The Rise and Resilience of Cerebras Systems: From Near Failure to AI Chip Powerhouse
Today, Cerebras Systems stands as a prominent public company, renowned for manufacturing cutting-edge AI chips for inference that power technology giants such as OpenAI and AWS. Their recent blockbuster IPO marked a significant milestone, catapulting its co-founders into billionaire status and elevating the company’s valuation to about $60 billion by the end of the week.
However, this success story almost never materialized. Back in 2019, just three years after its founding, Cerebras was on the brink of collapse, burning through cash at an alarming rate while wrestling with a seemingly insurmountable technical challenge.
The Early Struggles: Burning Cash to Solve an Unprecedented Problem
Founder and CEO Andrew Feldman recalled those difficult times in an interview with TechCrunch, saying, “We were spending about $8 million a month. At this point, we had incinerated nearly $200 million trying to solve one technical problem.” Every few weeks, Feldman endured the difficult task of reporting failures to the board, with each update revealing more money lost and no clear solution in sight.
Despite the mounting setbacks, Feldman understood that without overcoming this challenge, Cerebras had no future.
The company was founded on a bold premise. Unlike traditional microprocessors that increase speed and reduce costs by packing more transistors onto silicon wafers and slicing them into smaller chips, Cerebras aimed to revolutionize AI computing by creating a single, massive chip that used an entire wafer. This design promised faster computation by eliminating the need to network multiple smaller chips together, a process that often introduces latency and communication bottlenecks.
Yet, while the concept was elegant on paper, no one in the semiconductor industry had ever succeeded in building such a large, dense chip before. The scale introduced daunting engineering challenges, especially in the packaging phase—everything involved after the silicon wafer was manufactured.
The Packaging Challenge: Innovation Under Pressure
After successfully designing and manufacturing the chip with Taiwan Semiconductor Manufacturing Company (TSMC), Cerebras encountered its most formidable obstacle: packaging. This stage involves attaching the silicon wafer to a motherboard, delivering power, managing heat dissipation, and routing data efficiently—all of which become exponentially more complex with a chip 58 times larger than conventional designs.
Feldman explained, “We were using 40 times as much power as anybody had ever used.” With no existing heat sinks, vendors, or manufacturing partners equipped to handle such scale, the team was forced to innovate from scratch. Decades of attempts by leading microprocessor engineers to build large, dense chips had ended in failure, underscoring the enormity of Cerebras’ challenge.
Trial and error became their only option, resulting in the destruction of countless chips and the depletion of enormous financial resources. Yet, without a functional packaging solution, the chip’s potential was moot.
Through relentless analysis and iteration, Cerebras developed groundbreaking solutions for cooling and data movement. In one remarkable instance, they engineered a custom machine capable of simultaneously bolting in 40 screws to fasten the wafer to the board without causing cracks.
A Defining Moment: Success After Years of Perseverance
July 2019 marked a pivotal point for Cerebras. Feldman vividly remembers the day they powered up the packaged chip for the first time. “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.”
This moment was especially meaningful given the team’s prior success with SeaMicro, a pioneering cloud server startup they had built and sold to AMD for $334 million in 2012.
Image Credits: Cerebras Systems
Strategic Partnerships and Market Position
Interestingly, the day the chip finally worked came about two years after OpenAI had engaged in talks to acquire Cerebras, a fact confirmed by Feldman and corroborated by publicly revealed emails. Though those negotiations faltered amid internal disputes among OpenAI’s founders—some of whom are angel investors in Cerebras—the relationship evolved into a strategic partnership.
Today, OpenAI is both a customer and a partner, having loaned Cerebras $1 billion secured by warrants. These warrants potentially grant OpenAI about 33 million shares of Cerebras stock, which, at Friday’s closing price of $279 per share, are valued at over $9 billion.
As part of this deal, Cerebras agreed not to sell to certain OpenAI competitors—widely speculated to include Anthropic—although Feldman confirmed this restriction is temporary. “It’s limited in time, and it was designed to make sure that we could get OpenAI the capacity,” he said.
Feldman likened selling AI compute capacity to an all-you-can-eat buffet: “Instead of trying to stuff itself on all potential customers, 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.” This measured approach reflects Cerebras’ current capacity constraints and strategic focus on sustainable growth.
When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.
Read more about Cerebras Systems’ journey and recent IPO Here.
