Cerebras surged over 68% in its debut. Do you have confidence in its future performance?

深潮 TechFlow

joined discussion ·

Cerebras IPO: $48.8 billion valuation, is the 'NVIDIA challenger' a bubble or the new king?

Written by: Xiaohei, DeepTide TechFlow

Priced on May 13, with trading to begin on May 14; Nasdaq ticker symbol CBRS.

This is the largest IPO globally in 2026 so far. The underwriting syndicate consists of Morgan Stanley, Citi, Barclays, and UBS Group. During the roadshow phase, they secured 20 times oversubscription, which pushed the offering price from an initial range of $115-$125 up to $150-$160. The expected fundraising amount is $4.8 billion, corresponding to a valuation of $48.8 billion.

Just three months ago, Cerebras' secondary market valuation was at $23 billion. In other words, in the final stretch before the IPO, the company's book value more than doubled.

The "selling point" of this story has been repeated ten thousand times: the NVIDIA challenger, wafer-scale chips, inference speeds 21 times faster than the B200, and a computational power contract with OpenAI starting at $1 billion, with a potential maximum of $20 billion. This is a perfect script for an 'AI challenger,' featuring technological narrative, geopolitical narrative, star clients, and massive orders, each element precisely aligned with the main theme of AI infrastructure in 2026.

However, reading through the S-1 filing page by page reveals something strange: all public reports tell the same story, but the prospectus tells another.

Breaking down the prospectus item by item, Cerebras presents itself as a target defined by a "triple paradox."

First Paradox: Technologically it represents true Alpha, financially it's accounting magic.

The prospectus discloses: $510 million in revenue for 2025, a year-over-year increase of 76%, with GAAP net profit of $237.8 million. This sounds very impressive—a rapidly growing and already profitable AI hardware company, which is almost a "mythical-level" asset in today's valuation environment. CoreWeave was still loss-making when it went public in March this year, while Cerebras directly delivered a net margin of 47%.

However, the $237.8 million in "net profit" includes $363.3 million from a one-time, non-cash accounting adjustment related to the extinguishment of forward contract liability associated with G42. Excluding this amount and adding back $49.8 million in share-based compensation, the true non-GAAP net loss for 2025 is $75.7 million, which is 247% worse than the $21.8 million loss in 2024.

In other words, what the market sees is the "profitable + 76% growth" IPO golden boy, while the prospectus reveals a "rapidly growing company with expanding losses." Neither version is wrong; the difference lies in which one the market chooses to believe.

The second layer: superficially free from G42, but in reality replaced by OpenAI's nested loop.

The story of Cerebras' first failed IPO in 2024 is not complicated: G42, a client with UAE ties, contributed 85% of the first-half revenue. Following an investigation initiated by CFIUS, the company was forced to withdraw its application.

A year and a half later, they tried again. The customer list appeared more diversified, now including heavyweight clients like OpenAI and AWS. But when you look at the May 2026 S-1 filing, the 2025 customer composition looks like this:

MBZUAI (Mohammed bin Zayed University of Artificial Intelligence): 62%

G42: 24%

Combined total: 86%

G42 simply passed the "weight" to MBZUAI, which is also located in the UAE and is affiliated with G42. MBZUAI, as a single client, accounts for 77.9% of accounts receivable.

And this so-called "lifeline" from OpenAI itself is a nested structure. This contract is worth over 20 billion US dollars, with OpenAI committing to purchase 750 megawatts of computing power. However, the same document also discloses several things: OpenAI provided Cerebras with a 1-billion-dollar loan; OpenAI received nearly free warrants for 33 million shares of Cerebras; OpenAI's Master Relationship Agreement includes exclusivity clauses restricting Cerebras from selling to certain "named competitors."

In other words, OpenAI is simultaneously Cerebras’ customer, lender, upcoming shareholder, and to some extent, a strategic controller. An anonymous analyst made a harsh comment on an analysis published on Medium:When revenue is cyclical, valuations are cyclical, and IPOs are designed for those who generate the revenue to cash out, this isn't a market—it's financial engineering.

The wording may be too sharp, but in terms of facts, this statement is hard to refute.

Third layer: On the surface, it appears to be NVIDIA's 'challenger,' but in essence, it is NVIDIA's 'narrow-band filler.'

