Cerebras lists $24.6 billion of work in its backlog while reporting $510 million of revenue for 2025, a gap that highlights how the company’s public-market case depends on a handful of large, multiyear contracts. The AI chipmaker filed its S-1 registration on 17 April 2026, seeking a $22 to $25 billion valuation and roughly $2 billion in proceeds. The filing discloses large contracts with OpenAI and other customers and a concentrated revenue base that shapes both the growth case and investor risks.

Filing and financial snapshot

Cerebras formally filed to go public on Nasdaq under the ticker CBRS on 17 April 2026. The company provided its full S-1 registration in the public filing. The document lists $510 million in revenue for 2025.

The filing details the company’s contract backlog as $24.6 billion in remaining performance obligations as of 31 December 2025. Cerebras said it expects to recognise about 15 percent of that total in 2026 and 2027.

Different headline profit figures have circulated. The filing and press coverage show multiple measures that tell different stories. One report cited the filing as showing $87.9 million in net income for 2025. Another report parsed non-GAAP and GAAP measures, saying non-GAAP net income was $237.8 million while GAAP results reflected a $75.7 million net loss. The company’s S-1 provides the detailed numbers investors will review in the roadshow.

Large contracts and customer concentration

The S-1 lays out an expanded role selling computing capacity as a cloud service.

Cerebras has shifted from selling chips alone to operating systems built on those chips inside its own data centres for clients.

OpenAI is a central part of that strategy. The filing shows an arrangement to supply up to 750 megawatts of computing power to OpenAI between 2026 and 2028. That contract is valued at more than $20 billion, according to the filing. The deal calls for 250 megawatts per year in 2026, 2027 and 2028, and gives OpenAI the option to buy an additional 1.25 gigawatts through 2030.

Cerebras also disclosed that it issued OpenAI warrants to buy up to 33.4 million shares of non-voting Class N stock and took a $1 billion loan from OpenAI at a 6 percent annual interest rate to finance data centre build-out and services. The company can repay that loan in cash or by delivering products or services under the agreement. This warrants only vest in full if OpenAI purchases 2 gigawatts of capacity from Cerebras.

Those arrangements matter because the filing says the OpenAI alliance represents a substantial portion of projected revenue in coming years. At the same time, OpenAI can end parts of the agreement if Cerebras fails to deliver computing power on time or if service falls below agreed thresholds.

Who pays the bills now

The filing traces a rapid shift in which customers pay what share of revenue.

Mastheads

When Cerebras first prepared to go public in 2024, one customer, G42 of the United Arab Emirates, supplied the bulk of revenue in the first half of that year. That customer once accounted for as much as 87 percent of first-half 2024 revenue, the earlier disclosure shows.

By 2025 the mix had changed. The filing shows G42 accounted for 24 percent of revenue in 2025. Mohamed bin Zayed University of Artificial Intelligence, a public university in the UAE, provided 62 percent of 2025 revenue. Those figures underline a concentrated base of large buyers.

Concentration exposes investors to the fortunes of a few large contracts. At the same time, long-term agreements can lock in cash flows if the company executes. The S-1 presents both realities in numeric form.

Commercial moves and partners

Cerebras has also sought commercial partnerships with cloud providers and data-centre operators. In March the company signed a deal with Amazon to enable cloud services running on Cerebras chips and to allow Amazon to buy about $270 million in Cerebras Class N stock.

Oracle has discussed using Cerebras hardware in its cloud offerings, according to comments by Oracle’s CEO on an earnings call and the filing. Cerebras lists Amazon, Microsoft, Alphabet, Oracle and CoreWeave among firms that operate in the same market for AI compute.

Cerebras markets a wafer-scale chip called the Wafer-Scale Engine. The company argues the architecture gives an advantage for inference workloads because the chip packs large on-chip memory and many cores into a single die.

Industry coverage has highlighted technical claims such as a WSE-3 chip with around four trillion transistors, 900,000 cores and 44 gigabytes of on-chip memory. Some comparisons in the market state that the wafer-scale approach can run certain inference models many times faster than standard GPUs, according to public commentary and product material cited in filings and recent coverage.

The broader market is shifting greater share of compute to inference. Analysts and industry reports cited in market coverage note inference accounted for a rising share of AI compute needs in 2025 and 2026, which supports suppliers focused on inference performance.

Banking names lined up as underwriters, according to the S-1 and market reports. Morgan Stanley, Citigroup, Barclays and UBS are listed as joint lead underwriters. Mizuho and TD Cowen are bookrunners. Needham, Craig-Hallum, Wedbush Securities, Rosenblatt and Academy Securities are co-managers.

One market analyst, Stacy Rasgon, semiconductor analyst at Bernstein Research, noted that investor appetite for AI infrastructure is strong but unit economics will attract scrutiny. "Cerebras is entering the public markets at a moment when investor appetite for AI infrastructure is at an all-time high, but so is scrutiny of unit economics," Rasgon said.

The S-1 highlights several execution risks. Delivering large volumes of data-centre capacity on schedule is one. OpenAI’s right to end parts of its agreement if service falls short is another. The company also depends on a small set of customers for a large share of revenue today.

Financial metrics vary by accounting treatment. The contrast between non-GAAP profitability cited in market write-ups and GAAP loss figures in filings will be a focus.

Investors will parse which measures show the underlying business economics most clearly.

Related Articles

The S-1 shows $24.6 billion in remaining performance obligations, with about 15 percent expected to be recognised in 2026 and 2027.

This article was created with AI assistance.