Oversubscribed: DeepSeek's Pre-IPO Round Balloons Past $12 Billion as Tencent and CATL Pile In

The most sought-after ticket in Chinese venture capital just got bigger. According to a Bloomberg News report published October 6, 2026, DeepSeek is close to securing at least 80 billion yuan — roughly US$12 billion — in its latest funding round, blowing past the startup’s own capital-raising target ahead of a landmark initial public offering planned for early 2027.

The numbers keep climbing

When DeepSeek’s first external fundraise took shape this summer, the plan was comparatively modest: about 50 billion yuan (US$7.4 billion) at a valuation near 500 billion yuan (US$74 billion) pre-money. That round closed at roughly 350 billion yuan valuation, with founder Liang Wenfeng himself participating alongside outside investors.

Interest in the current round outstripped expectations after the successful release of the company’s latest model, V4 Flash, which reset industry expectations for cost versus performance against leaders like Anthropic and OpenAI. Based on term sheets already signed, people familiar with the matter told Bloomberg the final tally could approach 100 billion yuan — close to US$15 billion, nearly double the original goal. The round closes soon, though details may still shift in final negotiations.

Who is writing the checks

Two names dominate the commitments: Tencent Holdings, the WeChat operator that previously landed the single biggest investment in DeepSeek at 10 billion yuan, and Contemporary Amperex Technology Co. (CATL), the world’s largest battery maker. Their motivations sketch the broader picture of how Chinese industry is positioning around AI.

Tencent aims to integrate frontier AI into its sprawling platform ecosystem — a defensive and offensive move as chat assistants, agents, and AI-generated content reshape the social and gaming businesses it dominates. CATL, meanwhile, is eyeing an increasingly lucrative business supplying components to the AI industry, including data centers, where backup power and energy density matter as much as compute.

CATL declined to comment. Tencent had no immediate comment. DeepSeek did not respond to an emailed request during a Chinese public holiday.

The Huawei angle: 160,000 homegrown accelerators

Perhaps the most strategically significant detail in the report is where the money is going. DeepSeek plans to deploy at least 160,000 of Huawei’s top AI accelerators at a massive data center it is building in Inner Mongolia — which would create one of the largest known clusters of Huawei AI chips anywhere.

That matters for two reasons. First, it signals that China’s leading open-model lab is doubling down on a domestically produced compute stack rather than depending on sanctioned Nvidia hardware. Second, DeepSeek has publicly released software it built with Huawei for programming AI chips — a foray into the tooling layer, not just the model layer, executed in close partnership with China’s hardware champion. If the cluster materializes at scale, it becomes the reference deployment for the entire sovereign-AI compute argument.

From hedge fund side project to IPO runway

DeepSeek’s origin story is now famous: it grew out of High-Flyer, the quantitative hedge fund Liang Wenfeng built first, which bankrolled the lab’s early research. Since the 2025 breakthrough model that matched Silicon Valley capabilities at a fraction of the cost, DeepSeek has become the standard-bearer for China’s AI ambitions — and for the open-weights movement globally.

The money has already changed its founder’s standing: Liang’s net worth more than doubled to around US$36 billion when the first external round closed this summer. In at least one meeting with investors this year, Liang pledged to keep developing open AI models while pursuing artificial general intelligence, and has made clear that pushing the frontier — not monetization — remains the primary goal.

Once the current round closes, the next step is a corporate restructuring to prepare for IPO — the mechanics of turning a hedge-fund offshoot into a listable public company. Rival Moonshot is on a parallel track, targeting its own early-2027 listing after closing funding at a US$50 billion valuation, setting up a twin-debut race between China’s two most-watched AI labs.

A round that nearly didn’t happen

The fundraise was not without turbulence. DeepSeek briefly suspended the deal earlier this year after comments widely attributed to Liang — discussing China’s reliance on Nvidia chips and the country’s persistent lag in AI sophistication relative to the US — went viral on Chinese social media. The episode illustrated the tightrope Chinese AI leaders walk: candid technical assessments that circulate as praise abroad can read as heresy at home. The round resumed, and the eventual oversubscription suggests the controversy did little to dampen institutional appetite.

Why it matters

Three takeaways stand out. First, capital intensity is now universal: the argument that Chinese labs could win on efficiency alone is fading as even DeepSeek — the lab that built its reputation on doing more with less — raises double-digit billions for compute. Second, the sovereign stack is consolidating: DeepSeek plus Huawei plus Tencent plus CATL is a full vertical of Chinese AI capability, from chips to models to distribution to energy. Third, the IPO window is real: with DeepSeek and Moonshot both targeting early 2027, public markets will soon get their first direct exposure to China’s frontier-lab economics, and the valuations discovered there will ripple through every private AI mark on Earth.

For a company that once prided itself on needing less money than everyone else, DeepSeek is about to demonstrate just how expensive the frontier has become.