Cerebras Revenue Beats Estimates as Backlog Passes $26 Billion

Cerebras beat its own targets on the top line and narrowed the gap toward profitability, and that combination is what pushed the stock up nearly 18% in late trading. Core adjusted revenue for the third quarter reached $230.9 million, up 70% from a year earlier and comfortably ahead of the Street's $214.7 million target. The headline numbers are strong, but they need some unpacking before they say much about the underlying business.

Reading the numbers through the accounting

On a GAAP-style basis that excludes certain costs such as stock compensation, Cerebras lost 67 cents per share. Analysts had guided to a 14-cent loss. The two figures aren't directly comparable, and the reason lies in how the company books revenue. Cerebras includes some pass-through revenue that gets subtracted from the total. It also issues warrants to its major customers, and those warrants are amortized as noncash reductions in revenue under standard accounting rules. The adjusted number leaves both items out, which is why it reads higher than a straightforward reading of the reported total would suggest.

That's worth keeping in mind when comparing Cerebras with peers. Warrant-linked customer incentives are a real cost, even when they're excluded from the headline figure. Anyone modeling the company's margins should decide early which version of revenue they're using and apply it consistently.

The margin trend is more encouraging than the absolute level. Cerebras still isn't profitable, but its adjusted operating margin came in at -14%, against guidance of -24% given three months earlier. Moving that much in one quarter tells you more about the direction of the business than any single loss figure does.

Backlog growth, and the concentration behind it

Total backlog across hardware sales and the company's own cloud services rose from $25 billion in August to $26 billion at the end of the quarter. Chief Executive Andrew Feldman said a significant portion of that figure ties to the agreement with OpenAI Group PBC. That's a double-edged detail. On one hand, a large anchor customer provides visibility. On the other, it means the backlog's growth rate depends heavily on a single relationship, and a $1 billion increase in backlog says less about broad demand than the headline suggests.

Feldman argued that the concentration has a silver lining: if most of the backlog sits with one customer, other buyers may not have to wait as long for their orders. That's a reasonable read, though it's still a management claim that will be tested as deliveries roll out.

Customer wins and the timing problem

Cerebras recently signed deals with Amazon Web Services and Oracle, and on the same day it announced Microsoft as a new customer. The Microsoft announcement came with few specifics. Feldman said the agreement was signed about a month before the report, and that the company expects revenue from it to begin appearing in the first quarter of fiscal 2027.

The bigger question is size. The backlog increase was modest, which doesn't do much to address concerns that these deals are on the small side. Several new customers that are still ramping rarely move the needle in a single quarter. What matters is whether those accounts grow into meaningful, repeatable spending.

Financing is the open question

Feldman said the company is now weighing financing options to meet customer demand and work down its backlog. He didn't say what form that financing would take. That's a notable gap. Capital for AI hardware and infrastructure has become harder to secure in recent months, as investors grow more cautious about the scale of spending on AI buildouts. A chipmaker with a growing order book and negative operating margins needs working capital, and how Cerebras funds that growth will shape dilution, margins and its flexibility to take on more customers.

Nvidia's turbulence creates an opening, but not a quick one

Some analysts estimate Nvidia accounts for more than 90% of AI chip sales, which makes even a modest shift in buyer behavior significant for challengers. Feldman pointed to long lead times and rising costs for GPUs as an opening for Cerebras, particularly in hardware sales. Nvidia has also gone through a leadership change, with longtime Chief Financial Officer Colette Kress named interim CEO after co-founder and CEO Jensen Huang resigned.

My read is that this won't translate into share gains quickly. Enterprise and cloud buyers usually take months to qualify new hardware, and software stacks, developer familiarity and supply commitments all favor incumbents. Cerebras can capitalize on any hesitation, but the near-term benefit is more likely to show up as deal flow than as a visible change in market share.

Guidance and the full-year math

For the fourth quarter, Cerebras expects a core gross margin of 47% to 49%, with revenue between $310 million and $330 million. Wall Street had been looking for a core gross margin of 42.1% on $271 million in sales. If those targets hold, the company would be well ahead of consensus on both lines.

Feldman also said he expects revenue to triple by the end of fiscal 2027. That would imply full-year revenue of roughly $2.85 billion, slightly below Wall Street's long-term forecast of $2.92 billion for the same year. In other words, the company's own trajectory lands just short of what analysts already expect, which is a useful reminder that the stock's reaction reflects expectations as much as results.

What to watch next

Three things will determine whether this quarter is a turning point. The first is the financing structure: its terms will say a lot about how expensive growth will be. The second is whether Microsoft revenue starts arriving in the timeline the company gave. The third is backlog conversion, meaning how quickly the $26 billion turns into recognized revenue without the margin pressure that comes from large customer incentives. Cerebras has shown it can grow fast. The next few quarters will show whether that growth can carry the company toward profitability.