FEATURED

Cerebras (CBRS) Just Hit Nasdaq

The IPO market hasn’t seen demand like this in years. Cerebras Systems (CBRS) priced at $185 per share on Wednesday evening — well above the $150–$160 range it moved to after the original $115–$125 band proved far too low — and shares were indicated to open near $350 this morning. Nearly double the IPO price before the first trade cleared.

Thirty million Class A shares. $5.55 billion raised. A fully diluted valuation of $56.43 billion, per Reuters. The largest U.S. technology IPO of 2026, by a wide margin. The order book reportedly drew more than 20 times the available shares.

Worth noting: Arm Holdings and SoftBank approached Cerebras with an acquisition offer in the weeks before the IPO. Cerebras turned it down. That tells you something about where the AI hardware supply chain thinks this technology sits.

What Cerebras Actually Builds

Founded in 2016 and headquartered in Sunnyvale, California, Cerebras takes a fundamentally different approach to AI compute. Rather than networking thousands of smaller GPU chips together, its flagship Wafer-Scale Engine 3 (WSE-3) uses an entire 300mm silicon wafer as a single processor. The result: 4 trillion transistors, 900,000 compute cores, and 44 GB of on-chip SRAM — roughly 57 to 58 times larger than Nvidia’s H100 by die area.

The company claims the WSE-3 delivers up to 15x faster inference than leading GPU-based systems on open-source models, while consuming less power per unit of compute. CEO Andrew Feldman put it plainly this morning: “We built a chip the size of a dinner plate.” The pitch is speed and efficiency for inference workloads — not necessarily replacing Nvidia across the full AI stack, but winning where latency is the bottleneck.

Revenue comes from four streams: on-premises CS-3 AI supercomputer hardware, cloud-based AI inference services, multi-year compute capacity agreements with hyperscalers, and software subscriptions. Hardware was $358.4 million of 2025 revenue; cloud and services contributed $151.6 million — roughly 30% of the total, a mix that matters for long-term margin expansion.

The Numbers — With Full Context

• 2025 Revenue: $510 million, up 76% year over year (from $290.3M in 2024)
• Hardware Revenue: $358.4M | Cloud & Services: $151.6M
• Gross Margin: 39.0% in 2025 (down from 42.3% in 2024 as cloud infrastructure spending scaled)
• GAAP Operating Loss: $145.9 million
• GAAP Net Income: $237.8M — driven almost entirely by a one-time $363.3M non-cash gain on extinguishment of a G42 forward-contract liability
• Non-GAAP Net Loss: $75.7 million (after removing stock-based comp and the one-time gain)
• Remaining Performance Obligations: $24.6 billion — the vast majority tied to the OpenAI MRA
• IPO Price: $185/share | Fully Diluted Valuation: $56.43 billion
• Indicated Open: ~$350–$361 per share
• Oversubscription: 20x+ at the original range

The 47% net margin figure that circulated in early coverage requires a correction. The GAAP profit is real on paper — but it was manufactured almost entirely by that $363.3 million non-cash gain from restructuring the G42 liability. Strip that out, and the operational picture is a $145.9 million operating loss and a $75.7 million non-GAAP net loss. The revenue growth is genuine. The profitability framing is not.

The Customer Concentration Problem

This is the part that deserves more attention than it’s getting in the day-one excitement.

According to the S-1 filing, 86% of Cerebras’s $510 million in 2025 revenue came from just two Abu Dhabi-linked entities: Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) at 62% and G42 at 24%. G42 had previously accounted for 85% of 2024 revenue on its own, so the shift represents progress — but it’s not diversification in any conventional sense. It’s a reallocation within the same geographic and institutional base. MBZUAI alone held 77.9% of accounts receivable at year-end 2025.

There’s an additional data point that sharpens the concern: U.S.-billed revenue actually declined 34% year over year, falling from $282.7 million in 2024 to $187.6 million in 2025. The company’s headline growth story was entirely driven by international — specifically Abu Dhabi — bookings. Any tightening of U.S. export controls on AI compute to the Middle East, or any shift in UAE capital allocation, would hit the revenue base hard and fast.

Slight tangent, but it matters: Cerebras’s original 2024 IPO attempt was blocked by a CFIUS national-security review tied to G42’s investment in the company. G42 was removed from the cap table in March 2025 to clear that review. The revenue relationship with G42-linked entities, however, remains very much intact.

The OpenAI and AWS Relationships

The commercial case for diversification rests on two relationships that are very early stage relative to the UAE revenue base — but structurally significant.

In December 2025, Cerebras and OpenAI signed a Master Relationship Agreement under which OpenAI committed to purchase 750 megawatts of Cerebras inference compute capacity, with an option to expand to 2 gigawatts by 2030. The deal is valued at more than $20 billion. OpenAI also extended a $1 billion working capital loan at 6% interest — with accrued interest waived if repaid through capacity delivery — and received warrants for up to 33.4 million Cerebras shares. Earlier this year, OpenAI launched its first AI model running on Cerebras chips. That’s not a theoretical relationship. It’s operational.

