AI Compute · Wafer-scale inference. The high-beta NVDA bet.
$CBRSCerebras
Cerebras builds wafer-scale chips that win on inference latency and throughput vs GPUs, with OpenAI, G42, and AWS among its named customers. The trade is concentrated and binary. Inference winning faster than expected is the bull case; customer concentration is the structural risk.
The capex package committed this morning takes NVDA capacity up materially through 2027. The signal: management sees the demand curve flatter for longer than the consensus model.