Cerebras Stock Got Halved. The CEO Says Everyone Missed The Point.

Let's say you just went public at $311. Day one, you rip 68% above your reference price. Your chip obliterates NVIDIA's flagship on every metric that matters for AI inference. You've got OpenAI and AWS on multi-year deals worth north of $20 billion combined. You're growing revenue 94% year-over-year. And then — your first post-IPO earnings report drops, and you get absolutely obliterated. Down 22% in a single session. Stock sitting at $167, barely half its IPO close.

That's exactly what just happened to Cerebras Systems (CBRS). And according to CEO Andrew Feldman, the market panicked at something it fundamentally misunderstood.

Here's what everyone missed, starting with the hardware. The WSE-3 — Wafer Scale Engine 3 — isn't a chip trying to compete with NVIDIA's GPUs. It's an entirely different architecture. The entire die is a full silicon wafer: 46,225 square millimeters, 4 trillion transistors — 19 times more than NVIDIA's B200 — crammed onto one single unit. 900,000 AI-optimized cores. 44 gigabytes of on-chip SRAM. 125 petaflops of compute, which is roughly 28 times what NVIDIA's current flagship delivers. No GPU cluster, no networking nightmare, no interconnect tax. You plug it in and it runs models up to 24 trillion parameters. Their Condor Galaxy 3 supercomputer in Dallas — 64 CS-3 systems stacked together — puts out 8 exaflops. That's not incremental. That's generational.

Now look at who's writing checks. OpenAI signed a multi-year AI infrastructure deal. AWS is shipping Cerebras-powered inference on its cloud. Meta is using it for Llama API inference workloads. G42 and Core42 co-built the Dallas facility with them. IBM, Mayo Clinic, GSK, Cognition — the customer list reads like a who's-who of enterprises not willing to bet on a single-vendor NVIDIA future. This isn't a speculative startup. This is institutional commitment.

So what spooked the market? Q1 2026 revenue came in at $193 million, up 94% year-over-year. Revenue isn't the problem. The problem is gross margins. Cerebras flagged that margins would compress going forward. Traders read that as deteriorating unit economics and hit the bid.

But here's the thing. CEO Feldman went on the record after the selloff saying the margin outlook was misunderstood. Cerebras is transitioning from selling hardware — one-time, high-margin box sales — to cloud and infrastructure-as-a-service. Recurring revenue. Multi-year customer lock-in. Familiar playbook? It should be. Every great enterprise software company went through this exact moment. Margins compress in the near term because you're building out cloud infrastructure and acquiring customers on longer contracts. But lifetime value explodes. You trade quarterly margin peaks for decade-defining customer relationships. That's what Cerebras is doing right now.

Think about the numbers underneath all this. FY2025 revenue hit $510 million, up 76% from the year before. They've got $2.23 billion in cash. They're spending $243 million on R&D — nearly half of revenue — which tells you exactly where they expect to be in three years. Yes, operating income is still negative at -$145 million for FY2025. But that's the cost of building a second source for the entire AI compute stack.

And here's where I'd pay attention: Cathie Wood's ARK bought $22 million worth of shares on the selloff — literally yesterday. The 10 analysts covering the stock have a consensus price target of $299. That's not a modest upside call. That's ~80% from current levels. Every single one of them rates it Buy or Strong Buy. Zero holds. Zero sells.

Look, Cerebras isn't going to kill NVIDIA overnight. Jensen's got a decade-long ecosystem advantage that doesn't evaporate because one quarterly margin guide disappointed day traders. But the AI infrastructure buildout is a multi-hundred-billion-dollar opportunity, and the hyperscalers are not going to let one vendor own the entire stack. Cerebras is the credible second source. The tech is generational. The partnerships are real. The business model transition is exactly what you'd expect from a company maturing from box sales to cloud recurring revenue.

The market priced in panic. The fundamentals say otherwise. 80% upside to consensus with the largest AI chip ever built underneath it — that's not a value trap. That's a gift wrap on the best semiconductor story of 2026.

Disclosure: This article is for informational and educational purposes only and does not constitute financial, investment, or trading advice. The information presented reflects publicly available data and analysis as of the publication date and is subject to change without notice. Always conduct your own due diligence and consult with a qualified financial advisor before making any investment decisions. Past performance is not indicative of future results. Investing involves risk, including the possible loss of principal.