Cerebras Systems filed its S-1 with the SEC on Thursday, marking the first real Nvidia alternative to attempt a public listing during the current AI infrastructure boom. The company plans to list on the Nasdaq under the ticker CBRS, with Morgan Stanley and Citi as lead underwriters. The target valuation range is $22 to $28 billion.

This is Cerebras' second attempt. The company filed for an IPO in September 2024 but withdrew the paperwork late last year, stating that the original filing "no longer reflected the current state of our business." What changed: revenue surged 76% to $510 million in 2025, and the company swung from a $485 million loss to an $87.9 million profit. That is a $573 million swing in one year.

The Chip That Should Not Exist

Cerebras' flagship WSE-3 is built from an entire silicon wafer. It contains 4 trillion transistors organized into 900,000 cores. For context: it is 58 times the physical size of Nvidia's B200, the most popular high-end AI accelerator on the market. It delivers 27 petabytes per second of memory bandwidth, more than 200 times what Nvidia's NVLink interconnect offers.

Every major semiconductor company has attempted wafer-scale chip designs since the 1960s. All failed. Cerebras solved the defect tolerance and thermal management problems that killed every prior attempt. The company was founded in 2015 by five veterans of SeaMicro, a server startup that AMD acquired for $334 million in 2012.

The OpenAI Anchor

The S-1 reveals the full scope of Cerebras' OpenAI relationship for the first time. Last December, OpenAI agreed to purchase 750 megawatts worth of inference infrastructure from Cerebras. That deal is worth more than $20 billion. OpenAI also has the option to add 1.25 gigawatts of additional capacity through 2030.

Cerebras has issued OpenAI warrants for up to 33.4 million shares. Those warrants vest if OpenAI goes through with its plan to purchase 2 gigawatts of capacity by 2030. The filing states the contract "represents a substantial portion of our projected revenues over the next several years." Translation: OpenAI is not just a customer. It is the customer.

AWS also signed a deal last month to deploy the WSE-3 in a "disaggregated architecture" where AWS Trainium chips handle prefill calculations and Cerebras handles decode. This is significant: it means Cerebras is positioning itself not as a Nvidia replacement but as a complementary accelerator that works alongside other chips.

The Risk in the Filing

Customer concentration is extreme. In the first half of 2024, G42 and MBZUAI (both UAE-based) accounted for 86% of Cerebras' revenue. The CFIUS review that killed the original 2024 IPO filing was directly related to this UAE dependency. That review has now been cleared, but the concentration risk remains until OpenAI volumes ramp.

There is also the CUDA problem. Nvidia's software ecosystem has a decade head start. Every AI researcher, every framework, every tutorial assumes Nvidia hardware. Cerebras' software stack works, but it requires developers to learn a new workflow. That is the moat Nvidia has built, and it is deeper than any silicon advantage.

What This Means

If the listing holds at the $22 to $28 billion range, Cerebras would debut as one of the 10 largest semiconductor IPOs in history. Secondary-market pricing on Forge and Hiive recently quoted $102 to $107 per share, implying a pre-IPO valuation of $26 to $28 billion.

The Morgan Stanley credit facility will expand from $125 million to $850 million post-IPO, funding data center expansion for Cerebras' Training Cloud and Inference Cloud services. Add the $1 billion loan from OpenAI, and Cerebras will enter public markets with more than $2 billion in available capital.

Nvidia is not losing sleep tonight. But for the first time in this AI cycle, it has a publicly traded competitor with real revenue, real profits, and a chip architecture that solves a problem Nvidia's does not. That is worth paying attention to.

Sources: SEC S-1 filing, SiliconANGLE, Reuters, TradingView, MENA Fintech Association.