DeepSeek revenue is around USD 70mln YTD, 10x the FY25 figure, The Information reports citing sources. API access gross maring 82.9%.

This is a media-sourced revenue estimate on a private company rather than a reported print, so it carries the usual caveats of unattributed figures: no audited basis, no visibility on costs below the gross line, and no confirmation from the company itself. Reports of this kind on closely held AI developers have historically mattered less for the absolute number than for what they imply about the unit economics of low-cost model provision; a gross margin at that level on API access, if accurate, sits well above what most foundation-model providers have disclosed and would undercut the assumption that cheap inference is structurally loss-making. The figure is small relative to incumbent cloud and model revenues, so the read-through is competitive rather than macro: pricing pressure on the Western model providers and on the capex-heavy inference stack, rather than a shift in aggregate AI spending. What tends to follow in these episodes is contestation of the numbers, either a company denial or rival estimates, plus scrutiny of how much revenue is API versus consumer and how much is domestic. Worth noting the source is a single outlet citing unnamed sources, and prior scoops of this kind on private AI firms have sometimes been revised materially. The nearest listed proxies are the global model providers and the semiconductor names whose demand assumptions embed Chinese AI spending.