On the AWS side: in March 2026, Amazon Web Services announced a deal to deploy Cerebras CS-3 systems in its data centers, making Cerebras-powered inference available through Amazon Bedrock. The S-1 describes this as a binding term sheet rather than a final agreement — important distinction. It carries no contractual revenue commitment yet. But if it converts to a production deployment, it would represent the first meaningful hyperscaler distribution channel beyond the UAE anchor.

The $24.6 billion in remaining performance obligations on the books is the most important forward-looking number in the filing — and the vast majority of it is attributable to the OpenAI MRA. That backlog is what investors are buying at $56 billion.

Macro and Industry Context

The IPO market had a slow start to 2026 — only 35 companies went public in Q1, the fewest since Q1 2024, amid a risk-off backdrop and volatility tied to the conflict in Iran. The second quarter has opened differently. Cerebras lands into a market where AI-linked names have pushed the Nasdaq to record highs and where semiconductor stocks have moved sharply higher on infrastructure spending demand. The timing, whatever one thinks of the valuation, is not accidental.

The broader AI hardware race is relevant here. Cerebras is not positioned as a direct Nvidia replacement across the full stack — it operates outside the dominant CUDA ecosystem, which is both a competitive differentiator and a meaningful adoption barrier. Its strongest case is in inference: running already-trained models at extremely high speed for latency-sensitive applications. That market is growing fast. Whether Cerebras can scale within it while managing the OpenAI and AWS deployment obligations simultaneously is the execution question that the first several earnings calls will need to answer.

Bull / Base / Bear

Bull: OpenAI accelerates its 750MW deployment on schedule, the AWS Bedrock term sheet converts to a full production agreement, and a third hyperscaler signs. Cloud and services revenue grows as a percentage of the mix, improving gross margins above the current 39%. The $24.6 billion backlog gets pulled forward. Revenue diversification away from UAE customers becomes visible in the first two to three post-IPO earnings reports. At that point, the valuation argument changes substantially.

Base: OpenAI deployment proceeds but slowly. AWS remains a term sheet for longer than expected. UAE revenue stays elevated as a percentage of the mix through 2026. The stock trades in a wide range as investors wait for concrete proof points on diversification — locked-up insiders add overhang as the mid-2026 lock-up expiration approaches.

Bear: U.S. export-control policy tightens on AI compute shipments to the UAE, directly impacting the 86% revenue base. OpenAI delays or reduces capacity draws, triggering potential termination clauses in the MRA. Nvidia’s CUDA ecosystem proves stickier than anticipated with hyperscalers, and the AWS deal never moves beyond a term sheet. At a valuation that prices in years of execution before the company has demonstrated cash generation from a diversified base, there is very limited margin for error.

Technical Levels to Watch

With CBRS indicated to open near $350–$361 against a $185 IPO price, the first technical reference is the opening print itself. IPO-day gaps of this magnitude rarely offer clean entry — the first 30 to 60 minutes typically establish a range that becomes the short-term anchor. Watch for a pullback to the $280–$300 zone as the first potential support level if early momentum fades. The IPO price of $185 is the structural floor for the near term; underwriters hold the greenshoe option (up to 4.5 million additional shares) to provide stabilization support if needed. Longer-term, the lock-up expiration in mid-2026 is the next meaningful technical event.

What to Watch Post-IPO

• First earnings call: Quarterly revenue breakdown by customer — specifically the UAE vs. U.S. mix and whether OpenAI revenue has become a visible line item
• AWS Bedrock conversion: Term sheet to definitive agreement; any production deployment announcement would materially shift the customer diversification argument
• OpenAI deployment cadence: Is the 750MW capacity coming online on schedule? Any MRA modification or delay is a material event
• Export control developments: Bureau of Industry and Security (BIS) policy on AI chip exports to the UAE; Cerebras holds export licenses for CS-2, CS-3, and future CS-4 systems, but policy risk is real
• Gross margin trajectory: The 39% gross margin declining from 42.3% bears watching — cloud infrastructure build-out costs need to stabilize for the unit economics story to work
• Lock-up expiration (mid-2026): Insiders hold 99.2% of voting power immediately post-IPO via a Class B super-voting structure; insider selling behavior post-lock-up will be a significant signal

Bottom Line

Cerebras is a genuinely differentiated technology company with a real architectural advantage, a $20 billion-plus anchor contract, and 76% revenue growth. Those things are true. What is also true: 86% of that revenue comes from two Abu Dhabi-linked institutions, U.S. domestic revenue fell 34% last year, the operating business ran at a loss, and the headline GAAP profit was manufactured by a one-time non-cash accounting gain.

At a $56 billion valuation — potentially higher after today’s open — investors are paying for a version of Cerebras that doesn’t exist yet: a diversified, cash-generating AI infrastructure platform with multiple hyperscaler relationships. The OpenAI deal and the AWS engagement are the bridge to that version. Whether they get there fast enough to justify the price is the only question that matters.

The opening trade doesn’t answer it. The first few earnings calls will.

For informational purposes only.