OpenRouter’s Spending Data Shows Every Lab Growing — and Why the Growth Table Is Not a League Table

OpenRouter’s January–August 2026 spending figures show every lab on its platform growing — but a percentage change is a ratio between two of a single lab’s own numbers, and it says nothing about how that lab compares with any other.

What Happened

OpenRouter publishes spending data for the model labs available on its platform at openrouter.ai/data. The figures here are read from a published chart — not pulled from an API — and should be treated as approximate throughout. Measured against January 2026, August 2026 spending on the router was up approximately 2,425% for Moonshot AI, 1,925% for Z.ai (formerly Zhipu, listed on the Hong Kong Stock Exchange since January 2026), 1,000% for DeepSeek, 546% for Qwen (an Alibaba model family; Alibaba is publicly listed), 520% for OpenAI, 228% for Anthropic, and 163% for Google (Alphabet, publicly listed). OpenAI, Anthropic, Moonshot AI, DeepSeek, and OpenRouter are private companies.

Every lab in the dataset grew over the period. That sentence belongs alongside every number above and is usually omitted. This is not a winners-and-losers chart. It is a record of spending by one population — users of one router — over eight months, and all seven labs shown are above their January figure, though two of them fell month-over-month into August.

The chart carries a second panel showing month-over-month change into August, where the picture is less uniform. Moonshot AI rose 93% month-over-month. DeepSeek rose 40%, Qwen 33%, Google 23%, Anthropic 4%. Z.ai fell 13% and OpenAI fell 12% in the same month. A single month is not a trend, and nothing here should be read as one — but the month-over-month panel demonstrates that these series move in both directions at monthly resolution.

The Structural Read

The most common error made when reading this kind of market data is treating a growth table as a league table. They are different objects. A league table ranks entities against each other — it requires a shared metric with a common denominator. A growth table records the rate at which each entity changed relative to its own prior position. The two answer different questions, and conflating them produces conclusions the data cannot support.

This chart shows direction and rate for each lab individually. It supports no statement about the ordering between them, because the bases — the January 2026 starting amounts — are not shown. Without those bases, there is no scale, no revenue figure (pricing is also absent), and no market-wide denominator. So the data cannot establish which lab is ahead of or behind any other, and this article will not attempt to do so. Any size language — describing a lab as large, small, established, or emerging — would be importing information this dataset does not contain.

The month-over-month panel is worth reading carefully precisely because it is answering a different question. A vs-January figure falls whenever a given month’s spend is lower than the previous month’s, as OpenAI’s –12% month-over-month in August illustrates: the vs-January figure for OpenAI declined in August because August spending was lower than July’s. That is what month-over-month contraction means in this context. Two of seven labs contracted in August on that measure — in the same month that Moonshot AI rose 93% month-over-month. The range of outcomes at monthly resolution is wider than the cumulative picture suggests, and that is worth registering without over-reading a single month’s data point.

Four of the seven labs in this dataset are Chinese, and they are among those shown. That is an observation about this chart, not a finding about the industry, and the distinction matters. Without the starting amounts, the data cannot establish scale. Without pricing, it cannot establish revenue. Without a market-wide denominator, it cannot establish share. So nothing here supports a geopolitical, national, competitive, or strategic conclusion. Nothing here says any category of model is winning, gaining share, or displacing any other. No lab is characterised as rising or falling in the market, because this dataset cannot see the market.

Month-over-Month Change into August 2026 (Approximate)

91,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.

All figures in this article are read from a chart published at openrouter.ai/data and should be treated as approximate rather than precise. They measure the change in dollars spent on each lab’s models by users of one model router. They are not market share, not revenue, not total industry spending, and not a picture of enterprise buying. The chart does not show the starting amounts, so nothing here states or implies which lab is larger, smaller, ahead or behind any other; the data supports statements about each lab’s own direction and rate of change and no comparison of size between them. Every lab shown grew over the period, and nothing here frames the chart as winners and losers. A single month-over-month figure is not a trend. The observation that four of the seven labs listed are Chinese is an observation about the composition of this chart and not a finding about the industry. Without the starting amounts it cannot establish scale, without pricing it cannot establish revenue, and without a market-wide denominator it cannot establish share. Nothing here draws a geopolitical, national, competitive or strategic conclusion, claims that open-weight or Chinese models are winning, gaining share or displacing anyone, or characterises any lab as rising or falling in the market. Nothing here claims that spending by this population leads, lags, predicts or precedes spending by any other population; no dataset referenced measures such a relationship. No prediction of continuation or reversal is offered, no ratio is computed between labs or against any other dataset, and OpenRouter’s methodology is not criticised. OpenAI, Anthropic, Moonshot AI, DeepSeek and OpenRouter are private companies. Z.ai, formerly Zhipu, has been listed on the Hong Kong Stock Exchange since January 2026. Alphabet is publicly listed, and Qwen is a model family from Alibaba, which is also publicly listed. No claim is made about any share price. This is business analysis, not investment advice, no view is expressed on any security, and no recommendation is made.