chip firm

Today, Cerebras Systems stands as a publicly traded company, supplying AI chips for inference to industry titans like OpenAI and AWS. Its blockbuster IPO on Thursday made both co-founders billionaires, and by the end of the week, the company was valued at roughly $60 billion.

But just a few years ago, in 2019, when the company was only three years old, it teetered on the edge of collapse while burning through an astonishing amount of cash. The mission was to solve a technical challenge that the semiconductor world had long considered impossible.

“We were spending about $8 million a month,” founder and CEO Andrew Feldman told TechCrunch, reflecting on that period. “At this point, we had incinerated nearly $200 million trying to solve one technical problem.”

Every few weeks, Feldman made a painful trek to the boardroom, reporting yet another failure and another pile of money gone.

But he had no real choice. Without a breakthrough, Cerebras was doomed anyway.

The company’s founding idea was simple in theory. For over 50 years, the microprocessor industry had made CPUs faster and cheaper by packing more transistors onto a silicon wafer and slicing those wafers into ever-smaller pieces. But AI demands massive compute power, requiring many chips to be linked together and forced to communicate. Cerebras’ founders believed that turning an entire, larger wafer into one giant, powerful chip would be far more efficient.

The catch: no one had ever successfully done this before, for any reason, AI or not. Orchestrating that many microscopic components on a larger, but still thin, surface created compounding engineering nightmares.

Once Cerebras cleared the first hurdle of designing the mega chip and manufacturing it with TSMC, the real obstacle emerged.

They couldn’t solve packaging. This involves everything after the silicon is made: attaching it to a motherboard, supplying power, managing heat, and creating pathways for data to flow in and out, Feldman explained.

Cerebras’ chips “were 58 times larger. We were using 40 times as much power as anybody had ever used,” he said. No premade heat sinks existed. No vendors. No manufacturing partners. For decades, the brightest minds in microprocessor engineering had tried and failed to build such large, dense chips.

The team resorted to trial and error, a process in which “we destroyed an enormous number of chips” and an enormous amount of cash. Without functional packaging, the chip was useless.

After exhaustive analysis of each failure, the team finally cracked enough problems: how to cool the chip and move data around. In one case, they had to invent a machine that could simultaneously bolt in 40 screws to secure the wafer to a board without cracking it.

Feldman still remembers the day in July 2019 when it all, miraculously, worked.

They installed the packaged chip into a computer, powered it on, and the entire founding team “just stood in the lab and stared at it,” he said. “Watching a computer run is about as exciting as watching paint dry. But there we were watching lights flashing on the computer, stunned that we’d solved this.”

“That was one of the greatest moments of my life,” he added. That’s significant because this same founding team had previously built and sold a pioneering cloud server startup, SeaMicro, to AMD for $334 million in 2012.

The day the chip finally worked also came about two years after OpenAI had approached Cerebras about an acquisition. Feldman confirmed to TechCrunch that the publicly revealed emails describing those talks were accurate.

Those negotiations fell apart amid growing internal disputes among OpenAI’s founders, several of whom are angel investors in Cerebras.

Today, OpenAI is both a customer and a partner. It loaned Cerebras $1 billion, secured by warrants that conditionally grant OpenAI about 33 million shares of Cerebras stock. As disclosed in the S-1, those 33 million shares were worth over $9 billion at Friday’s closing price of $279.

Interestingly, as part of that loan deal, Cerebras also agreed not to sell its products to specific OpenAI competitors. Feldman wouldn’t confirm that the obvious company this involves is Anthropic. He did, however, say the restriction is temporary.

“It’s limited in time, and it was designed to make sure that we could get OpenAI the capacity,” he said.

The reality is that Cerebras hasn’t yet grown large enough to serve multiple fast-growing model makers simultaneously. Feldman compared selling AI compute capacity to an all-you-can-eat buffet. Instead of trying to serve every potential customer at once, “We’re going to work with part of the buffet only, and we’re going to get comfortable with that, before we attack the rest.”