OpenRouter’s routing data — across thousands of U.S. organizations, January 2025 through June 2026 — shows Chinese-origin models went from a rounding error to nearly half of all token volume. This is commoditization in production, not in theory.
OpenRouter Signal — Mid-2026
~46%
Chinese-origin share of all routed tokens (platform-wide, mid-2026)
~4.5%
Chinese-model share among U.S. orgs — H1 2025 baseline
16.3%
DeepSeek’s share of ALL OpenRouter token volume — #1 provider
~36%
U.S.-origin models’ platform share — down from clear majority a year prior
OpenRouter — the AI model-routing marketplace that sits between U.S. developers and the full menu of available models — published data covering January 2025 through June 2026. The chart tells a story that no survey or forecast could manufacture: the share of tokens from U.S. organizations routed to Chinese-origin models climbed from roughly 4.5% in the first half of 2025 to approximately 45–46% by mid-2026. That is a roughly 10x move in eighteen months, measured in actual production routing behavior across thousands of U.S. organizations.
The climb was not smooth. Early spikes into the 20s tracked major Chinese model releases — moments when developers ran experiments and evaluations on newly dropped weights. Then a sustained surge from early 2026 pushed the line past 40% and held it there. Platform-wide, the inversion is now complete: Chinese-origin models account for roughly 46% of all routed tokens versus roughly 36% for U.S.-origin models, a reversal from twelve months earlier when U.S. models held the clear majority.
The driver is not sentiment. It is price. DeepSeek V4 Flash lists at approximately $0.14 per million input tokens. OpenAI’s GPT-5.5 runs around $5.00 per million input tokens — roughly a 35x gap on the input side alone. DeepSeek, as a result, now commands about 16.3% of all token volume on OpenRouter, making it the single largest provider on the platform, ahead of Google, Anthropic, and OpenAI individually.
The Routing Shift — Timeline
H1 2025 — Baseline
Chinese-model share of U.S. org tokens on OpenRouter: ~4.5%. U.S. models hold a dominant majority platform-wide.
Mid-2025 — First Spikes
Major Chinese model releases trigger early routing experiments; share spikes into the 20s, then partially retraces. Monthly volatility: 30–46% swings.
Early 2026 — Sustained Surge
DeepSeek V4 Flash and peer open-weight releases push Chinese share past 40% and sustain it. DeepSeek becomes OpenRouter’s #1 provider at 16.3% of all token volume.
Mid-2026 — Inversion Complete
Chinese-origin: ~46% of all routed tokens. U.S.-origin: ~36%. The platform majority has flipped. Export controls and security warnings have not reversed the economics.
Token Share by Origin — OpenRouter, Mid-2026
Chinese-origin models~46%
U.S.-origin models~36%
DeepSeek alone (of all tokens)16.3%
The key insight: This is not a developer preference survey or a model benchmark leaderboard. It is production routing behavior — the decisions that U.S. organizations make when actual tokens, actual compute costs, and actual applications are on the line. A 10x rise in Chinese-model share in 18 months, inside an active U.S.-China AI cold war, is the clearest possible signal that price is the primary clearing mechanism at the inference layer.
The Structural Read
Before the three reads, the honest hedge: OpenRouter is one aggregator, and its user base skews toward cost-sensitive, experimentation-heavy developers — indie builders, early-stage startups, and API-first teams rather than heavily regulated enterprises. Token counts include testing, evaluation, and CI/CD traffic alongside production inference. Monthly swings of 30–46% mean any single month is noise. What is signal is the secular direction: a 10x move over 18 months in the same dataset, measured the same way throughout.
Intelligence Is Commoditizing Where It’s Hardest to Spin
The inference layer is not differentiating on brand, safety posture, or national origin among cost-sensitive buyers. It is differentiating on price per token and capability-per-dollar. When the cheapest credible token increasingly carries Chinese weights — and those weights are open, self-hostable, and API-accessible — the moat of the closed U.S. frontier lab at the commodity tier collapses. This is the production barbell in action: premium closed models for the tasks that justify the price; cheap open-weight models — increasingly Chinese — for everything else.
Read 1 — Demand-side proof, domestically. The price gap between GPT-5.5 and DeepSeek V4 Flash — roughly 35x on input tokens — is not an abstraction. U.S. developers are acting on it at scale, in production, right now. The OpenRouter data is the demand-side confirmation that the inference barbell is real and accelerating. Cost-sensitive workloads are leaving U.S. frontier models not because of capability gaps but because of price gaps.
Read 2 — Price beats flag in software. U.S. firms are adopting Chinese models en masse despite export controls, active security warnings, and a U.S.-China AI cold war that produced moves like Alibaba restricting Claude Code on its platforms. At the token layer, cost and openness are beating national-security caution. This is the mirror image of the hardware story: in atoms, provenance still wins — buyers pay a premium for a non-red, China-free supply chain in semiconductors and drones. Software commoditizes past the flag. Atoms don’t — yet.
Read 3 — Open weights are the vector. This is not a Chinese closed-API story. DeepSeek’s weights are open. That means a U.S. developer can route to the API for price, or self-host the weights for sovereignty. The open-vs-closed contest is being settled in production by economics: openness is what converts a price advantage into actual switching behavior, because you can run the weights yourself without being locked to a foreign API endpoint. As with Vercel’s AI Gateway data, the routing layer is the vantage point — the gateway sees what the labs’ own dashboards never will.
“Openness is not a feature — it is a distributionstrategy. When a model’s weights are open, every price advantage compounds because any buyer can act on it immediately, without negotiating an API contract, without accepting a foreign dependency, and without waiting for a procurement cycle. The routing data shows that the open-weight price advantage has crossed the threshold where it changes aggregate behavior at scale.”
Three Implications
IMPLICATION 1 — U.S. Frontier Labs Are Losing the Commodity Tier
Gennaro is the creator of FourWeekMBA, which reached about four million business people, comprising C-level executives, investors, analysts, product managers, and aspiring digital entrepreneurs in 2022 alone | He is also Director of Sales for a high-tech scaleup in the AI Industry | In 2012, Gennaro earned an International MBA with emphasis on Corporate Finance and Business Strategy.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
