OpenAI and Anthropic’s Valuation Risk: When Frontier Intelligence Stops Being Scarce

OpenAI and Anthropic’s latest private rounds imply a combined ~$1.8 trillion valuation built on frontier intelligence staying scarce and premium-priced—yet Moonshot’s Kimi K3 landing near top systems on Artificial Analysis weakens that scarcity story. Training the absolute frontier still climbs (~3.5× per year in Epoch AI estimates), but inference cost at a fixed capability has been collapsing on the order of ~40× per year, so last year’s best model becomes this year’s cheap—and often open-weight—commodity. Open weights sever who trained a model from who gets paid each time it runs, shifting rents toward chips, energy, data centers, distribution, and physical machines, as Nvidia’s stack and Grok’s cost/performance posture illustrate. Labs still hold distribution modes—ChatGPT’s huge weekly audience, Claude’s enterprise workflows—but those must replace pure model scarcity. Overall, this points to durable value in energy, silicon, and embedded physical AI rather than whichever lab trains the smartest chatbot this month.

Check the video here.