A Price Hike, Not a Price Cut
DeepSeek raised its API prices between 2.3 and 4.5 times last month, depending on the model, with peak-hour pricing on its V4 models roughly quadrupling. That is not the number anyone expected from the lab that spent the better part of two years being cited as proof that AI inference was racing toward free.
Founder Liang Wenfeng told investors at a briefing that demand held anyway. Customers stayed, he said, even at the higher prices, and DeepSeek's annualized revenue run rate has reportedly reached about 1 billion dollars, more than double the under 500 million dollar pace of just a few months earlier.
The Number That Breaks the Race-to-Zero Story
A company racing prices to the floor to buy market share runs on thin or negative margins by design. DeepSeek's API business ran an 82.9 percent gross margin through July, a figure that belongs to a high-margin software business, not a loss leader.
That is the detail that matters more than the headline number. DeepSeek's low prices were never proof that Chinese open-weight labs could not charge more. They were a choice, made while the company built volume, and the August increase shows the ceiling on that choice was higher than the market assumed. DeepSeek's own prices reportedly remain among the lowest of any major model even after the hike.
From Disruptor to IPO Candidate
DeepSeek is finalizing a second outside funding round, targeting 7.5 billion dollars at a valuation of roughly 75 billion dollars, ahead of a planned listing on the Shanghai Stock Exchange. The company only began accepting outside capital at all in 2026, after closing an earlier round of about 7 billion dollars with founder Liang Wenfeng himself supplying much of it.
A loss-leading disruptor does not usually chase a public listing at that scale. A business demonstrating it can raise prices, keep customers and run software-grade margins does.
The Shift, By the Numbers
The move from August to now is easiest to see side by side.
| Figure | Detail |
|---|---|
| Annualized revenue run rate | About 1 billion US dollars |
| Run rate a few months earlier | Under 500 million US dollars |
| API price increase, August | 2.3x to 4.5x, by model |
| Peak-hour V4 pricing change | Roughly 4x |
| API gross margin through July | 82.9 percent |
| New funding round target | 7.5 billion US dollars |
| Target valuation | About 75 billion US dollars |
| Planned listing venue | Shanghai Stock Exchange |
What This Means for Your Own AI Budget
Plenty of enterprise AI budgets for 2027 assume the same curve DeepSeek was supposed to represent: more capability, lower cost, every year, driven by Chinese open-weight competition undercutting the majors. That assumption just took its first real hit from the lab that was cited as its best evidence.
None of this means AI inference gets more expensive across the board. It means the specific bet that competition would keep pushing prices toward zero indefinitely was never guaranteed, and the company best positioned to prove it wrong just did, without losing the customers it was supposed to need cheap prices to keep.
Read next: China Turns AI Tokens Into a Phone-Bill Perk | Gemini 3.8 Flash: Three Weeks, Pricing Risk
