2026 is set to see quite a few blockbuster initial public offerings (IPOs). Both OpenAI and Anthropic are expected to make their public debuts this year, but another notable artificial intelligence (AI) company may beat them to the punch.
Cerebras filed its S-1 registration statement with the Securities and Exchange Commission (SEC) earlier this month, suggesting its IPO could be imminent. It also gave investors a glimpse of its recent financial results, and there's a lot to like about the chipmaker, which aims to rival Nvidia (NVDA +3.97%) for AI compute dominance. That said, there's significant risk and uncertainty in the business that investors can't ignore, especially given its recent valuation and potential IPO price.
Here's what investors should know before the Cerebras IPO.
Image source: Getty Images.
Unlocking faster artificial intelligence
Cerebras makes AI chips unlike any others in the market. Its wafer-scale engine architecture uses an entire silicon wafer, resulting in chips nearly 30 times the size of Nvidia's Blackwell B200 package, stuffing 19 times as many transistors per chip, according to the company's S-1 filing.
The advantage of wafer-scale chips is that they can move data quickly with lower power requirements and fewer networking devices. And when you have clusters of thousands of graphics processing units (GPUs) all sharing data with each other, the time and costs can add up. Cerbras' chips eliminate much of the overhead of data transfer between chips by placing more compute cores and memory on a single piece of silicon, connecting them directly to one another during chip fabrication rather than relying on external networking or packaging.
The savings can be substantial, but the time saved may be even more important in the age of reasoning models. With reasoning models that iteratively call a large language model to refine answers, the time it takes for a response can grow significantly. Cerebras says its wafer-scale chips are up to 15 times faster than the leading GPU-based solutions when it comes to inference.
The technology is compelling. So much so, in fact, that OpenAI signed a $20 billion deal with the chipmaker to purchase 750 megawatts worth of AI inference capacity between 2026 and 2028, with the potential to add an additional 1.25 gigawatts of capacity by the end of 2030. Additionally, Amazon signed a deal to bring its CS-3 system to Amazon Web Services (AWS) for inference integrated with its custom Trainium3 chips.
With those two deals, Cerebras' balance sheet shows $24.6 billion in remaining performance obligations, which is absolutely massive for a company that generated just $510 million in revenue last year. But remaining performance obligations are not revenue, and there are a lot of risks to consider before investing.
The big challenges and risks facing the company
There are a handful of challenges Cerebras will have to overcome before it can fully realize its backlog of orders.
Cerebras will have to dramatically scale its data center infrastructure operations to service OpenAI. It's never taken on anything of this scale before, and it presents a significant operational challenge.
What's more, it'll need to scale production, and it's entirely reliant on Taiwan Semiconductor Manufacturing (TSM +0.62%) for its wafer-scale chips. No other foundry is capable of producing them. Meanwhile, TSMC is focused on retrofitting the 5nm process it uses for Cerbras' chips to its 3nm process, which is used by its biggest customers, including Nvidia, for high-end AI accelerators. Combined with the fact that yields on wafer-scale chips can be significantly lower than smaller chips (because a single mistake can ruin the entire wafer), and scaling its production could prove costly and challenging.
If Cerebras fails to deliver capacity on time or its hardware underperforms, it could result in a significant reduction in OpenAI's commitment. And that points to another key risk: Practically all of its backlog is tied to the OpenAI deal. The only way to mitigate that risk is to prove to potential customers that it can successfully scale its production and data center operations.
Granted, those kinds of risks aren't uncommon for a young company making its public debut. However, they don't appear to be priced into the valuation. The company is reportedly targeting a $35 billion valuation, although it refrained from putting an official price in the S-1 filing. That puts it at 70 times trailing sales, which is a huge premium to pay for the business. While it's set for significant revenue growth as it delivers on the OpenAI contract, it's still a premium valuation. For reference, Nvidia trades for 23 times trailing sales and 13 times 2026 revenue expectations. And while it's not growing as quickly, its dominant position in data centers is fairly secure.
So, while I like the technology Cerebras has developed, the price and execution risk will keep me from buying into the IPO. That said, it'll be worth keeping an eye on the company to see how well it executes and whether the stock's risk/return profile becomes more appealing.
