Cerebras Systems: How a California Start-up Is Reshaping the AI Chip Industry

Cerebras was founded in 2016 by five engineers who already knew each other from a previous shared venture: Andrew Feldman (CEO), Gary Lauterbach, Michael James, Sean Lie and Jean-Philippe Fricker. All five had previously worked together at SeaMicro, a company specialising in energy-efficient servers that was acquired by chipmaker AMD in 2012 for around 334 million US dollars. Armed with that experience, they took on a challenge that many experts considered unsolvable: building a working processor that, rather than consisting of a small piece of silicon in the conventional way, uses an entire wafer as a single processor. A wafer is the circular silicon disc from which hundreds of individual chips are normally cut. This concept, known as wafer-scale integration, had been discussed in the semiconductor industry for decades as a theoretical possibility, but was widely regarded as technically unworkable. Cerebras solved the problem in part through software-based redundancy that automatically routes around defective areas of the chip.

The centrepiece of Cerebras is the Wafer Scale Engine 3 (WSE-3), currently the largest commercially available processor in the world. To put its dimensions in perspective: the WSE-3 has a surface area of more than 46,000 square millimetres, contains four trillion transistors, and features 900,000 processing cores along with 44 gigabytes of on-chip memory. By comparison, Nvidia's flagship GPU, the B200, is roughly 58 times smaller. This enormous size allows Cerebras to process AI tasks significantly faster: according to the company, up to 15 times faster for inference (applying trained AI models) and up to ten times faster for training new models. Cerebras does not sell individual chips alone, but complete AI systems through its CS-3 platform, which bundles processor, memory, networking and software. The company also offers computing capacity via the cloud, allowing customers to access the technology without having to purchase their own hardware.

Cerebras counts some of the biggest names in technology among its most important customers and partners. In the first quarter of 2026, the company signed a multi-year agreement with OpenAI worth more than 20 billion US dollars, under which OpenAI intends to deploy 750 megawatts of Cerebras infrastructure. In March 2026, a partnership with Amazon Web Services followed, giving AWS customers access to Cerebras technology. The company's revenue grew by around 76% in 2025 to 510 million US dollars, with roughly 70% coming from hardware sales and the remainder from cloud services. The profit trajectory is particularly noteworthy. After a loss of nearly half a billion dollars the previous year, Cerebras reported its first net profit in 2025, amounting to around 238 million dollars. Historically, the company was heavily dependent on customers from the United Arab Emirates: the technology group G42 accounted for 85% of revenue in 2024. That share fell sharply to 24% in 2025, pointing to a broader and healthier customer base.

On 14 May 2026, Cerebras listed on the Nasdaq under the ticker symbol CBRS, surpassing all expectations. The company priced its shares at 185 US dollars after the initial price range had been raised several times. At the opening of trading, the stock jumped to 350 dollars, a gain of 90%. In total, Cerebras raised around 5.5 billion dollars. Observers described the listing as the largest semiconductor IPO of all time and an important sentiment test for the valuation of AI companies on public markets. CEO Andrew Feldman, who holds around 4.6% of the company, became a billionaire overnight as a result of the listing. Despite the euphoria, challenges remain. The valuation is extremely high, with a price-to-sales ratio of around 130, the market continues to be dominated by Nvidia, and Cerebras must prove that it can sustain its growth momentum over the long term. One thing, however, is clear: the company from Sunnyvale has transformed itself from an unknown start-up into a serious contender in the global AI race in record time.