This point is most easily overlooked by the market.

Cerebras' technology is indeed solid. The WSE-3 boasts 4 trillion transistors, 900,000 AI cores, and 44GB of on-chip SRAM, turning an entire wafer into a single chip and bypassing the inter-chip communication bottlenecks that all GPU clusters must face. Independent benchmark tests from Artificial Analysis show that running Llama 4 Maverick (400 billion parameters), the CS-3 outputs over 2,500 tokens per second per user, while NVIDIA’s flagship DGX B200 achieves around 1,000 tokens. Groq and SambaNova achieve 549 and 794 tokens respectively.

Numbers don't lie,Cerebras has a generational advantage over GPUs in this specific scenario of inference.

The keyword here is 'inference.' Cerebras’ own prospectus clearly states that its strength lies in latency-sensitive inference workloads. It does not have the ability or intention to challenge NVIDIA in large model training or general-purpose computing. The CUDA ecosystem, which has accumulated nearly 20 years since 2007, including toolchains for model training, developer communities, and third-party libraries, remains within NVIDIA's moat.

More crucially, the market hasn’t stood still. NVIDIA unveiled the Vera Rubin architecture at GTC 2026, with 336 billion transistors and performance claimed to be five times that of Blackwell; AMD MI400 has already reached 320 billion transistors; Google TPU v6, Amazon Trainium 3, and Microsoft Maia 2—hyperscale vendors are all developing their own chips. NVIDIA’s R&D investment for the fiscal year 2025 exceeds $18 billion, and last December it spent $20 billion acquiring assets from AI inference startup Groq, followed by a $4 billion investment in two photonics technology companies in March.

Therefore, a more accurate description would be: Cerebras is not looking to replace NVIDIA; rather, it is seizing a differentiated position in NVIDIA’s 'inference' narrow band. This is a real business opportunity, but a valuation of $48.8 billion against $510 million in revenue implies a price-to-sales ratio of 95 times.

Beyond the numbers, we need to talk about the soul of this company.

Andrew Feldman is an underrated 'serial entrepreneur' in Silicon Valley. He is not a technical genius founder and didn’t come from an ivory tower. After graduating from Stanford Business School, he served as the Vice President of Marketing at Riverstone Networks (which went public in 2001) and as the Vice President of Products at Force10 Networks (which was acquired by Dell for $800 million in 2011).

In 2007, he co-founded SeaMicro with Gary Lauterbach to build 'energy-efficient servers,' stacking clusters of low-power processors with small cores to compete against mainstream high-power servers with large cores at the time. This idea was very avant-garde, but the market wasn't ready. In 2012, AMD acquired SeaMicro for $334 million. Feldman worked as a VP at AMD for two years before leaving.

Then he founded Cerebras.

Looking at Feldman’s career path as a whole, one interesting thing stands out: he is not a 'chip designer,' but rather an 'alternative bettor on compute infrastructure.' SeaMicro was a bet that 'small cores would beat big cores,' which was only partially wrong. AMD bought it intending to use its Freedom Fabric interconnect technology to develop its own server CPU platform, but that route didn’t pan out, and the SeaMicro brand quietly disappeared. Cerebras, on the other hand, was a bet that 'big chips would beat small chips,' exactly the opposite proposition of SeaMicro.

In a sense, Feldman has been doing the same thing: finding overlooked or seemingly 'impossible' paths in computing architecture, making bold bets, and then pushing them into the market with strong sales capabilities. Back in the SeaMicro days, he managed to bring the sales team from Force10 under his control; AMD was interested in his sales network. With Cerebras, the most important thing he did right this time was securing G42, allowing a hardware company that still derived 80% of its revenue from a single Middle Eastern customer in 2024 to eventually sign a $20 billion contract with OpenAI.

The footnote to this story is: Feldman is a product-sales-focused CEO, not a visionary technologist CEO. He excels at selling a 'seemingly crazy' product to customers willing to pay a premium for differentiation — this is his alpha.

Understanding this point is crucial because it directly impacts the judgment of Cerebras’ investment value.

Looking at the three layers of paradox stacked together, the answer is actually much more complex than simply 'buy' or 'don't buy'.

