DeepSeek Seeks Over $70 Billion at a $74 Billion Valuation

SHANGHAI — DeepSeek, the Hangzhou lab that turned cut-rate inference into a geopolitical weapon, is now asking investors to underwrite the bet at scale. The company is reportedly seeking to raise more than $70 billion at a roughly $74 billion valuation — a jump from about $50 billion just weeks earlier — while quietly laying groundwork for an onshore listing on Shanghai's STAR Market as soon as next year, according to people familiar with the talks.

The number is arresting precisely because the revenue behind it is not. DeepSeek's annualized revenue is estimated at just $400 million to $500 million, a fraction of what Western frontier labs are pulling in. A $74 billion valuation on half a billion dollars of sales is not a bet on this year's cash. It is a bet that open-weight models will swallow the high-volume, price-sensitive core of the AI market — and that DeepSeek will be the reference implementation everyone else is measured against.

A valuation built on price, not profit

The mechanics of the raise are aggressive by design. DeepSeek is aiming to pull in roughly 50 billion yuan against a contemplated valuation near 500 billion yuan, or about $74 billion, according to reporting from Reuters and others. That follows the company's first outside financing round in June, which brought in a comparable sum at a post-money valuation of roughly $50 billion. In other words, DeepSeek has repriced itself upward by nearly 50 percent in a matter of weeks.

What is funding that markup is not margin — it is momentum. DeepSeek V4 has topped open-weight leaderboards for months, and its stable release is expected around July 24. The model's pricing is the number the rest of the industry braces against: V4 Pro runs roughly $0.435 per million input tokens and about $0.87 output, with a cheaper V4 Flash tier far below that. Rounded, that headline figure of about $0.44 sits some 75 percent below comparable closed rivals. DeepSeek made a 75 percent cut to V4 Pro permanent earlier this year, and the shock waves dragged competitor pricing down across the Chinese market.

The Shanghai listing plan is the other half of the story. The company has begun early work on a STAR Market flotation — China's Nasdaq-style venue — with an internal goal of filing later this year, though people familiar with the discussions caution that both the fundraising terms and the IPO timeline could still shift. Keeping the listing onshore keeps the value inside China's own capital markets, a choice that fits neatly with a week defined by decoupling: Xi Jinping used the World AI Conference in Shanghai to launch a 29-country governance body and to champion open-source AI as China's offer to the Global South.

The playbook: subsidize now, own the standard later

DeepSeek is running a recognizable platform strategy. Cut the price far enough to make adoption a no-brainer, capture the developers and enterprises building high-volume workloads, and monetize later through scale, services, and ecosystem lock-in. The company said after its June round that it planned to double its workforce across data-center and AI-agent teams — the kind of build-out that a per-token price war alone could never fund, which is exactly why the equity raise matters.

The tension is that the weapon cuts both ways. A 75 percent discount is a margin ceiling as much as a market lever. DeepSeek is effectively spending profit it does not yet earn to win a standard, and the strategy only pays off if prices eventually recover or if services revenue arrives to fill the gap. If the open-weight price war never lets pricing normalize, the company could find itself owning the market and still struggling to earn on it.

DeepSeek is not alone in the offensive, and that is part of the pitch to investors. Moonshot AI's Kimi K3 topped the Frontend Code Arena days ago with open weights due within weeks, and Thinking Machines — the startup founded by former OpenAI CTO Mira Murati — released the 975-billion-parameter open model Inkling on a reported $2 billion seed. Four of July's most-discussed model releases are open-weight. The camp now has both the models and the capital to force every closed lab to defend its pricing.

Analysis: what a $74 billion bet is really pricing

Strip away the leaderboard drama and the valuation is a wager on where value settles in AI. If frontier capability keeps commoditizing — and open weights topping the charts is the strongest evidence yet that it is — then the durable money moves away from the model itself and toward the layers around it: inference infrastructure, enterprise services, and the standards that high-volume workloads standardize on. DeepSeek is trying to own that last layer.

That is why the modest revenue is not the disqualifier it looks like. Investors buying in at $74 billion are underwriting a thesis, not a P&L: that open-weight models dominate the price-sensitive bulk of AI demand, and that DeepSeek becomes the default. The comparison that stings for Western labs is Anthropic, which reportedly filed confidentially this month toward a possible trillion-dollar IPO on roughly $47 billion of annualized revenue. DeepSeek is asking for a fraction of that valuation on a fraction of a percent of that revenue — the pure-play version of the open-weight bet.

The risks are real and stackable. Onshore listings are subject to Chinese regulatory approval and market conditions that can move fast. The revenue base is thin. And the price war DeepSeek started is not obviously winnable in the sense of ever producing fat margins. But the strategic logic is coherent, and the capital lining up behind it suggests the market is taking the open-weight thesis more seriously than it did even a quarter ago.

What to watch next

- The July 24 V4 release. A stable, chart-topping open-weight model on schedule keeps the valuation narrative intact; any slip hands ammunition to skeptics. - Whether the raise closes at $74 billion. Terms are early and could change. A completed round at that mark would validate the repricing; a haircut would signal cooling enthusiasm for pre-revenue open-weight bets. - The STAR Market filing. Watch for an actual submission later this year and how Chinese regulators treat a marquee AI listing. - Pricing discipline across the sector. If DeepSeek, Kimi, and Inkling keep undercutting, watch whether closed labs cut prices in response — the clearest sign the open-weight camp is winning the argument on cost.

Reporting drawn from Reuters, Tech Startups, BenchLM, and Build Fast with AI. Revenue and valuation figures are reported estimates and remain unconfirmed by DeepSeek.