Cerebras CEO insists dinner-plate-sized chip startup will still go public
Inference service launched a month before IPO filing turns out to have been a much bigger business than initially thought
Just days after announcing a $1.1 billion Series G funding round, AI chip startup Cerebras Systems pulled its S-1 IPO filing without so much as an explanation.
Amid growing concerns of an AI bubble, has Cerebras CEO Andrew Feldman gotten cold feet? Apparently not. Taking to LinkedIn on Sunday, Feldman apologized for the lack of transparency, and presumably any panic the decision might have caused.
"On Friday, Cerebras Systems withdrew our S-1. We didn't explain why - that was a mistake," he wrote, adding that he still has every intention of taking the company public.
Founded in 2015, Cerebras develops dinner-plate-sized AI accelerators. These accelerators were initially pitched as an alternative to GPUs for AI model training. But while Cerebras has by no means abandoned training as a workload, that business has taken a back seat to inference — a service it launched barely a month before announcing its intention to go public in September last year.
As Feldman points out, a lot has changed since then. Inference turns out to have proved a particularly potent workload for Cerebras' accelerators, each of which is equipped with more than 40 GB of SRAM, a class of memory that's etched directly into the compute die and makes the HBM used by rivals Nvidia and AMD look glacial by comparison.
This has allowed the company to offer API access to models at speeds topping 3,000 tokens a second for models like OpenAI's gpt-oss-120B, far faster than is possible using GPUs.
These capabilities have helped Cerebras win over several high profile customers, including Mistral AI, AWS, Meta, IBM, Cognition, AlphaSense, Notion, and Perplexity, a major change from this time last year.
In the startup's IPO filings from September 2024, the company reported that the United Arab Emirates' (UAE) AI crown jewel G42 accounted for 87 percent of its revenues in the first half of 2024 — a fact that raised no shortage of concerns among members of the US government.
To support its growing customer base, this spring Cerebras announced it was expanding operations to four new datacenters across the US, in addition to one in Montreal and another in France. With its bank accounts flush with another $1.1 billion funding, the company now plans to further expand its US manufacturing and datacenter capacity.
- OpenNvidia could become the AI generation's WinTel
- Stargate is nowhere near big enough to make OpenAI's tie-ups with AMD and Nvidia work
- Deloitte refunds Aussie gov after AI fabrications slip into $440K welfare report
- Oracle will have to borrow at least $25B a year to fund AI fantasy, says analyst
However, as Feldman points out, none of this growth is reflected in its S-1 filings.
"Given that the business has improved in meaningful ways we decided to withdraw so that we can re-file with updated financials, strategy information including our approach to the rapidly changing AI landscape," Feldman wrote, emphasizing that the move doesn't reflect a shift in strategy.
"We will re-file our S-1 when the updated materials are ready," he wrote. ®