Key Takeaways:
DeepSeek's first-seven-month revenue hit 4.75 billion yuan ($70.7 million), about 10 times its full-year 2025 total, as the Chinese AI start-up closes a funding round valuing it at roughly $74 billion.
DeepSeek's first-seven-month revenue reached 4.75 billion yuan ($70.7 million), about 10 times its full-year 2025 total, according to The Information, as the Hangzhou-based AI start-up nears a funding round valuing it at roughly 500 billion yuan ($74 billion) before investment. The company is seeking to raise about 50 billion yuan in the round, expected to close before the end of August, people familiar with the matter said.
Existing backers including venture funds Monolith and Shixiang Capital and battery maker Contemporary Amperex Technology Ltd are participating, joined by new investors CPE, Legend Capital and Stony Creek Capital, a semiconductor-focused private equity firm linked to ChangXin Memory Technologies chairman Zhu Yiming. Funds backed by memory chip designer GigaDevice and Hefei state investment vehicles are also involved, the people said.
The API business, which sells model access to outside developers, posted an 82.9 percent gross margin in the first seven months, while overall gross margin came in at 44.6 percent. That compares with OpenAI's 39 percent gross margin on $5.7 billion in first-quarter revenue and Anthropic's expected 63 percent for 2026, up from 40 percent in 2025. DeepSeek's net loss narrowed to 715 million yuan from 935 million yuan in 2025, according to Bloomberg.
Pricing power and infrastructure spend
DeepSeek raised API prices this month, lifting flagship V4-Pro output fees to $3.96 per million tokens at peak times and $1.98 off-peak, from $0.87 previously. The increase still leaves it far below Moonshot AI's Kimi K3 at $15 per million tokens and Anthropic's Claude Opus 5 at $25, preserving the cost advantage that drove adoption of its open-weight V4 family, particularly the lightweight V4-Flash, among global developers.
Infrastructure spending has climbed in step with revenue. The company spent about 11 billion yuan on AI servers and chips in the first seven months, versus 1.2 billion yuan for all of 2025, a near-tenfold jump that reflects a push to expand compute capacity for model iteration and user growth. The Information last month reported DeepSeek's annualized revenue had reached $400 million to $500 million, a range the latest figures confirm.
IPO path
DeepSeek completed its first 500 billion yuan funding round in June and began a second almost immediately, hiring investment banks to prepare for a Shanghai Star Market listing. The company could file as early as year-end, targeting a 2027 debut, people familiar with the matter said. DeepSeek did not respond to requests for comment.
The funding and IPO plans give DeepSeek a capital base to sustain its cost-efficient model strategy, which has pressured US rivals on price. Its 82.9 percent API margin shows the economics of running models on fewer chips can rival the largest American labs even as revenue remains a fraction of OpenAI's or Anthropic's. A Shanghai listing would also hand Chinese investors a direct stake in the country's most prominent AI champion, a milestone for a sector where most value has accrued to US-listed names. The narrowing loss and high-margin structure strengthen the case for long-term commercial viability as the company scales.
This article is for informational purposes only and does not constitute investment advice.
