Ben Okonkwo: I've been staring at this one number for two days and I still find it — hm — almost perversely funny. Cerebras Systems beats every metric anyone publicly modeled. Revenue, loss, the whole thing. And the market's response is to send CBRS down eleven percent after hours. I need someone to explain that without using the word 'sentiment.'
Eleanor Crane: Without the word sentiment. Alright.
Ben Okonkwo: Hard mode.
Eleanor Crane: Here's the plain version. Investors in a growth stock are not buying what the company did — they're buying what it's going to do. So when Q1 comes in at a hundred and ninety-three point four million, ninety-two percent year-over-year, crushing the FactSet consensus — that's already priced in the moment the IPO opens at three-fifty. What wasn't priced in was the Q2 margin guide.
Ben Okonkwo: The earnings beat versus forward guidance divergence. Yeah. But — okay, the IPO was May 14th, a hundred and eighty-five dollar price. Opens at three-fifty. That's nearly double on day one. So the market was already way ahead of the fundamentals. Doesn't that mean the selloff is just... correction?
Eleanor Crane: That's exactly the question. And the answer lives in what changed between three-fifty and eleven percent down — which is the margin guidance, not the revenue number.
Eleanor Crane: Core gross margin. That's the number doing the damage. And I want to be precise about what it is, because this isn't the GAAP figure — it strips out the pass-through data-center costs, strips out the warrant amortization, so you're left with what Cerebras actually earns on its hardware and cloud services. Q1 2026: forty-six point five percent. Q2 guidance: thirty-six to thirty-eight. That's a ten-point drop in a single quarter.
Ben Okonkwo: Ten points in one quarter. While revenue is still growing.
Eleanor Crane: And the year-over-year picture actually looks fine — Q1 2025 core gross margin was roughly forty-two percent, so the business improved. It's the sequential move that spooked people. Something shifted in the mix between quarters. And the obvious candidate, announced the same day — June 23rd — is the twenty billion dollar multi-year deal with OpenAI.
Ben Okonkwo: Okay, I need to sit with that for a second. Andrew Feldman signs what is almost certainly the largest deal in AI infrastructure history, and it immediately becomes the reason nobody wants to hold the stock? Is the OpenAI deal a moat or is it — I mean, is Cerebras now dependent on a single customer for, what, the majority of forward revenue?
Eleanor Crane: That's the question I can't fully resolve. Large anchor customers extract pricing concessions — that's not controversial. But whether that makes OpenAI a moat or a liability depends on whether Cerebras can walk away from the table. And I'm not sure the Wafer Scale Engine gives them that leverage yet.
Ben Okonkwo: Nvidia doesn't face this. Not at that scale. So is this a Cerebras problem or is it what every non-Nvidia AI hardware company discovers the moment a hyperscaler actually shows up at the door?
Ben Okonkwo: Wait, but — okay, I want to press on something that doesn't add up for me. CBRS priced at a hundred and eighty-five. Opened at three-fifty, touched three-eighty-five. Then fell twenty-eight percent from peak before June 23rd even happened. But after all of that — after the after-hours drop too — it's still nearly fifty percent above the IPO price. If the earnings story is genuinely broken, wouldn't it have fallen past a hundred and eighty-five?
Eleanor Crane: That's a real observation. And I don't want to dismiss it.
Ben Okonkwo: It suggests the market still believes the Wafer Scale Engine thesis — that this architecture is genuinely differentiated against Nvidia. The floor held somewhere. So maybe what we watched isn't a broken story. Maybe it's just... euphoria being repriced.
Eleanor Crane: The three-fifty open was euphoria, yes. But someone at Cerebras knew when they filed that Q2 margins would land in the thirty-six to thirty-eight range. Andrew Feldman knew. The board knew. The underwriters priced at a hundred and eighty-five with that information in the room. Retail opens it at three-fifty without it. That gap isn't just a market pattern. That's an information asymmetry.
Ben Okonkwo: Hm. Though — actually, there's research on this. Analysts systematically over-react to optimistic sentiment in earnings call narratives and under-react to the risk signals. So the IPO surge and the post-guidance selloff could both just be the same behavioral bias running in opposite directions. It doesn't require anyone to have hidden anything.
Eleanor Crane: And that's worth holding. But does that behavioral pattern explain why the margin drop feels structural when it might actually be a mix-shift artifact — the Q2 guide reflecting a customer-concentration moment rather than a ceiling on what the Wafer Scale Engine can earn?
Eleanor Crane: Which is the question I can't get out of my head. Because Cerebras's full-year margin guidance implies a potential recovery — back toward the mid-forties by Q4 2026. So either Q2 at thirty-six to thirty-eight is a floor, meaning OpenAI extracted a one-time concession and the business finds its footing, or it's a landing zone — meaning this is what the Wafer Scale Engine actually earns when a hyperscaler is in the room. And those are not the same company.
Ben Okonkwo: And we can't tell from the outside. We don't have the contract terms.
Eleanor Crane: No. And it wasn't just Cerebras on June 23rd and 24th — the Nasdaq, the S&P, South Korean markets, the whole AI trade cooled at once. So some of that eleven percent is macro. Which makes it even harder to isolate what the market was actually saying about Cerebras specifically.
Ben Okonkwo: Right, and that's — I mean, that's the thing nobody can answer yet. If Q2 at thirty-six to thirty-eight is where every non-Nvidia AI hardware company ends up the moment a real hyperscaler signs, what does that mean for every AI hardware IPO that follows Cerebras through that door?
