Cerebras listed on NASDAQ on May 14 at $185 per share and opened at $311. The $66B market cap puts the wafer-scale engine company on par with established EDA vendors. The chip itself (58x the die area of a GPU, 15x the inference throughput on standard benchmarks) was never the question. The question was whether the market would pay for non-NVIDIA AI silicon, and the answer is now on the tape.
The mechanism that made the IPO work was not chip sales. Cerebras pivoted from hardware vendor to vertically integrated inference provider: they built and operate their own data centers, serve customers via API, and let the chip be a competitive advantage that customers never have to buy directly. That is the architecture that cleared the market. Teams that evaluated Cerebras in 2022 for hardware procurement could not get their heads around the procurement model. The cloud API route removed that friction entirely.
Custom silicon for AI inference is now a proven, public-market-validated architecture choice, not a research project. The teams building next-generation inference chips have a $66B comparable to cite when making architecture decisions. The GPU monoculture in AI inference has a public challenger with a balance sheet. That changes the risk calculus for every chip team that has been sitting out the non-GPU path waiting for market proof.