DeepSeek's books leak: $70.7 million revenue, $106 million loss
The most-watched lab in open weights just got something none of its Western rivals have volunteered: a public look at the P&L.
DeepSeek generated roughly $70.7 million in revenue during the first seven months of 2026 while posting a $106 million net loss, according to figures reported by The Information, citing sources. That top line is about ten times what the Hangzhou lab earned in all of 2025 — around $7 million of revenue against a $139 million annual loss. The company has not confirmed the numbers itself, so treat them as sourced rather than filed, but the shape of the curve is the story: revenue scaling fast while the burn actually shrank year over year.
Why it matters: frontier labs are famously opaque, and the few US peers that disclose anything report losses measured in billions per quarter. DeepSeek reaching eight-figure monthly-scale revenue on a nine-figure half-year burn suggests its much-discussed efficiency playbook — sparse expert routing, low-bit training, and aggressive memory compression that squeeze more capability out of fewer scarce high-end chips — isn't just good engineering, it's a survivable business model under export-control constraints. The money comes mostly from selling API access to its models, and the ramp lines up with the product cadence: we covered the launch of its paid flagship in DeepSeek ships V4 Pro, ending its flagship's four-month preview, and the pricing reset that followed in China's LLMs quit the price war as DeepSeek API jumps 11x. Raising prices tenfold while revenue grows tenfold is what margin repair looks like on a chart.
The other reason these numbers surfaced now: DeepSeek is widely reported to be preparing for public markets. A Chinese regulatory filing last month implied a valuation of around $52 billion, and the lab — historically funded by founder Liang Wenfeng's quant fund High-Flyer — has reportedly taken outside capital this year to bankroll bigger compute purchases. If a Shanghai listing is the destination, today's leak reads like a pre-marketing exercise: small next to OpenAI or Anthropic, but the only frontier-scale lab whose costs and revenues an investor can actually inspect. Cheap tokens, thin losses, a tenfold growth rate — that's a pitch.
What to watch: whether an official filing eventually confirms these figures, and whether DeepSeek keeps raising API prices as demand compounds.
Would you buy stock in a frontier lab that undercuts everyone on price and still loses money on every model it trains? Tell us in the comments.
Sources: The Information · Techmeme · LLMs Blog · Reuters
