DeepSeek, the Chinese lab that stunned the industry in early 2025 with a frontier-grade model trained for a fraction of the usual cost, is reportedly about to raise a great deal of money — and spend almost all of it on the one thing its cheap-training story was supposed to make less important: compute.

According to the Wall Street Journal and the South China Morning Post, DeepSeek is nearing a round of about 50 billion yuan — roughly $7.4 billion — at a pre-money valuation of around 500 billion yuan, or about $74 billion. Neither DeepSeek nor its investors have publicly confirmed the financing, and the round is described as close but not yet closed, expected to complete around the end of August. Every figure here is reporting, not an announcement.

What the money is for

The striking detail is where it is reportedly going: about one gigawatt of new computing capacity, plus the salaries needed to win an increasingly brutal talent war.

That is worth sitting with. DeepSeek's whole reputation was built on efficiency — doing more with fewer chips. A raise of this size, aimed squarely at raw compute, is a quiet admission that efficiency and scale are not alternatives. You still need an enormous amount of hardware to stay at the frontier, however clever your training recipe; the efficiency just changes how far each chip goes, not whether you need a great many of them.

Who is reportedly backing it

The reporting names existing backers said to be participating — among them Monolith, Shixiang Capital and the battery giant CATL — alongside a list of new investors said to be in discussions, including CPE, Legend Capital and state-linked investment vehicles from Hefei. Earlier DeepSeek rounds drew in bigger consumer-tech names such as Tencent, JD.com and NetEase; those are prior backers rather than confirmed participants in this one.

The common thread is that this is largely Chinese domestic capital, much of it with a state or strategic-industry flavour, funding a company Beijing plainly regards as a national AI champion.

The listing in the background

The round is being read as a step towards a stock-market debut. DeepSeek is reported to be targeting a listing on Shanghai's Star Market, with a filing possibly as early as the end of 2026 and a listing in 2027. That timing is a plan, not a commitment — the kind of forward-looking target that shifts easily and has not been confirmed by the company.

Why it matters

If the numbers hold, they place DeepSeek among the most valuable private AI companies in the world, and they say something about the shape of the race. The story of 2026 has increasingly been about the physical layer — power, chips, data centres — rather than model cleverness alone. A lab famous for frugality reportedly raising billions to buy a gigawatt of compute is that story in miniature.

The honest caveat is the one the sources themselves carry: until DeepSeek or its backers say so, this is a well-reported plan, not a done deal. It is worth watching whether the round closes on these terms — and whether the Star Market listing actually follows.