Chinese open-source AI is starting to win over U.S. businesses

Ramp’s AI Index shows a shift in enterprise purchasing: the share of businesses paying for model-serving platforms rose to 6.1% of AI-spending businesses in July from 4.5% in January 2026, indicating more adoption of open/Chinese alternatives. While U.S. leaders remain dominant (Anthropic up 1.1pp to 43.5% market share; OpenAI up 0.23pp to 39.7%), premium models appear capped—Anthropic’s Fable 5 is only 6% of tokens and 11.4% of spend despite ~ $10 per million tokens (2x OpenAI’s GPT-5.6 Sol), as buyers increasingly choose cheaper “good enough” options.

Analysis

Open-weight adoption is a pricing-war story before it is a share-shift story. The first-order damage is to premium model monetization per workflow; the second-order winners are platforms that can capture usage, hosting, governance, or workflow integration rather than raw model ASP. That makes GOOGL relatively more insulated than a pure frontier-model narrative would imply, while BABA gains strategic optionality from the Chinese open-model stack and could benefit if enterprise fine-tuning demand shifts toward cheaper, locally controlled deployments.

TRI looks like a cleaner beneficiary than the market may appreciate: legal/document workflows are exactly where domain specialization and lower inference costs should expand margins and reduce dependence on third-party model vendors. RAMP’s own business is not directly threatened, but its AI-spend data may increasingly signal a mix shift away from high-ARPU frontier usage and toward commoditized serving layers, which can pressure AI multiples even if aggregate spend keeps rising.

The contrarian miss is assuming Chinese open-model leadership immediately becomes broad U.S. enterprise substitution. Procurement, security, and data-residency friction should slow that over the next 1-3 quarters, so the near-term trade is about expectation resets and margin compression, not a sudden collapse in demand. Falsifier: if the next two Ramp prints and upcoming earnings calls show frontier-model spend holding steady while enterprise model-serving spend reaccelerates, the pricing-ceiling thesis is premature.

Key Decisions for Investors

• Long TRI on weakness for a 3-6 month horizon: specialized legal AI should improve unit economics and retention as customers migrate from premium frontier APIs to in-house tuned models; cut if product quality or renewal commentary deteriorates.
• Add GOOGL on AI-multiple pullbacks, not on headline hype: open-weight commoditization can still lift cloud inference volumes and favor distribution-rich incumbents; thesis breaks if Cloud growth or Vertex AI consumption decelerates for two consecutive quarters.
• Small tactical long BABA as optionality on the Chinese open-model ecosystem, but keep sizing modest until there is evidence of sustained non-China enterprise adoption; reduce if U.S. procurement/security restrictions tighten.
• Watch RAMP’s next AI Index release as a signal, not a trade trigger: if model-serving share keeps rising while frontier spend share stalls, expect multiple compression across AI software rather than a direct fundamental hit.