Cerebras Targets CS-5 Production for 2027 as Backlog Hits $26.1 Billion

Cerebras Systems ended the third quarter of 2026 with a revenue backlog of $26.1 billion and $7.36 billion in cash, even as hardware sales fell 28.8 percent to $68.9 million while the company built out cloud capacity for its OpenAI contract.

The decline was expected. Cerebras spent 2026 ramping manufacturing and standing up its cloud services business in support of the $10 billion inference deal signed with OpenAI in January 2026. That agreement covers 750 megawatts of gear that OpenAI can rent for fast inference, and figures of $20 billion to $30 billion between 2026 and 2032 have circulated since. The larger stream depends on OpenAI exercising an option for an additional 1.25 gigawatts of capacity by the end of 2030, with services running through 2032. Provided OpenAI goes public and has the cash, there is no reason to believe it will not stretch the deal to the maximum.

What dragged margins down to 11 percent?

Gross margins came in at just 11 percent of revenue in Q3, far below the company's long-term goal of 60 percent. Two forces drove the shortfall. The Cerebras factory is still ramping to a new production level, and an early customer — presumably cloud provider G42 — negotiated aggressive pricing years ago for machines being installed now.

Chief executive Andrew Feldman also used the Wall Street call to reject talk of a yield problem at the company. The argument has merit. The WSE-3 in the CS-3 systems and the WSE-3 Turbo in the CS-4 systems are built on very mature 5 nanometer processes at Taiwan Semiconductor Manufacturing Co, not bleeding-edge nodes such as the 2 nanometer process AMD is using for its just-launched "Venice" Epyc 9006 server CPUs.

How far along is the WSE-5?

The WSE-3 Turbo is a stop-gap, double-clocked waferscale processor that Cerebras shipped in CS-4 systems in August. There will be no WSE-4, but a WSE-5 is in the works for next summer's CS-5 machines.

"Our product road map continues to advance rapidly," Feldman told analysts. "Our CS-4 will soon be in full production, doubling both performance and throughput. Our next-generation processor, the Wafer Scale Engine 5, which will be deployed in our fifth-generation CS-5 system, is back from the fab ahead of schedule."

Feldman gave a precise timeline. The WSE-5 taped out on June 15, has been brought up, is already running models and is undergoing in-chassis testing. "Moving from tape-out to system test in less than five months is exceptional execution," he said. "The CS-5 will more than double the speed of CS-4. It will enter production in early summer of 2027. It is built on TSMC's 5 nanometer node, where competition for wafers is less intense and wafer costs are much lower."

Who are the new hyperscaler customers?

Beyond OpenAI, Cerebras has now signed Amazon Web Services, Oracle Cloud and Microsoft Azure. Those deals are not yet in the backlog and are not yet producing revenue, and Cerebras has been vague about their terms. Even so, three of the world's largest clouds will work CS iron into their inference workloads, presumably for decode operations. The CS-5, with twice the performance of the current WSE-3 Turbo, is almost certainly the system they want.

Cerebras is also pairing its wafers with AMD GPUs for prefill in its own cloud, which by definition has to be cheaper for inference than buying Nvidia GPUs for prefill and Nvidia Grok LPUs for decode. The CS-4 and CS-5 will compete with new GPUs from Nvidia and AMD, LPUs and other XPUs from Grok, SambaNova and Rebellions, plus homegrown gear from hyperscalers, cloud builders and AI model builders.

What does the financial picture look like?

Hardware sales at supercomputer makers are lumpy, and the CS family is comprised of supercomputers, so annual figures tell the cleaner story. On a GAAP basis, Cerebras should close 2026 with roughly $885 million in revenue, up 73.4 percent from 2025. Operating income and net income were hit by a spike in research and development spending and stock-based compensation.

Feldman said the company expects to triple its sales in 2027. As manufacturing ramps and margins improve on hardware built with partners Flex and Sanmina, gross profit and operating income should strengthen materially next year.

One open technical question surrounds the WSE-5's design. Cerebras revealed at Hot Chips this past summer that it will use 3D stacked DRAM with its wafers, and possibilities such as silicon photonics links between wafers and SRAM stacking remain on the table. Given the premium on fast inference, double-clocked wafers also seem likely from here onward — dollars per million tokens matters more than tokens per watt at this stage of the GenAI market. The answers should arrive with the CS-5 launch next summer.