The takeaway: DeepSeek is preparing to raise fresh capital at a valuation of about $74 billion and has begun early work toward an initial public offering on Shanghai's STAR Market, according to a Reuters report on 20 July. It comes just weeks after the company's roughly $7.4 billion June round — and it tells you something that the firm that made frontier AI cheap is now raising like a frontier lab, and doing it entirely on home turf.

The numbers

DeepSeek is looking to raise as much as 50 billion yuan in a new round that would value the Hangzhou company at about 500 billion yuan — roughly $74 billion. Alongside it, DeepSeek has started early deliberations on a listing on the STAR Market, Shanghai's Nasdaq-style board for tech firms, with an internal target to complete an IPO filing before the end of the year. Reuters, citing sources, cautions that the fundraising and IPO plans are at an early stage and that the terms and timetable may still change.

Weeks, not quarters

What makes the figure land is how recent the last one is. In June, DeepSeek closed its first big outside raise — about $7.4 billion, at a post-money valuation near 450 billion yuan. That round was itself a departure for a company that had prided itself on staying lean and largely self-funded; it drew in founder Liang Wenfeng, Tencent, CATL and China's national AI fund, and DeepSeek said afterwards it planned to double headcount across data centres and AI-agent work. A jump from that round to a ~$74 billion valuation in a matter of weeks is not the trajectory of a company short of interest.

Why "onshore" is the story

The more telling detail is where the money is coming from and where the listing would be. A STAR Market IPO keeps DeepSeek funded in yuan, by domestic backers, and public on a Chinese exchange — at a moment when US public markets are largely out of reach for leading Chinese AI firms, amid national-security scrutiny on both sides. For DeepSeek, that is less a constraint than a design choice: it can raise megaround-scale capital and reach public markets without ever touching Western money. It is the financial mirror of the company's technical pitch — competitive frontier models, built and now funded inside China's own system.

There is an irony worth naming. DeepSeek's whole reputation was built on doing more with less — matching Western models at a fraction of the training and inference cost, and giving much of the recipe away. Raising at $74 billion doesn't contradict that so much as complete it: efficiency got it in the door; scale is what it is buying now.

The caveats

None of this is done. Reuters is explicit that the deliberations are early and the plans could change — no filing has been made, and valuations quoted before a round closes are targets, not facts. But the direction is unambiguous, and it rhymes with the rest of the week: Moonshot had to cap new subscriptions to its days-old Kimi K3 model because demand outran its GPUs. Between a record raise at one lab and a compute shortage at another, the same signal keeps showing up — Chinese appetite for home-grown frontier AI is running well ahead of the capacity to serve it.

Sources: Reuters via Investing.com, The Manila Times, Business Standard.