AI pricing has been falling for two years straight
That logic may be about to break. The release of Anthropic’s Mythos model — initially gated to a select tier of enterprise customers — has sent a visible signal through the market. But the more consequential story isn’t technical. It’s structural.
Here’s the economic mechanism worth tracking: when demand for frontier compute outpaces supply, access becomes rationed. That rationing doesn’t happen randomly. It follows revenue concentration, strategic partnerships, and negotiated enterprise agreements. The companies that locked in preferred vendor status early are now operating under a fundamentally different cost structure than those still on standard API pricing.
Two dynamics are converging. First, the AI infrastructure buildout — despite record CapEx from Microsoft, Google, and Amazon, collectively on pace for $200B+ in 2025 — is running behind model demand growth. Second, consolidation at the frontier is narrowing real choice. OpenAI, Anthropic, and Google DeepMind account for an estimated 70%+ of enterprise frontier model spend. That’s a supplier concentration ratio that would concern any CFO in a traditional vendor review.
The deflationary era of AI pricing rested on one assumption: unlimited competitive substitution. Scarcity economics removes that floor.
For any finance or tech leader with AI embedded in the product stack or cost structure, the relevant framework right now is vendor dependency scoring — mapping which workflows are model-agnostic versus locked to a single provider, and what a 30–50% pricing increase would do to unit economics.
That analysis is worth running before pricing power consolidates further.
How are you pressure-testing AI vendor concentration in your own planning?
Source: https://www.ft.com/content/53f9bb30-3abc-4f4d-bf0d-99410d0ab77f
