Cerebras는 2027년 출시를 목표로 CS-5 시스템을 준비 중이며, 추가 하이퍼스케일러 고객을 확보했다고 The Next Platform이 보도했다.

Upstart waferscale AI compute maker Cerebras Systems has been very upfront that 2026 is a year to build out its cloud and ramp up its manufacturing capability so it can meet the demands of contracts it has signed with OpenAI and, more recently, Amazon Web Services, Oracle Cloud, and Microsoft Azure. So far, most of its revenue backlog and all of its incremental revenue is coming from the OpenAI deal, which is finally ramping.

Next year is going to be a banner year for many AI system providers, and there is a non-zero chance that it could be the peak spending year for the GenAI boom. We shall see, though even AI systems do a terrible job of predicting the future. All anyone knows is what their customers tell them, and all those organizations know is what their own customers tell them, and so on down and up the chain. Sometimes this bullshit begets even bigger bullshit, but sometimes it isn't bullshit and people really do buy a ton of systems. That has happened for the past four years, despite fears that the gravy train might run out of coal.

Here is the thing, though: GenAI is real. It is just a matter of time, driven by the ever-decreasing cost of chewing and spitting out tokens, before it becomes pervasive in the world's applications, or around them. Some go too far and think it will become the world's applications, but they do not understand the nature of risk avoidance among the millions of corporations around the world. There is mainframe code older than I am at the center of our banking and insurance systems. Wave after wave of technological evolution since I entered the IT sector four decades ago has augmented that code, but a very large portion of it is still there, like a granite mountain.

In the third quarter, Cerebras did not sell as much of its CS-3 and CS-4 hardware as it has in prior quarters, which stands to reason because the company was building out its cloud services business to support the $10 billion inference deal it signed with AI model builder OpenAI in January 2026.

Since then, that deal, covering 750 megawatts of gear that OpenAI can rent for fast inference, has been characterized as worth $20 billion, and sometimes $30 billion, between 2026 and 2032, inclusive. That larger revenue stream is apparently possible if OpenAI has Cerebras build an additional 1.25 gigawatts of capacity by the end of 2030 and the services contract runs out to 2032. Provided OpenAI goes public and has the cash, there is no reason to believe it will not stretch this deal to the max. This option, by definition, has to be cheaper than buying Nvidia GPUs for inference prefill and Nvidia Grok LPUs for decode, particularly with Cerebras using AMD GPUs for prefill, as it does in its cloud.

In the third quarter, hardware sales at Cerebras fell 28.8 percent to $68.9 million. Because the Cerebras factory is still ramping to a new production level, gross margins were poor at only 11 percent of revenues, versus the long-term goal of 60 percent. Andrew Feldman, co-founder and chief executive officer, told Wall Street analysts on a conference call that margins in Q3 2026 were also hurt by the aggressive pricing an early customer, presumably cloud provider G42, negotiated years ago for machines being installed now.

There has been some jawboning online that Cerebras has a yield problem, and Feldman put that idea to rest. It stands to reason. The WSE-3 used in the CS-3 systems and the WSE-3 Turbo used in the CS-4 systems are based on mature 5 nanometer processes from Taiwan Semiconductor Manufacturing Co, nothing as exotic as the 2 nanometer process AMD is spearheading with its "Venice" Epyc 9006 server CPUs, which just launched this week. The WSE-3 Turbo is a stopgap, double-clocked waferscale processor that Cerebras put on the market in August in the CS-4 systems. A WSE-5 is in the works for next summer for the CS-5 machines. (There is no WSE-4, because there isn't going to be one.) These CS-4 and CS-5 systems will compete with new GPUs from Nvidia and AMD, LPUs and other XPUs from Grok (essentially owned by Nvidia), SambaNova, and Rebellions, and with homegrown gear from the hyperscalers, cloud builders, and AI model builders themselves.

"Our product road map continues to advance rapidly," Feldman said on the call. "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. The WSE-5 taped out on June 15. The chip is back, has been brought up, and is already running models and undergoing in-chassis testing. Moving from tape-out to system test in less than five months is exceptional execution. 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."

It is safe to say the WSE-5 has been in development for a long time, and I strongly suspect that as technology has advanced, it has been tweaked to incorporate new things. There are all kinds of technological possibilities for this engine, including silicon photonics optical connections linking wafers and memory servers, and SRAM stacking. I favor those two, for the sake of Cerebras. At Hot Chips this past summer, Cerebras revealed it would use 3D stacked DRAM with its wafers.

Given the need for fast inference, I also think the wafers will be double-clocked from this point forward. Performance trumps power consumption and heat dissipation these days; dollars per million tokens matter more than dollars per million tokens per watt at this stage of GenAI commercialization. Electricity is cheaper than the systems, even if it is getting harder and harder to come by. Cerebras can always offer two clock speeds at the same price and let customers decide.

We shall see next summer.

Given the lumpiness of hardware sales at all supercomputer makers, and make no mistake, the CS family is made up of supercomputers, it is best to look at the company's financials on an annual rather than quarterly basis. Here is how 2022 through 2025 look, with projections, for GAAP revenue (mine), not the non-GAAP revenue, which is larger, that Cerebras cites in its financial presentations for Q3 and Q4.

For those who like the raw numbers, here are the quarterly and annual results in the typical table.

My best guess is that on a GAAP basis, Cerebras will close 2026 with around $885 million in revenue, up 73.4 percent from 2025. Operating income and net income were whacked by a spike in research and development spending and stock-based compensation. Feldman said on the Wall Street call that the company expects to triple its sales in 2027. It is reasonable to assume that as manufacturing ramps and Cerebras improves its hardware margins with partners Flex and Sanmina, gross profit and operating income will be much improved in 2027.

Cerebras ended the third quarter with $7.36 billion in cash and equivalents and a revenue backlog of $26.1 billion, mostly from the OpenAI deal. The AWS, Oracle, and Microsoft deals are not yet in the backlog and are certainly not yet bringing in revenue. Cerebras has been fairly vague about what these deals entail, but it is important to note that three of the biggest clouds in the world are going to be working CS iron into their inference workloads, presumably for decode operations. Whatever the WSE-5 and resulting CS-5 system turn out to be, this is almost certainly what they want, especially since the WSE-5 will offer twice the performance of the WSE-3 Turbo compute engine available now.