DeepSeek's Bet on Pricier AI Pays Off: Revenue Crosses $1 Billion

A year ago, DeepSeek was mostly known for one thing: being cheap. Now the Chinese AI lab has more than doubled its annualized revenue to around $1 billion, and it did that partly by making its API more expensive, not less.

The price hike nobody walked away from

Founder Liang Wenfeng told investors at a recent meeting that DeepSeek's revenue run rate has jumped from roughly $500 million just a few months ago to about $1 billion now. The company got there in part by raising API prices — the per-token cost developers pay to call its models through code — by 2.3 to 4.5 times last month. That's a steep jump by any standard, yet reports from people briefed on the numbers say the customer base barely shrank.

That's the surprising part. Conventional wisdom says a price hike of that size on a developer-facing product should trigger an exodus, especially in the AI space where switching to a rival model is often a one-line code change. Instead, DeepSeek's API business is now reportedly running at close to 83% gross margin through July, a number that puts it in the same league as the margins US frontier labs talk about, despite DeepSeek operating at a fraction of their scale and infrastructure spend.

Cheap got DeepSeek noticed. Sticky customers are what's letting it charge more and keep them anyway.

Why this matters beyond one company's balance sheet

DeepSeek is reportedly finalizing a second funding round of roughly $7.5 billion at a $75 billion valuation, with a listing on Shanghai's STAR Market — the tech-focused board of the Shanghai Stock Exchange — expected to follow. A real revenue number, rather than just usage and download stats, is exactly what a company needs to justify that kind of valuation to public-market investors. It also changes the story around DeepSeek from "surprisingly good open model that spooked Silicon Valley" to "AI company with an actual, profitable business," which is a harder thing to dismiss.

For the wider AI industry, it's a data point in an argument that's been running all year: can anyone other than OpenAI, Anthropic and Google actually make money selling model access, or is everyone just burning cash to buy market share? DeepSeek's answer, at least so far, is that a lab can raise prices sharply on a loyal developer base and still grow.

What it means for Indian developers and startups

DeepSeek's low-cost API was a genuine draw for Indian AI startups and independent developers who built products on top of it specifically because it undercut OpenAI and Anthropic pricing by a wide margin. A 2.3x to 4.5x price increase changes that calculation fast, especially for early-stage Indian SaaS teams running on thin margins where API costs are one of the biggest line items.

There's a second layer to this for Indian users specifically. DeepSeek is a Chinese company, and India's Ministry of Electronics and IT (MeitY) has previously flagged concerns around Chinese apps and data handling, on top of the broader compliance questions any company processing Indian user data has to think about under the DPDP Act, India's data protection law. Startups weighing whether to route customer data through a Chinese AI provider now have to balance that against a fast-changing price it charges them, which makes homegrown alternatives like Sarvam AI and Krutrim more attractive on cost predictability alone, even before governance concerns enter the picture.

• DeepSeek's API price jumped 2.3–4.5x last month, yet its customer base reportedly held steady.
• Annualized revenue run rate: roughly $1 billion, up from about $500 million months earlier.
• Reported API gross margin through July: around 83%.
• Next funding round targets roughly $7.5 billion at a $75 billion valuation, ahead of a planned Shanghai STAR Market listing.

Code24 covered DeepSeek's funding talks when they first surfaced, and this is the clearest sign yet that the numbers behind that round are real rather than just growth-at-any-cost. The bigger question now isn't whether DeepSeek can raise money. It's whether developers everywhere, including in India, keep tolerating price increases from a lab they once picked specifically because it was the cheap option.