DeepSeek's $12-15 Billion Funding Round Ahead of IPO

If DeepSeek's round closes near its upper range, the implied valuation lands around $75 billion - a fraction of OpenAI's reported $1.4 trillion target in its concurrent $30 billion raise. That gap is notable given both companies are chasing the same thing: capital to buy compute and fund a landmark IPO. Market commentary around the deal has floated two competing explanations - either a persistent discount on Chinese AI assets, or inflation in US valuations - and it is genuinely unclear which read is correct.

The skepticism around pricing predates this round. Kevin Xu, founder of Interconnected Capital, has argued that DeepSeek already proved Chinese labs can build frontier models despite export-control constraints ^[1], which cuts against the idea that a lower valuation reflects lower technical capability. PitchBook's Melanie Tng goes further, arguing DeepSeek's cost-efficient approach to model-building challenges the sustainability of billion-dollar training budgets elsewhere ^[2]- which would mean the "discount" is DeepSeek undercutting the price of frontier AI, not trailing it.

DeepSeek's spending plans point to a change in identity, not just a bigger bank account. The company intends to deploy at least 160,000 Huawei AI accelerators in a new data-center buildout, part of a broader push - driven by the same US export controls that limit access to Nvidia's top chips - for Chinese AI firms to build out their supply chain domestically ^[3]. That shift explains why Tencent and CATL, not just AI-focused funds, wanted in: Tencent is working to embed DeepSeek's models directly into WeChat and its other consumer apps, while CATL - better known for EV batteries - wants to supply power and cooling hardware to AI data centers as compute demand grows ^[4]. Both bets only pay off if DeepSeek keeps scaling infrastructure rather than just publishing papers - a shift that starts to resemble how OpenAI and Anthropic increasingly operate, less pure research lab, more compute-hungry infrastructure buyer.

The number attached to this financing has moved fast. DeepSeek closed its first-ever external round at about $7.4 billion in June 2026, with a post-money valuation above $50 billion ^[5]. By July it was reportedly seeking fresh capital at a valuation near $74 billion ^[6], and in August it resumed an $8 billion round with Monolith reportedly in the mix ^[7]. The round now being reported started with a target of roughly $7.4-7.5 billion and has since blown past it ^[8], with signed term sheets pushing the total toward 80-100 billion yuan - $12 to $15 billion - after demand surged following September's release of DeepSeek-V4.1-Flash ^[9]. In four months, the number attached to DeepSeek has gone from a $50 billion valuation to a fundraising target nearly triple that size.

DeepSeek's capital structure carries an unusual signature for a company raising at this scale. In its first external round, founder Liang Wenfeng put in roughly $3 billion of his own assets, and outside investors participated through a limited partnership he manages that carries no voting rights. The only outside party granted direct equity and voting privileges was a state-backed AI fund ^[5]. That arrangement keeps control concentrated with Liang even as the company takes on Tencent- and CATL-sized checks, and it is worth watching once DeepSeek restructures ahead of its planned Shanghai listing - a public listing typically demands far more conventional governance than a founder-controlled limited partnership allows.