Cerebras Signals Strong IPO Demand Above Price Range
AI chip maker Cerebras told investors of strong demand during its roadshow and plans to set the IPO price above the initially marketed range.
Cerebras IPO at $185: Bet on Inference, Not Training, and SoftBank's Hidden Play
The fact that on May 13, 2026, Cerebras Systems set its IPO price at $185 per share, rather than the previously announced range of $150-160, is not just a sign of AI hype. It signals a fundamental shift in semiconductor market preferences. Investors, including SoftBank, are no longer chasing training compute—they are making a huge bet on the computations that make already-built AI work. In this race, Cerebras's giant chips look like a weapon against the Nvidia empire.
The Core: What's Really Happening
Behind the buzz of 20x oversubscription and a final valuation of $56.4 billion lies a much deeper tectonic shift. The market has recognized that the era of expensive LLM (Large Language Models) training is giving way to the era of inference. Previously, the main goal was to build a model; now the task is to make it run fast, cheap, and in real time, serving millions of requests from "agentic AI." This is where Nvidia's architecture (GPUs originally designed for gaming) starts to struggle, and Cerebras's wafer-scale engine (WSE-3), which is 58 times larger than Nvidia's Blackwell chip, opens up incredible opportunities.
The insider intrigue lies in the failed acquisition. SoftBank and Arm Holdings actively explored buying Cerebras right before the IPO. Masayoshi Son, known for his instinct for infrastructure monopolies, wanted to grab this asset before it went public, obviously realizing that acquiring the company now would be cheaper ($40-50 billion) than trying to catch up later. However, Andrew Feldman, CEO and founder, chose to go public himself. This is a classic move: when you have a technology that can disrupt the market, you don't sell to giants; you demand a premium from the public market that embeds the potential for multiple growth against a fading leader.
Timeline and Context
What we are seeing now is Cerebras's second attempt to go public. The first, attempted back in 2024, failed. The reason was mundane but fatal: CFIUS (Committee on Foreign Investment in the United States) launched a review because the lion's share of the startup's revenue came from G42 in the UAE. Dependence on a single Middle Eastern client threatened sanctions.
The lesson was learned. By May 2026, the picture is fundamentally different. Revenue soared from $290 million in 2024 to $510 million in 2025, and instead of G42, two giants appeared on the scene: OpenAI and Amazon Web Services. Their multi-year contracts became the fuel that drove IPO demand.
Furthermore, the January mega-deal with OpenAI to deploy 750 megawatts of computing power worth over $20 billion was a signal to the entire market. The paradox is that OpenAI itself, as a client and holder of warrants worth $14.4 million (through co-founder Greg Brockman), is betting on Cerebras, even while simultaneously developing chips with Broadcom. This proves that even the creators of ChatGPT do not believe Nvidia can alone satisfy the appetites of future artificial superintelligence (ASI).
Who Wins and Who Loses
SoftBank wins, but as a brilliant tactician. They tried to buy the company directly, got rejected, and likely entered the IPO through their channels (UBS and Barclays, which are underwriters) to build a position in the secondary market in the first days of trading. For them, it's a win-win: if they couldn't get the whole company, they get a share of its growth.
Data centers win. The $1.1 billion deal between Digi Power X and Cerebras shows that demand for energy for new chips will only grow. This pulls along the entire power supply and cooling sector.
Nvidia loses—in the long-term psychological perspective. Many analysts rightly point to Nvidia's moat: the CUDA software ecosystem that developers are used to. But Cerebras offers what Nvidia cannot: a chip the size of a silicon wafer that is 15 times faster for inference tasks and significantly cheaper to maintain. As one expert aptly noted, this is a test of the market's desire to buy "anything with an AI label besides Nvidia stock."
What the Media Isn't Saying
First, Elon Musk as a hidden beneficiary. Court filings revealed that OpenAI considered merging with Cerebras back in 2017 to create AGI. Now Musk, as an enemy of OpenAI, may watch as his competitors accidentally enrich his potential future partner. Given that Musk is building his xAI, Cerebras's high public valuation sets a benchmark for xAI's future rounds. If the market gives 100 times annual revenue for inference hardware, what is xAI itself worth?
Second, the illusion of diversification. We are told that dependence on a single client is resolved. But now 90% of revenue is tied to two: OpenAI and AWS. This is better than before, but it does not make the business stable. If OpenAI solves its funding problems for its unprofitable business and cuts CapEx, Cerebras will collapse. Moreover, OpenAI employees' stake in Cerebras through warrants creates a dangerous conflict of interest: when making chip purchasing decisions, they affect the value of their own portfolio.
Third, financial magic in the reports. Yes, the company posted a net profit of $88 million for 2025, but a close look at the documents shows: GAAP reporting records operating losses. The profit came from non-cash items and financial maneuvers. This means the company spends more on development and production than it earns from sales. At this scale and valuation, there is simply no room for error.
Forecast: Next 30 Days and 90 Days
30 days (by mid-June): Shares will show wild volatility. Huge oversubscription and the fact that institutions received insufficient allocation guarantee a sharp spike in the first days. But then reality will set in. Traders will start shorting the stock as a "bubble," unless Cerebras announces a new major partnership. Likely price range in a month: $150 to $220 per share.
90 days (by mid-August): Here, macroeconomics and competition come into play. The key trigger will be the company's first report as a public entity. If it confirms revenue growth rates and, crucially, margins, the stock will settle above $200 and ignore market corrections. But there is a huge geopolitical risk: WSE-3 production depends entirely on TSMC (the sole supplier). Any escalation in the Taiwan Strait will hit Cerebras harder than anyone else, simply because their wafers are unique and cannot be made on backup capacity.
Insider takeaway: This IPO is not just a fundraising; it's an attempt to lock in a huge market segment from giants like Nvidia. Once inference costs exceed model training costs, Nvidia's multiples will compress to mature company levels, and the growth premium will flow into Cerebras shares. That's why SoftBank wanted to buy them before the market fully realizes this.
— Editorial Team
