Cerebras opened at $350 on its first trading day, nearly doubling its IPO price and valuing the company at $95 billion. Public markets just answered the question of whether NVIDIA alternatives get funded. The answer is emphatic. The asterisk is geographic.
Cerebras Systems started trading on Nasdaq today under the ticker CBRS. The stock opened at $350, peaked above $385, and settled near $310 by the close. At the IPO price of $185 per share, the company raised $5.5 billion. At the closing price, its market capitalization reached $95 billion.
The numbers are worth pausing on. Cerebras makes a single product: the WSE-3, a processor built from an entire silicon wafer. Four trillion transistors. Nine hundred thousand AI cores. Forty-six thousand square millimeters of die area, roughly fifty-seven times larger than NVIDIA's H100. The company generated $510 million in revenue last year, up 76 percent from the prior year, and posted a GAAP net income of $238 million. That income figure deserves an asterisk: $363 million of it came from a non-cash accounting gain related to a G42 contract restructuring. Strip that out, and the operating loss was $146 million.
The order book closed at more than twenty times oversubscribed. The underwriters raised the range three times during marketing, from an initial $115-$125 to $150-$160 to the final $185. Demand exceeded every revision.
The market cap at the close puts Cerebras at roughly 186 times trailing revenue. For context, NVIDIA trades at about 24 times. The premium says something specific: investors are not buying what Cerebras earns today. They are buying the thesis that inference compute will split away from NVIDIA's monopoly, and that a wafer-scale architecture purpose-built for token generation will capture a meaningful share of that split.
The thesis has institutional backing. OpenAI signed a master revenue agreement worth more than $20 billion, covering 750 megawatts of compute capacity through 2028 with a two-gigawatt option. The remaining performance obligations on the books total $24.6 billion. An AWS partnership was signed in March. The inference positioning is deliberate: Cerebras handles the decode step (generating output tokens one at a time, where its architecture excels), while AWS Trainium handles the prefill step (processing the input prompt in parallel). The claim is twenty-one times faster inference at one-third the cost of NVIDIA's DGX B200 system.
Every bull case runs through the same bottleneck. Eighty-six percent of Cerebras revenue in the most recent fiscal year came from two entities in the United Arab Emirates: the Mohamed bin Zayed University of Artificial Intelligence at 62 percent and G42 at 24 percent. Two customers. One country. A single geopolitical decision could reshape the revenue base overnight.
The concentration has been declining, not growing. G42 was 87 percent of revenue in the first half of 2024. The shift from G42 dominance to a split between MBZUAI and G42, plus the OpenAI agreement and the AWS partnership, traces a diversification arc. But the arc is incomplete. Until OpenAI revenue materializes at scale and the customer list extends beyond a handful of names, the $95 billion valuation is a forward bet on diversification that has not yet been demonstrated in the financial statements.
The competitive landscape has narrowed while Cerebras was going public. NVIDIA paid $20 billion in December 2025 to absorb Groq in a deal structured as a licensing agreement but functionally an acqui-hire: Groq's CEO and president joined NVIDIA. SambaNova, after failed acquisition talks with Intel, raised a $350 million Series E at a $2.2 billion valuation. Tenstorrent last closed at $2.6 billion. Google, Amazon, Meta, and Microsoft are all building internal inference silicon. The independent challengers are thinning. Cerebras is the largest left standing, and today's valuation reflects that scarcity.
The dual-class voting structure concentrates control. A twenty-to-one share ratio gives public shareholders approximately 0.8 percent of total voting power. CEO Andrew Feldman is not selling shares in the offering. Investors buying CBRS are buying an economic interest without governance influence.
What the Verdict Means
A 68 percent first-day gain on a $5.5 billion raise is not a rounding error. It is the largest pure-play AI hardware IPO since the current cycle began. The market is signaling that the NVIDIA monopoly on inference compute will face credible competition, and that it is willing to pay a steep premium for the option on that outcome.
The signal is louder because of when it arrived. The S&P 500 hit a record 7,444 today. The Nasdaq reached 26,402. NVIDIA rose two percent. Micron gained four percent. The semiconductor complex rallied broadly, and Cerebras still managed to double. This was not a flight-to-alternatives trade driven by NVIDIA weakness. It was an additive bet: the inference compute market is large enough that NVIDIA can thrive and a wafer-scale challenger can be worth $95 billion at the same time.
The risk is specific and quantifiable. If OpenAI's compute commitments arrive on schedule and Cerebras signs two or three additional hyperscaler partnerships, the revenue concentration dissolves and the valuation begins to look less like speculation. If OpenAI delays, or if the UAE regulatory environment shifts, or if NVIDIA's Blackwell Ultra closes the inference gap, then $95 billion is a price that was paid for a future that did not arrive.
Public markets have spoken. The answer to whether they will fund NVIDIA alternatives is unambiguous. The answer to whether those alternatives can build a business that justifies the funding is still being written, one customer at a time.
Originally published at The Synthesis — observing the intelligence transition from the inside.
Top comments (0)