Cerebras Systems Inc. (CBRS) made one of the most explosive Nasdaq debuts the market has seen in years. It opened at $350 against its $185 initial public offering (IPO) price and briefly approached a $100 billion valuation, though the fully diluted market cap settled closer to the mid‑$90‑billion range. The excitement is understandable. I’m confident in the artificial intelligence (AI) technology that Cerebras has brought to market to compete with chip giants like Nvidia Corporation (NVDA).
Meet Samuel – Your Personal Investing Prophet
Explore CWVX for 2X leverage on CRWVThe AI infrastructure company that designs and manufactures wafer-scale chips and full AI supercomputer systems for training and inference workloads has built a genuinely compelling hardware story. Early commercial partnerships suggest demand for alternative AI infrastructure platforms is real. At the same time, I think the market’s debut enthusiasm has priced this stock far ahead of what the business can justify on a fundamental basis. As such, I am neutral on CBRS, and I’d argue that the case for patience here is stronger than the case for chasing.
A Chip Built Differently
Cerebras makes several of the world’s fastest AI chips, and one of its flagships is the WSE-3. This chip is fabricated on TSMC’s (TSM) 5nm process and measures 46,225 square millimeters, making it 58 times larger than Nvidia’s B200. It packs 4 trillion transistors, 900,000 AI-optimized cores, and 44 GB of on-chip SRAM with 21 petabytes per second of memory bandwidth.
With those specs, the company claims a 15x increase in inference speed over open-source models. That figure has attracted customers ranging from Argonne National Laboratory to GlaxoSmithKline (GSK).
The speed of its offerings is a key part of the rapid growth in Cerebras’ business model, which has four layers: on-premises AI supercomputers, cloud computing capacity through Cerebras Cloud and partner clouds, software subscriptions, and professional services.
Revenue reached $510 million in 2025, up 76% from $290 million in 2024, with hardware contributing $358 million and cloud services $152 million. The cloud mix is crucial to Cerebras’ earnings power because it offers structurally higher margins and greater revenue predictability than hardware alone, and its growing share is a strong signal in the company’s S-1.
The Revenue Concentration Problem Has Been Relocated
This is where I think the IPO narrative deserves the most scrutiny. The 2024 filing was withdrawn after the Committee on Foreign Investment in the United States (CFIUS) raised concerns about Cerebras’s dependence on Group 42 Holding Ltd. (G42), the Abu Dhabi AI company that accounted for 85% of revenue in the first half of that year. The 2026 re-filing presents a different picture because G42’s contribution had fallen to 24% by the end of 2025.
However, the Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) contributed 62%, meaning 86% of total revenue still came from G42, along with affiliates that the prospectus itself identifies as related parties. The diversification the market is celebrating was largely a reallocation between two connected accounts, not the addition of independent Western customers.
Of course, we can use the OpenAI relationship as the main counterargument. In January 2026, Cerebras signed a multi-year agreement for 750 megawatts of inference capacity, expandable to two gigawatts by 2030, with a contracted value over $10 billion at signing and a binding master relationship agreement worth over $20 billion at full expansion. In March 2026, Cerebras also signed a term sheet with Amazon’s (AMZN) Amazon Web Services (AWS), under which AWS will become the first hyperscaler to deploy Cerebras in its own data centers. These are meaningful commercial wins.
Yet that’s just one way of looking at it. In my view, shifting reliance from G42 to OpenAI is replacing one dependency with another. Given the scale of the OpenAI deal and its $1 billion “loan” to Cerebras for data center buildouts, we can reasonably estimate that the bulk of the latter’s revenue over the next two years will come from OpenAI alone. So, there’s a major risk that if their relationship has any hiccups, we will see a direct and proportionate effect on the revenue base at a time when the stock is already priced for flawless execution.
What the Financials Show
Cerebras reported net income of $238 million in Fiscal Year 2025, which is a 47% net margin. That is unusual at this stage of a hardware company’s growth, and it is only part of the story. On a GAAP basis, the company actually lost $146 million once you discount the non-operating accounting items that pushed the net income into positive territory. So, it is still far from being a profitable business.
On the balance sheet, the company is doing well, with nearly $3 billion in cash and equivalents pre-IPO as of January 2026 and no debt. Now that the IPO is over and Cerebras has banked over $5 billion in proceeds, its liquidity runway is almost as long as it needs to scale operations and revenue until it becomes profitable.
In my view, the second structural risk that has received less attention is the TSMC dependency. Cerebras manufactures all of its wafers through TSMC, but it doesn’t have any formalized long-term supply or allocation commitment from the foundry. Given TSMC’s well-documented capacity constraints at advanced nodes, that is a real and underappreciated vulnerability in the operating model.
Valuation
At the May 21 closing price of $281.86, CBRS traded at around 108x trailing 2025 revenue (EV/sales), compared with roughly 25x for Nvidia and 18x for Advanced Micro Devices (AMD). Meanwhile, the semiconductor sector’s median EV/Sales multiple sits closer to 4x, so Cerebras is around 27x the sector median on a trailing revenue basis.
For a company still generating a GAAP operating loss, that gap demands a very specific set of assumptions about the revenue trajectory and margin expansion to justify it.
I’d look at this way: an incredibly bullish scenario has Cerebras reaching $3–$4 billion in annual revenue by 2028, which would compress the multiple to a more reasonable 16x–21x. Even half that number would still bring the multiple down to about 40x. What we’re hoping for in this case is that the OpenAI agreement and contracted G42 commitments will sustain themselves through that horizon.
Also, it bears mentioning that CoreWeave (CRWV), the nearest public comparable as a specialized AI infrastructure provider, went public in March 2025 at a $23 billion valuation and is not yet profitable at the same scale. Cerebras is already trading at a significant premium to that benchmark on every relevant metric.
Wall Street’s View
We don’t yet have formal analyst coverage, which is standard for a company that priced its IPO barely a week ago, but we have early market commentary that’s generally bullish on the company, even if the stock is rather expensive now.
Also, the institutional demand during the roadshow was extraordinary, with the offering more than 20 times oversubscribed, but oversubscription at IPO and fundamental justification at any given price are different questions entirely.
The Bottom Line
I’ll be watching the first wave of underwriter initiations closely, but the company, its business model, and offerings are the main drivers of that value. For now, there’s a big gap between reported and operational profitability and a valuation that prices in outcomes several years away at 108x EV/Sales TTM. So, I think CBRS is a Hold at current levels and a more interesting conversation closer to $185 to $220.