If the goal is to capitalize on the first-day IPO pop, with 20 times oversubscription, AI hardware being the hottest track, and a lack of pure NVIDIA alternative listings, CBRS will likely spike on the first day. This is event-driven short-term trading that doesn’t require deep analysis.

However, if making a 'long-term hold' investment decision, there are three things to consider carefully:

First, is Cerebras worth a price-to-sales ratio of 95 times?

CoreWeave went public in March this year with a price-to-sales ratio of around 15 times. NVIDIA’s current price-to-sales ratio is about 25 times. A company with projected revenue of $510 million in 2025, an 86% customer concentration, and still operating at a loss, priced at 95 times sales implies the market expects it to grow its revenue to $3 billion to $4 billion within three to four years while achieving consistent profitability.

Can this be achieved? The key lies in whether OpenAI’s $20 billion contract can materialize as planned. According to the prospectus, about 15% of remaining performance obligations, roughly $3.5 billion, are expected to be recognized by 2026 and 2027. If this pace holds, Cerebras’ revenue could exceed $2 billion by 2027, potentially bringing its price-to-sales ratio down to a reasonable range. However, any delay, strategic shift by OpenAI, or loss of a major client could make this valuation instantly unsustainable.

Second, how wide is Cerebras’ moat?

The architectural advantage of WSE-3 is real, but how long will this edge last? NVIDIA Vera Rubin, AMD MI400, and Google TPU v6 are all advancing. The semiconductor industry has generational cycles of 18-24 months. If Cerebras slows down even slightly, its technological lead could be erased. Its R&D spending as a percentage of revenue is already high, but in absolute terms, it remains orders of magnitude behind the giants.

A deeper question is: Is the wafer-scale chip route a mainstream path that will be widely adopted, or is it a 'special forces' unit that can only survive in niche scenarios? There is no definitive answer to this. The optimistic view is that when inference workloads rise from 30% of total AI computing today to over 70% in the future, Cerebras’ niche could become the main battlefield. The pessimistic view is that if NVIDIA simply boosts Rubin’s inference performance, the niche will remain just a niche.

Third, governance structure and geopolitical risks.

The prospectus revealed two easily overlooked but important points:

First, Cerebras adopts a dual-class share structure with Class A and Class B shares. After the IPO, insiders will hold 99.2% of the voting rights. Even if the founding team only holds 5% of the outstanding shares in the future, they will still control the company. This means that external minority shareholders will have almost no say in corporate governance.

Second, the company disclosed two 'material weaknesses in internal control over financial reporting.' As an emerging growth company, it can be exempt from the SOX 404(b) auditor attestation for up to five years after the IPO. This is a warning sign—though not a major one, it's worth noting.

On the geopolitical front, CFIUS has resolved the voting rights issue with G42, but export controls (shipment licenses for CS-2, CS-3, and CS-4 to the UAE) remain a long-term variable. The Trump administration’s policy direction on AI chip exports to the Middle East has yet to stabilize completely, and any policy shift could reignite tail risks for CBRS.

CBRS's IPOAs an event, it is the most anticipated AI hardware capital market event of 2026, setting the valuation benchmark for AI infrastructure in the secondary market, and its performance will influence pricing across all related assets.

As a long-term holding, it represents a typical 'high-risk, high-reward' bet, wagering on the macro narrative of 'inference supremacy,' the micro execution of 'Cerebras leveraging OpenAI to carve out a narrow monopoly,' and the valuation assumption that 'the market will continue to pay a 95x price-to-sales premium for AI hardware.' All three conditions must hold simultaneously for outsized returns; if any one fails, the drawdown will be severe.

For institutional investors, the strategy is typically to avoid chasing on the first day, waiting instead for the Q3 earnings report, updates on key customers, and valuation adjustments. For individual investors, considering it as a small tail asset within an AI hardware allocation is reasonable; treating it as an all-in faith-based investment requires revisiting the triple paradox outlined above.

More noteworthy than whether CBRS will surge at tomorrow's opening is another layer of significance to this matter:When a company that generates 86% of its revenue from two related entities in the UAE and is still operating at a loss can be valued by the market at 48.8 billion US dollars, it tells everyone just how far the capital frenzy surrounding the AI infrastructure sector has gone.

82K Views