Cerebras Goes Public: What Investors Need to Know

Cerebras Systems has spent years building something genuinely different in the semiconductor world — a chip the size of an entire wafer, purpose-built for AI workloads at a scale that makes conventional GPUs look incremental. Now, after a failed IPO attempt in 2024, the company is trying again. It filed to list on Nasdaq under the ticker CBRS, and this time, the numbers look considerably more convincing.

Whether that's enough to satisfy public markets is a different question entirely.

Understanding Cerebras' Business Model

Most chip companies design silicon. Cerebras redesigned the unit of silicon itself.

Its Wafer Scale Engine isn't a chip in the traditional sense — it's an entire silicon wafer functioning as a single processor, containing hundreds of billions of transistors and eliminating the inter-chip communication bottlenecks that slow down conventional AI training and inference. That architectural decision has real consequences: Cerebras systems can run large language model inference significantly faster than GPU clusters for certain workloads, which is exactly what hyperscalers and sovereign AI projects need.

But Cerebras isn't just a hardware company — it's also a cloud service provider, selling access to its compute infrastructure rather than simply shipping boxes of silicon. That distinction matters enormously for how investors should think about the business. Hardware companies get hardware multiples. Infrastructure-as-a-service companies, particularly those with long-term contracted revenue, get something much better.

The cloud services angle also means Cerebras takes on capital expenditure risk — it has to build data center infrastructure before it can recognize revenue. That's a harder business to operate, but it creates stickier, recurring relationships with customers who'd rather buy compute than manage exotic hardware themselves.

The Financials: A Closer Look

The headline numbers are genuinely impressive. Cerebras reported $510 million in revenue for 2025, up 76 percent from the prior year, and swung to $87.9 million in net income — a dramatic reversal from a $485 million net loss in 2024. For a pre-IPO AI company, that kind of profitability is rare enough to warrant attention.

The backlog is even more striking. As of December 31, 2025, Cerebras carried $24.6 billion in remaining performance obligations — contracted future revenue that hasn't been recognized yet. The company expects to recognize roughly 15 percent of that in 2026 and 2027 combined. Spread that out, and you're looking at a multi-year revenue runway that most technology IPOs can only claim theoretically.

To put the $24.6 billion figure in context: that's nearly 48 times 2025 revenue. Even discounting for contract risk and execution uncertainty, that backlog gives Cerebras a financial cushion that's genuinely unusual for a company at this stage. The February 2026 funding round — $1 billion at a $23 billion valuation — suggests institutional investors see it similarly.

Where things get more complicated is concentration. More on that shortly.

Strategic Partnerships: OpenAI and Beyond

The OpenAI deal is the centerpiece of this IPO story, and it's worth understanding precisely what was agreed.

OpenAI will consume 750 megawatts of computing power delivered through 2028, under a $10 billion agreement announced in January. It also holds an option to purchase an additional 1.25 gigawatts through 2030 — a number that, if exercised, would dwarf the initial contract. For reference, 1.25 GW of compute infrastructure is roughly equivalent to a large campus of hyperscale data centers.

The structure of the deal goes further than a typical supply agreement. OpenAI extended a $1 billion loan to Cerebras at six percent annual interest specifically to fund data center build-out. That's a creditor-customer hybrid relationship that aligns incentives — OpenAI wants Cerebras to succeed because it needs the compute — but it also creates a financial obligation Cerebras must service regardless of market conditions.

The real insider read here is that OpenAI isn't buying chips — it's buying AI inference capacity at scale, which tells you something important about where the market is heading. The shift from "we'll buy hardware and operate it ourselves" to "we'll contract for compute outcomes" is accelerating, and Cerebras is positioning itself squarely in that transition.

The UAE partnerships are a separate thread. G42, the Abu Dhabi-based AI and cloud company, accounted for 24 percent of Cerebras' 2025 revenue — down dramatically from 87 percent in the first half of 2024, which was one of the red flags that derailed the original IPO filing. Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) represented 62 percent of 2025 revenue, which immediately raises the same concentration concerns that G42 once did. Notably, both G42 and MBZUAI partnered on a February 2026 deal to deploy Cerebras systems in India — suggesting these UAE-linked entities are functioning as a coordinated sovereign AI ecosystem rather than independent customers.

Potential Risks and Market Concerns

Revenue concentration is the obvious risk, and the filing doesn't try to hide it. Swapping 87 percent dependence on G42 for 62 percent dependence on MBZUAI isn't diversification — it's substitution. Both entities are UAE-affiliated, and both are now co-investing in the India expansion, which means Cerebras' revenue base is arguably more geographically and politically concentrated than the raw numbers suggest.

The OpenAI deal adds a different kind of concentration risk. Cerebras explicitly acknowledges that OpenAI "represents a substantial portion of our projected revenues over the next several years" and that OpenAI can terminate part or all of the agreement if compute delivery falls behind schedule. For a company building infrastructure from scratch, that's a meaningful execution risk. Missing a data center commissioning deadline isn't an abstract possibility — it's a real operational hazard in a market where power, permitting, and equipment all routinely cause delays.

There's also the competitive reality: Nvidia controls the GPU market with a grip that's proven resistant to every challenger who's tried to loosen it. Cerebras has found genuine product-market fit with specific AI inference workloads, but scaling that into a durable competitive moat against a company with Nvidia's ecosystem depth, software stack, and manufacturing partnerships is a different order of difficulty.

The prior IPO withdrawal in 2024 reportedly involved national security review concerns related to the G42 relationship. The filing doesn't address that directly, but investors should note that the UAE partnership structure has grown more complex, not simpler, since then.

What to Expect Post-IPO

Cerebras enters the public market at an interesting moment. AI infrastructure investment is accelerating — data center construction, power procurement, and compute procurement are all running at historically elevated levels. The question isn't whether Cerebras addresses a real market need; it's whether the company can execute at the scale its contracts demand without stumbling on the operational complexity that scale creates.

The $24.6 billion backlog, if Cerebras can execute against it, represents a revenue trajectory that would make the current $23 billion valuation look conservative by the time 2027 arrives. That's the bull case. The bear case is simpler: two or three large customers deciding not to renew, or a significant delivery failure with OpenAI, would reshape the financial picture dramatically.

For investors evaluating this as an AI infrastructure investment, the most useful lens isn't semiconductor comparables or cloud company multiples — it's project finance. Cerebras is essentially a build-and-operate business with contracted long-term cash flows and real execution risk. Underwriting it requires the same discipline you'd apply to any large infrastructure commitment: look past the headline growth rate, examine the counterparty quality, and think hard about what happens when something goes wrong.

The Cerebras IPO will tell us something about where market appetite for frontier AI infrastructure sits in 2026. Given the backlog, the OpenAI relationship, and the profitability turnaround, there's a real story here. Whether public market pricing rewards the opportunity or punishes the concentration risk will depend on which narrative investors find more compelling — and right now, both are defensible.

[INTERNAL LINK: Cerebras Systems Overview]

[INTERNAL LINK: AI Infrastructure Trends]

[INTERNAL LINK: Semiconductor Industry Insights]

Ready to explore the future of AI infrastructure? Visit InfraSale Marketplace to learn more!