Each leading-edge training generation costs roughly an order of magnitude more than the last. Per-token inference cost falls by roughly an order of magnitude per year. For OpenAI and Anthropic this means recurring training cycles and pre-listing dilution.
Groq's LPU architecture is purpose-built for inference. Success is becoming a structural participant in the inference market, not displacing GPU incumbents wholesale.
SpaceX's launch-cost compression curve is the cleanest precedent for what AI silicon cost-down does to the addressable market.
OpenAI Pre-IPO Anthropic Pre-IPO Groq Pre-IPO SpaceX Pre-IPO IPO Index Investor Platform AI Infrastructure Landscape Pre-IPO Operating Manual