Anthropic soars to $965bn valuation, leapfrogging OpenAI

This is less a celebration of one company than a signal that the private AI capital stack is still in a reflexive phase: the highest-quality model provider is now being priced like a platform winner before public-market discipline can test unit economics. That tends to benefit the picks-and-shovels layer first, because every incremental dollar of frontier-model spend translates into more inference, more training, and more enterprise deployment intensity — all of which are upstream demand multipliers for GPU, networking, power, and data-center capacity.

The second-order effect is competitive pressure on the broader software stack. If Claude is gaining share in coding and enterprise workflows, the near-term risk is not just to other model labs but to application vendors whose differentiation depends on generic AI features; model quality compression can shift margins from software to infrastructure. Over the next 3-6 months, the market may continue to reward “AI exposure” indiscriminately, but over 12-24 months the winners should be the firms that control scarce supply, not the ones that merely consume it.

The contrarian issue is that headline valuations can mask a coming bottleneck in monetization. Private-market pricing may be extrapolating adoption curves while underweighting the possibility that compute costs, model commoditization, and procurement concentration at large enterprises force a later-stage margin reset. If public investors start demanding proof of durable ROIC rather than usage growth, the air-pocket risk is highest in richly valued AI software proxies, while infrastructure leaders should remain relatively insulated.

For NVDA and GOOGL specifically, the article is directionally supportive but not equally so: the bigger near-term beneficiary is Nvidia via sustained training/inference spend, while Alphabet faces a more nuanced read-through because faster frontier competition raises both capex intensity and product pressure across search/ad tools. The cleanest market expression is to own the toll collectors and hedge the application layer until the IPO window validates actual cash generation rather than narrative momentum.