Cerebras' Margin Scare Was a Misread — and the Street Is Repricing CBRS

Why the Margin Scare Was a Misread

CBRS didn't fall because the business is broken — it fell because the first earnings report as a public company landed an EPS miss alongside soft margin guidance. The quarter itself was strong: core revenue of $191.3 million, up about 92% year over year, with the net loss narrowing to $14 million from $23.9 million a year earlier. What spooked the tape was the full-year gross-margin guide of 38–41% against the 47% Cerebras printed in Q1, with Q2 guided to just 36–38%. Shares sold off roughly 17%.

CEO Andrew Feldman pushed back the next day, saying investors had misunderstood the guidance. His explanation: the lighter near-term margin comes from a temporary decision to rent its own systems back from a customer to bring capacity online faster — a deployment-speed choice, not structural pricing pressure. If he's right, the 38–41% guide is a floor distorted by a one-off, not a new baseline. The market spent the back half of the week deciding he's probably right.

The Street Reprices the Dip

The sell-side didn't blink. Morgan Stanley lifted its target to $273 from $250 and kept an Overweight, calling it a strong debut quarter with top-line and gross-margin upside — and noting the margin outlook still sits well above its own ~24% estimate. Wedbush went to $280 from $270 with an Outperform. Mizuho reiterated Outperform with a $300 target, and its note carried the detail that actually matters for the bull case: cloud revenue up about 60% quarter over quarter on the OpenAI ramp, with Cerebras pulling roughly 50% higher pricing on incremental fast-inference demand.

That pricing line is the tell. The bear thesis is margin compression; the data says Cerebras has pricing power on exactly the inference workloads that are scaling fastest. With three desks clustered at $273–$300 against a stock near $191, the gap between price and target is doing a lot of the work in this bounce.

GPT-5.6 Sol Puts a Date on the $20B Deal

Underneath the bounce is the fact that the $20 billion OpenAI deal — a multi-year, 750-megawatt compute commitment running in tranches through 2028 — is turning into a shippable product. OpenAI's new flagship, GPT-5.6 Sol, is set to go live on Cerebras silicon in July at up to 750 tokens per second. That converts an abstract backlog number into a dated, benchmarkable launch — the kind of thing a market can actually price.

There's a wrinkle: Sol's initial access is restricted to select partners through the API and Codex at the direction of the US government, a policy OpenAI publicly calls unsustainable. Demand isn't the constraint here; distribution is. For Cerebras the relevant fact is simpler — its hardware is the inference layer for OpenAI's fastest model, and OpenAI is enough of a counterparty that it also extended Cerebras a $1 billion working-capital loan now sitting on the balance sheet.

What Could Still Break It

The margin debate isn't settled by a CEO soundbite. Q2 core gross margin is still guided to 36–38%, and if the rent-back drag persists past a quarter or two, the bears get their structural-compression story back. Sol's government-gated rollout could slip, and a model that can't reach developers doesn't pull through inference volume on schedule. And this is a name that only IPO'd in May — the debut popped 68% to a roughly $95 billion cap — so it still trades with post-IPO volatility and an eventual lockup overhang. On the perp specifically, the ~$22.5 million of 24h volume on this market is thin relative to the underlying equity, so basis to spot can gap around weekend and after-hours moves.