Cerebras’s First Earnings Test Isn’t Profit — It’s Who’s Buying

When Cerebras Systems posts its first quarterly report as a public company after the closing bell on June 23, the number most headlines will lead with — a per-share loss — is close to meaningless. Cerebras was never going to show a profit this quarter, and Wall Street is not asking it to. The figures that decide where CBRS trades into the summer sit lower on the page: how fast a $24.6 billion backlog is turning into revenue, and whether the company has found a single major customer outside Abu Dhabi.

This is the first hard data the market has had since Cerebras priced one of the loudest AI listings of the year. The stock opened at nearly double its IPO price, briefly carried a valuation north of $80 billion, and has since handed back roughly a third of that. June 23 is the moment the wafer-scale chipmaker stops being a story about momentum and becomes a story about a financial model.

Article Brief

Key Takeaways

5 Points30s Read

1. 01The date–Cerebras reports fiscal Q1 2026 after the close on June 23 — its first results since the May IPO.
2. 02Ignore the loss–Consensus is a small per-share loss. The reported 2025 “profit” was a one-time accounting gain, not operations.
3. 03The real number–A $24.6B backlog, mostly OpenAI, is the bull case — the tell is how fast it converts to revenue.
4. 04The real risk–About 86% of 2025 revenue came from two UAE-affiliated customers; G42’s decline was replaced by MBZUAI.
5. 05The overhang–An unusual lock-up clause can release insider shares early once market cap clears $40B — it already has.

A 68% pop, then a long walk back

Cerebras priced its IPO at $185 a share on May 13 — above its marketed range — and raised about $5.55 billion, the largest US technology offering of the year. The debut was the kind that defines a tape: the stock opened near $350, touched $385 intraday, and closed at $311, up roughly 68%. Then gravity took over.

By the June 18 close, CBRS sat near $235 — still well above the IPO price, but down about a third from that first-day high, inside a 52-week band that already spans $197 to $386. That round trip is the backdrop for this print. The buyers who chased the pop are underwater; the buyers eyeing an entry are waiting for a reason. A first earnings report is exactly the catalyst that breaks that standoff one way or the other. For the running tally of price, valuation and targets, Cerebras’s live quote page carries the current data.

Why the headline loss is the wrong number

Cerebras will almost certainly report a loss for the quarter — consensus sits near a 14-cent per-share deficit — and that is the right thing to expect from a company still scaling into its backlog. The trap is the prior year. The reported 2025 “profitability” was an artifact of accounting, not a sign of operating leverage.

Cerebras booked GAAP net income of roughly $238 million for 2025. Almost all of it came from a single non-cash gain of about $363 million tied to the change in fair value of a forward-contract liability, recognized entirely in the second quarter. Strip that out and the picture inverts: Cerebras lost money on a GAAP basis in the other three quarters, and its non-GAAP net loss for the year was about $75.7 million — wider than 2024. Anyone calling Cerebras profitable is reading one line and skipping the footnote.

The line that does deserve attention is revenue, and it is moving the right way. Cerebras grew from $99.5 million in the first quarter of 2025 to $171.4 million in the fourth — a 72% climb across the year that accelerated into the finish, with Q4 up 26% over Q3. Full-year revenue landed at $510 million, up 76%. Gross margin is the offsetting worry: it eased from about 42% in 2024 to 39% in 2025, the wrong direction for a company asking the market to pay up for scale. The Q1 guide, and any comment on where margin goes next, will reveal whether the ramp is healthy or merely busy.

The $24.6 billion question

The single most important figure in Cerebras’s filing is its backlog. As of December 31, the company reported $24.6 billion in remaining performance obligations — contracted business not yet recognized as revenue. Against $510 million of actual 2025 revenue, that is nearly 48 times promised-to-delivered, and it anchors every bull case on the stock.

The catch is what sits inside it. The overwhelming majority traces to one counterparty: OpenAI. The agreement, first announced in January and formalized late in 2025, commits OpenAI to deploy 750 megawatts of Cerebras inference capacity through 2028, in a deal Cerebras values at more than $20 billion. A backlog anchored to one well-funded AI leader is both the bull case and the risk: it is real, contracted demand, but its conversion runs on a single customer’s buildout schedule. Secondary estimates put only about 15% of the backlog converting to revenue by the end of 2027, with the bulk arriving in 2028 and later — so the figure that matters on June 23 is not the backlog itself, but the pace at which it becomes recognized revenue.

One piece of genuine diversification is worth tracking. Cerebras has paired with Amazon Web Services to offer its inference inside the AWS ecosystem — a relationship that, if it scales, would put Cerebras silicon in front of mainstream cloud customers rather than a handful of sovereign buyers. A hard number attached to AWS on the call would matter more than another OpenAI headline.

The concentration nobody actually fixed

Here is the problem the bull case tends to skip. Cerebras spent 2024 as effectively a single-customer company: G42, the Abu Dhabi AI group, was 85% of revenue. The IPO narrative said that dependence was easing — and on G42 specifically, it did, falling to 24% in 2025. But the gap did not fill with a diversified book of business. It filled with a second Abu Dhabi entity.

In 2025, Mohamed bin Zayed University of Artificial Intelligence — MBZUAI — became Cerebras’s largest customer at 62% of revenue. Add G42’s 24% and roughly 86% of the company’s revenue still came from two UAE-affiliated buyers. The concentration did not dissolve; it changed names. For a US-listed chipmaker whose customers’ AI ambitions track a single government’s strategy, that is a single-country dependency as much as a single-customer one — and it carries the geopolitical sensitivity that delayed the listing in the first place. Cerebras’s earlier filing stalled for the better part of a year while a US national-security review examined G42’s stake. Investors who want this story to mature need to see a recognizable American or European logo on the customer list. June 23 is the first chance to learn whether one exists.

