The most revealing number in DeepSeek’s second funding round is not the amount raised. It is the amount that was not raised. On October 13, 2026, Caijing reported that DeepSeek’s latest round has grown from a planned 500 billion yuan to roughly 1000 billion yuan — a doubling in under a fortnight — while its pre-money valuation stays anchored at 5000 billion yuan. A round that doubles in size without moving the valuation is not a company scrambling for cash. It is a company rationing access to itself.
This is DeepSeek’s second external financing in less than half a year. The first, closed in June 2026, raised 510 billion yuan at a near-4000 billion yuan post-money valuation and set the record for the largest single round in China’s AI industry. That round was led by roughly 200 billion yuan from founder Liang Wenfeng personally, with Tencent, CATL, NetEase, JD.com, and IDG Capital participating, plus a strategic follow-on from the National AI Industry Investment Fund. Within weeks of closing, DeepSeek opened the second round at 500 billion yuan targeting a 5000 billion yuan valuation. The market simply refused to fit inside that envelope.
The structural fact underneath the doubling is oversubscription. According to Caijing, more than 1000 billion yuan of capital expressed investment intent in the first round; only 500 billion was accepted. The money that did not get in simply waited for the second round and fought for the new slots. Doubling the round, then, is not DeepSeek deciding it needs twice as much capital. It is DeepSeek deciding to let twice as many strategic investors through the door at a price it has already set.
The Investor Roster Tells the Real Story
The composition of the round is more important than its size. New entrants include Shixi Capital (backed by GigaDevice), SMIC Polaris (the investment arm of SMIC), CPE Source-Fung, Legend Capital, and Geely. Returning shareholders include CATL, Tencent, IDG Capital, Lisi Capital, and Shixiang Capital. CATL is reportedly adding about 100 billion yuan, roughly doubling its first-round commitment of around 50 billion.
Three investor clusters stand out, each with a different strategic logic:
- Compute supply chain. SMIC Polaris and Shixi Capital (GigaDevice) represent the silicon layer. Their participation is not financial diversification; it is a forward integration play that ties model training demand to domestic chip manufacturing capacity.
- Power and infrastructure. CATL’s doubling-down maps onto its thesis of supplying energy storage and data-center power infrastructure to the AI buildout. For CATL, an AI data center is a battery customer at industrial scale.
- End-market demand. Tencent (AI integration across its platform stack) and Geely (automotive intelligence) represent buyers of model capability who want a seat at the table rather than a pure API dependency.
The round is, in short, a supply-chain syndicate masquerading as a financing. DeepSeek is not just raising money; it is locking in the upstream and downstream relationships it needs to scale compute and distribution.
Revenue Quality, Not Just Revenue Quantity
The willingness of industrial capital to pay an anchored price depends on the unit economics beneath the headline ARR. The numbers, as reported by The Information and recounted in investor meetings, are striking:
- Annualized revenue run-rate has crossed $1 billion, more than doubling from under $500 million a few months earlier.
- The core driver was the mid-August API price increase, which raised call prices to 2.3–4.5x their previous level, with cached-hit pricing rising the most.
- Crucially, the price increase did not cause customer churn. Demand remained strong.
- For the first seven months of 2026, revenue was roughly 475 million yuan — about ten times full-year 2025 revenue — while the API business gross margin reached 82.9% and the combined gross margin was 44.6%.
- The net loss narrowed to roughly 715 million yuan for the period, down from 935 million for all of 2025.
This is the profile of a company that used low pricing to build a market moat and is now converting that moat into margin. The price-elasticity story — raising prices 2–4x without losing customers — is the single most important signal in the whole financing. It means DeepSeek’s API demand is, at the margin, inelastic: customers have built workflows on the model and cannot easily substitute away.
There is also an untapped revenue lever: the consumer chatbot app remains free and ad-free, contributing no direct revenue. The $1 billion ARR was achieved without activating consumer monetization, which leaves meaningful upside for the public-market narrative.
The Valuation Math and the IPO Math
A 5000 billion yuan pre-money valuation against roughly $1 billion ARR implies a P/ARR of about 70x using a straight yuan-to-dollar conversion. Reported comparisons vary: Rhodium Group estimates put DeepSeek’s P/ARR at roughly 163x, versus about 34x for OpenAI and 21x for Anthropic, while a simpler 10x ARR read lands near 70x. Any way you calculate it, the multiple is high — and it is high specifically because the growth rate is high. ARR went from under $500 million to over $1 billion in about three months. If that cadence continues into 2027, the P/ARR compresses quickly.
The IPO preparations are moving in parallel. On September 21, former Hillhouse Ventures partner Yan Wentao joined as CFO, filling a role that had been vacant since the company’s founding. Yan’s resume includes investments in ByteDance, Zhipu AI, MiniMax, and J&T Express — experience spanning the full financing-to-listing lifecycle. Reuters reported on September 9 that DeepSeek has engaged CITIC Securities to prepare for a STAR Market (Shanghai Science and Technology Innovation Board) listing, with due diligence already underway. South China Morning Post had earlier reported a potential filing as early as late 2026 and a 2027 listing target.
Where the Money Goes: Compute in Inner Mongolia
The financing’s ultimate destination is compute. DeepSeek is reportedly planning a large data center in Inner Mongolia deploying at least 160,000 high-end Huawei AI accelerators, and has open-sourced infrastructure components for Huawei’s AI chip platform to deepen that ecosystem binding. Founder Liang Wenfeng has told investors that the company allocates over 70% of compute to model training, leaving under 30% for inference serving — a ratio that explains both the pressure to build training capacity and the strategic logic of lightweight models that can run on consumer GPUs for inference.
The Inner Mongolia deployment is the concrete expression of the compute migration already underway: DeepSeek is accelerating its shift to domestic chips for training, while using model efficiency to reduce inference dependence on high-end silicon.
What the Round Signals
The combination — doubled round size, unchanged valuation, industrial-capital-heavy roster, $1 billion ARR, narrowing losses, and an accelerating IPO — sends a coherent signal. DeepSeek does not need capital; it needs strategic capital that brings supply-chain and distribution leverage. The 5000 billion yuan valuation anchor is not a number to be negotiated upward; it is a filter for investors willing to accept DeepSeek’s terms.
For the broader industry, the round confirms two things. First, the model layer in China is consolidating around a small number of winners that can absorb capital at this scale. Second, the relationship between model providers, chip manufacturers, and power-infrastructure providers is becoming equity-based rather than transactional. DeepSeek is turning its financing rounds into a mechanism for vertical integration by proxy.
The question the public markets will eventually have to answer is whether $1 billion ARR at a 5000 billion yuan valuation can be defended by sustained growth, and whether the Inner Mongolia compute buildout and domestic-chip training efficiency can deliver the next order of magnitude of model capability. Between now and the 2027 listing window, those are the metrics that matter.