DeepSeek Reports $70.7M Revenue in First 7 Months, API Gross Margin Reaches 82.9%
Nashnova编辑部
DeepSeek generated roughly RMB 475 million (~$70.7M) in the first seven months of 2026 — ten times its full-year 2025 revenue — while pushing a second funding round at a RMB 500 billion pre-money valuation and preparing a Shanghai STAR Market listing.
Revenue up tenfold — where is the money coming from?
First-seven-month revenue hit roughly RMB 475 million, about ten times the full-year 2025 figure. This means → DeepSeek has crossed from "tech darling" into a real revenue ramp.
The core driver is broad adoption of the flagship V4 model family, especially the smaller V4-Flash variant, popular with developers worldwide for its low cost and competitive capability.
This month DeepSeek sharply raised API pricing: flagship V4-Pro output jumped from $0.87 to $3.96 per million tokens at peak hours — roughly a 3.5× increase. Even so, the price remains far below Kimi K3 at $15 and Claude Opus 5 at $25.
An 82.9% API gross margin — what does that number mean?
Overall gross margin for the seven months was 44.6%; API revenue — selling model access to developers — carried a 82.9% margin.
In plain terms = for every dollar of API revenue, the cost was under 20 cents. That is exceptionally rare in AI.
This reflects DeepSeek's sustained push to cut inference costs: better infrastructure efficiency means fewer chips per query, turning "compute cheaply" into a profit edge.
For context, OpenAI posted a 39% gross margin in Q1; Anthropic projects its margin will rise from 40% to 63% by 2026. DeepSeek's API margin already exceeds both.
Losses are still growing — but slower than revenue?
Net loss for the first seven months was roughly RMB 715 million, versus a full-year 2025 net loss of RMB 935 million. This means → seven months consumed about 76% of last year's total loss.
The key point: revenue grew tenfold; losses grew far less. In plain terms = revenue is outrunning the burn, and the company is trending toward breakeven.
Infra spending surged eightfold — where did it go?
AI infrastructure spending — server leases, chip and compute-equipment purchases — reached roughly RMB 11 billion in seven months, versus only RMB 1.2 billion for all of 2025, an increase of about eight times.
In plain terms = last year the company spent RMB 1.2 billion on compute for the whole year; this year it spent RMB 11 billion in seven months. It is racing to stockpile capacity.
This reflects a dual strategy: squeeze unit inference costs for high margins on one side, and massively expand the compute base for model iteration and commercial scale-up on the other.
RMB 50B raise plus a STAR Market IPO — where is the timeline?
A first RMB 50 billion round closed in June; a second round of the same size launched almost immediately, at a pre-money valuation of roughly RMB 500 billion.
The company is in active talks with existing and new investors, aiming to close by late August. Returning backers include Panlin Capital, Shixiang Capital, and CATL; new parties in discussions include CITIC PE, Legend Star, semiconductor-focused Stony Creek Capital, a GigaDevice-linked fund, and a Hefei state-capital platform fund.
In parallel, DeepSeek has hired investment banks to prepare a Shanghai STAR Market listing, with a filing possible as early as year-end and a target listing in 2027. This means → fundraising and IPO tracks are running simultaneously, and accelerating revenue is the hardest bargaining chip in valuation talks.
Can the RMB 500 billion valuation hold?
Rapid revenue growth and high API margins give the valuation a data foundation, but whether RMB 500 billion can be realized at listing still depends on sustained commercial scale-up.
In plain terms = strong numbers are necessary but not sufficient — investors will ultimately judge whether the revenue growth curve holds through IPO.
市场有风险，内容仅供研究参考，不构成投资建议。