Where Cerebras genuinely leads

None of that risk changes what Cerebras actually built, which is the most distinctive piece of AI hardware in production. Its Wafer-Scale Engine — the WSE-3 — is a single chip the size of a dinner plate, etched with 4 trillion transistors and 900,000 cores on one continuous slab of silicon, where a conventional design would be sliced into dozens of separate GPUs. Keeping an entire model on one wafer removes the networking bottleneck that throttles clusters of discrete chips.

That architecture has found its moment in inference — running already-trained models — just as the industry’s spending tilts from training toward serving. Cerebras markets eye-watering speed, claiming its systems generate output for large open models like Llama at well over 2,000 tokens per second, multiples faster than GPU-based services. Those are company benchmarks and deserve the usual skepticism, but the direction holds: as AI shifts from building models to running them at scale, raw inference throughput becomes the product, and it is the field where Cerebras has the cleanest technical edge. It is the same supply-and-speed logic now reshaping the rest of the stack, from scarce high-bandwidth memory to power.

The competition is closing from three directions

Cerebras’s edge is real, but it is not alone, and the threats arrive at different speeds:

• Direct speed rivals. Groq, with its inference-tuned LPU, is the closest head-to-head on tokens-per-second, and SambaNova competes on the same full-system, sovereign-AI pitch — though SambaNova has stumbled through layoffs and a down round, leaving Groq the sharper near-term competitor.
• The incumbent. Nvidia is the benchmark everyone is measured against; its Blackwell generation dominates training and is the volume option in inference. Cerebras wins on niche speed, not breadth.
• The structural threat. The hyperscalers are building their own silicon — Google’s TPUs, Amazon’s Trainium, Microsoft’s Maia, Meta’s MTIA. In-house chips are the long-term squeeze on every merchant AI-silicon vendor, Cerebras included, because the largest buyers would rather own the supply than rent it.

The competitive question is whether wafer-scale stays a defensible advantage or becomes a feature others approximate. Earnings will not settle it, but management’s commentary on design wins and capacity will hint at how durable the lead is.

The balance sheet buys time; the customer list spends it

Cerebras does not have a funding problem. It ended 2025 with roughly $1.3 billion in cash and securities, and the IPO added about $5.5 billion more — a war chest near $7 billion against a company burning well under $100 million a year on a non-GAAP basis. Runway is not the bear case. The pressure point is the other side of the OpenAI agreement: serving 750 megawatts of inference means building or leasing enormous capacity, and the filing flags take-or-pay style commitments to do it. The question the cash raises is not whether Cerebras survives, but how much it must spend up front to earn a backlog that converts mostly after 2027.

That spending is why valuation is the real argument. Near $235, CBRS trades around 90 times trailing sales — a multiple that makes sense only if revenue compounds toward the multibillion-dollar level analysts pencil in for 2027 and beyond. Nvidia, the most profitable name in AI silicon, changes hands at a fraction of that sales multiple while generating enormous cash; Cerebras is priced as if its backlog were already revenue. A single quarter cannot validate a 2027 model, but it can show whether the slope is steep enough to grow into the price — or whether the market has paid a premium for a backlog that belongs to one customer and a build-out that has to be funded before it pays.

The lock-up clause that could cap the upside

One technical detail deserves more attention than it gets. Cerebras’s IPO lock-up — the window barring insiders from selling — carries an unusual early-release trigger. Beyond the standard expiry tied to the third-quarter report, shares can be freed once the company’s market capitalization clears $40 billion on a sustained basis, or once certain OpenAI payments arrive. At roughly $235, Cerebras is already worth more than $50 billion, which means that threshold has plausibly been met. A strong reaction to earnings could, paradoxically, accelerate insider selling rather than reward holders — a supply overhang worth pricing in before chasing any post-print rally.

Analysts have been constructive in their first calls: Morgan Stanley opened coverage at $250 and Wedbush at $270, both beneath an average target near $294 that still implies room above the June 18 close. Every one of those targets, though, rests on the same two assumptions — that the backlog converts and the customer base widens.

What to watch when Cerebras reports

Skip the EPS line. These are the disclosures that will actually move the stock:

• The revenue guide. Anything that extends the Q4 acceleration past a ~$171 million run-rate confirms the ramp; a flat or soft guide does the opposite.
• A new name. Any customer outside G42, MBZUAI and OpenAI — especially a US or European enterprise — would be the most bullish line in the release.
• Gross margin. After sliding to 39%, direction matters more than level. Scale should help; if it does not, the premium is harder to defend.
• Backlog conversion. Listen for how much of the $24.6 billion management expects to recognize in 2026 and 2027, not just 2028.
• Lock-up and cash. Commentary on the early-release mechanics and on capital commitments for the OpenAI buildout will shape the supply-and-spending picture.

Cerebras enters its first earnings as a genuinely impressive engineering company wrapped around an uncomfortable customer list, priced at about 90 times sales while still operationally unprofitable. The IPO answered whether investors wanted the story. June 23 begins to answer whether the business underneath it can broaden fast enough to justify the price. The backlog says demand is real; the revenue line will say how real, and how soon; the customer list will say how safe. For the longer view on valuation and access, an earlier breakdown of how to think about Cerebras still frames the trade.