Cerebras Burned $200M Building an 'Impossible' Chip
SkimNews Take
The market's enthusiasm for AI, rather than Cerebras's specific technology, appears to have been the primary driver of its IPO success, masking its earlier near-failure and potentially setting a precedent for other high-burn AI startups.
Get the Finance newsletter
Daily finance — markets, central banks, M&A, the prints that move money. Free.
- Cerebras Systems went public this week at a roughly $60 billion valuation, making both co-founders billionaires and marking 2026's biggest tech IPO so far
- OpenAI and AWS are now customers of Cerebras's AI chips, which the company sells specifically for inference workloads
- Andrew Feldman, Cerebras's founder and CEO, disclosed the company was burning $8 million a month in 2019 and had incinerated nearly $200 million chasing a single technical problem
- The wafer-scale concept — building an entire silicon wafer as one giant processor — had never been successfully manufactured before, for AI or any other use case, according to Feldman
- Feldman was forced to report repeated failures to the board every few weeks during the crisis, but had no alternative: without a solution, Cerebras was dead regardless
- Cerebras's bet directly contradicted 50-plus years of conventional microprocessor design, which had focused on shrinking transistors onto ever-smaller silicon dies
Why it matters: Cerebras's $60B valuation validates a contrarian bet against 50-plus years of conventional chipmaking: building one giant wafer-scale processor rather than stringing many smaller chips together. With OpenAI and AWS as inference customers, the bet has paid off in the market, though Cerebras currently sells only inference chips — not the training workloads where NVIDIA still dominates.
