The Nasdaq debut of Cerebras Systems is a prime example of the type of stock market launch that has significance beyond the financial aspects of the offering itself. The Sunnyvale-based AI chip company eventually went public this month for $185 per share, substantially beyond the upper end of its already-raised price range. The company has spent years positioning itself as the most reliable rival to Nvidia in the data center computation industry.
On the first day of trade, the stock opened at $350. At $311.07, it closed. Cerebras’s first market value of almost $65 billion, which is significant for any IPO and noteworthy for a business that, only three years earlier, was largely regarded as a long-shot gamble against the dominating chip incumbent, was achieved with a first-day gain of 68%. The deal became the biggest U.S. tech IPO since 2021 with a total of $5.55 billion raised, which is a significant market signal about where institutional capital has been choosing to invest.
Cerebras’s success is essentially an argument about how the AI chip market will develop, so it is important to grasp the company’s basic story. The CEO and founder of the firm, Andrew Feldman, designed Cerebras around a chip architecture that is fundamentally different from the one that has made Nvidia the leading provider of AI training hardware. Cerebras creates monolithic wafer-scale chips that are around 58 times larger than ordinary GPUs, whereas Nvidia and its rivals make standard-sized GPUs that are then networked together in data centers to handle massive training workloads.
The computing power that would typically require dozens of GPUs coupled by intricate networking infrastructure is effectively contained on a single chip. This architecture is based on the wager that the wafer-scale method yields significantly faster training durations and lower power consumption than the multi-GPU option for specific types of AI applications. The IPO has now given public market investors the task of determining whether that wager will pay off in the larger market.
By all standards, the institutional demand for the service was exceptional. Institutional investors together requested to buy 20 times the number of shares that were really available at the offering price since the book was oversubscribed by more than 20 times. In a market where the majority of IPO performance has been disappointing, that kind of oversubscription is the most obvious indication that astute investors think Cerebras has struck gold. The financial figure that explains the institutional interest is the $24.6 billion backlog the company disclosed in its papers, which is anchored by an estimated $20 billion three-year deal with OpenAI. Such a backlog in a recently publicly traded firm indicates that the demand for Cerebras’ chips is not hypothetical or speculative. Some of the most important clients in the AI ecosystem have made a contractual commitment to it.
However, the cynicism that has been subtly growing around AI stocks finds its strongest expression in the valuation article. By all historical standards, Cerebras is trading at an excessive multiple of almost 25 times its projected 2026 sales. Mature semiconductor companies, on the other hand, usually trade between five and ten times their revenue. Even Nvidia has traded at multiples much below 25 times revenue for the majority of its growth period, despite its remarkable recent results. In other words, the Cerebras value is pricing in a significant portion of future success that has not yet materialized. For the post-IPO valuation to appear reasonable in retrospect, the company must increase revenue significantly over the following three to four years while maintaining gross margins that support the existing multiple. That is doable. It’s not guaranteed.
The aspect of the Cerebras narrative with the biggest strategic ramifications is the OpenAI partnership. Outside of the typical Nvidia partnerships that rule the larger industry, the rumored $20 billion, three-year agreement with OpenAI is among the biggest AI infrastructure commitments made by a single customer to a single supplier. No early-stage AI chip business has ever had revenue visibility as Cerebras does because to the OpenAI promise. The partnership offers OpenAI a significant diversification away from the Nvidia-dominated supply chain, which has been the company’s sole viable option for training infrastructure for a number of years. By all standards, the partnership is the best proof that Cerebras has established itself as a reliable alternative compute provider rather than only a curiosity at the research stage.
The hyperscalers and big cloud providers continue to support the larger AI infrastructure investment case with real, consistent capital expenditures. Through 2026, Microsoft, Google, Amazon, and Meta have all maintained or raised their intentions to invest in AI infrastructure, with a total of hundreds of billions of dollars in capital expenditures. The multibillion-dollar OpenAI-backed AI infrastructure buildout known as the Stargate project, which was unveiled earlier this year, has created an additional layer of demand that is only now starting to materialize into orders for particular chips and data centers. The demand for AI compute hardware will sustain robust revenue growth throughout the market’s infrastructure layer as long as that capital flow persists. Businesses who secure a significant portion of that demand—possibly including Cerebras—will gain a great deal.
But compared to a year ago, the bear situation has been receiving more attention. According to a number of eminent studies, the efficiency of AI models is increasing more quickly than the headline computing demand forecasts had predicted. The advances made by Chinese DeepSeek earlier this year showed that competitive AI capacity could be attained with a lot less computing than what American hyperscalers had been using.
According to Stanford research on energy-efficient AI systems, as efficiency gains accumulate, the overall compute demand may eventually stabilize or even decrease. The rapid demand increase that AI infrastructure stocks have been pricing in might not fully materialize if those trends persist. In this way, institutional investors’ vote of confidence that demand will persist is represented by the Cerebras IPO. Although it has fewer well-known proponents, the opposite wager—that demand would moderate as efficiency increases—has a strong underlying case.
The practical challenge for investors considering whether to purchase Cerebras at current prices is whether the company’s unique competitive position warrants the premium. Cerebras argues that there is sufficient demand for several winners in the larger market for AI infrastructure, that the wafer-scale design gives real benefits for specific workloads, and that the OpenAI partnership delivers revenue visibility that rivals do not. The argument against it is that Nvidia’s dominant position is actually hard to challenge, that once established client connections in this market are typically persistent, and that value multiples of 25 times revenue allow very little leeway for any performance errors over the next years.