DeepSeek's new funding round could reach $15B, Bloomberg reports

DeepSeek’s second outside funding round has grown far beyond its target: the Chinese AI lab set out to raise roughly $7.5 billion, and is now nearing at least $12 billion, according to Bloomberg. People familiar with the matter said the round could end up close to $15 billion, with Tencent and battery maker CATL writing the largest checks.

The size of the round puts fresh attention on who owns a piece of DeepSeek, and on the role of Chinese state capital in a lab best known for open-weight models. In DeepSeek’s first round in June, a state-backed AI fund reportedly received voting rights that other investors did not get. None of the figures in this article have been confirmed by DeepSeek; all of them come from news outlets citing unnamed people.

What is reported about DeepSeek’s new round

The jump is large for a single round. Even the original $7.5 billion target carried a valuation of about $75 billion, according to the report. If the round lands near $15 billion, it would bring in roughly twice what DeepSeek raised in its first external round in June.

Tencent and CATL are putting in the largest amounts. None of the three companies would discuss the deal: CATL declined to comment, and neither Tencent nor DeepSeek responded.

The round did not run smoothly. DeepSeek paused it earlier this year after remarks by founder Liang Wenfeng about the company’s reliance on Nvidia chips went viral. Once the round closes, DeepSeek plans to restructure ahead of an initial public offering (IPO) in early 2027, according to the same report.

Item	What is reported	Basis
Original target, new round	About $7.5 billion, at a valuation of about $75 billion	Bloomberg, via coverage
Current size, new round	At least $12 billion, possibly close to $15 billion	Bloomberg, people familiar
Largest backers, new round	Tencent and CATL	Bloomberg, via coverage
First round (June), size	Around 50bn yuan (Series A)	South China Morning Post
First round, valuation	About 400bn yuan ($59bn) post-investment	South China Morning Post
IPO timing	Early 2027, after a restructuring	Coverage of Bloomberg report

How DeepSeek’s valuation climbed in six months

The rise has been fast. In May, DeepSeek’s valuation was reported to have jumped from $20 billion to $45 billion within weeks as it raised its first venture capital round. At that point, the round was expected to be led by China’s state chip investment fund, with Tencent and Alibaba also in talks to participate.

By June, the South China Morning Post reported that DeepSeek had closed that first external round, raising around 50bn yuan at a post-investment value of about 400bn yuan, or $59bn. A later account describes the same June round as raising about $7.4 billion at a valuation above $50 billion. The reports use different currencies and rounding, so the exact numbers should be treated as approximate.

For comparison, the June valuation was reported as higher than Moonshot AI at $30bn and Hong Kong-listed MiniMax AI at around $17.7bn, but lower than Zhipu AI, whose Hong Kong-listed shares are worth about $93bn.

Where state money sits in DeepSeek’s cap table

The clearest public picture of DeepSeek’s ownership comes from the June round. According to the South China Morning Post, Liang funded about 20bn yuan of that round from his personal fortune, nearly half the total. Tencent and CATL reportedly contributed 10bn yuan and 5bn yuan, making them the largest outside backers, while NetEase and JD.com each put in 3bn yuan.

The state’s direct stake in that round was small by comparison. China’s National Artificial Intelligence Investment Fund (NAIIF) reportedly contributed nearly 1bn yuan. NAIIF is a joint venture between state-backed Guozhi Investment (Shanghai) Private Equity Fund Management and Big Fund Phase III, a state investment vehicle.

What stands out is the terms. A Beijing-based investor told the Post that none of the investors were expected to have board seats, and that the investments would carry a five-year lock-up, a period during which investors cannot sell, intended to discourage short-term profit-seeking. The Information, however, reported that NAIIF was an exception: it was given both voting rights and freedom from the lock-up.

That matters because the round was built to keep control with the founder. Liang’s large personal contribution gives him significant control, so that he can keep the company aligned with his focus on artificial general intelligence (AGI), meaning AI that equals or surpasses human intelligence. At an investor meeting, he said what “DeepSeek wants to do is anything relevant to enhance the intelligence” of AI models, and that the company is not interested in anything else, according to the Post. A state fund with voting rights is the one reported exception to that founder-first design.

The sources available for this article do not break down how much of the new, larger round comes from state-owned investors, or whether any new backers received NAIIF-style rights. Those are the key details to confirm as the round closes.

Why DeepSeek is raising at all

DeepSeek was initially funded by Liang’s High-Flyer Quant hedge fund and has kept a focus on technical work over profit-seeking. It is understood to have sought outside money mainly to set a valuation benchmark that helps it retain staff amid aggressive poaching by competitors. The May reporting made the same point: the fundraise is driven by the need to offer employee equity to keep researchers.

Investor appetite, according to the October report, comes down to one model: DeepSeek’s V4-Flash, which has set new marks for cost and performance against Anthropic and OpenAI. That claim is the publication’s characterisation, not an independent benchmark result cited in the source. The same report notes that V4-Pro and V4-Flash came out in April as preview versions, not finished products.

Questions about how DeepSeek builds its models have also drawn attention; Unicorn API covered a report that DeepSeek and Moonshot trained new models on OpenAI’s GPT-6 Astra. Chinese AI companies are raising large sums more broadly, including Kuaishou’s Kling AI video unit.

Chips: from Nvidia to Huawei to DeepSeek’s own

Hardware is central to the story. The viral remarks that paused the round were about DeepSeek’s reliance on Nvidia chips. The company is now building a data center that will hold at least 160,000 AI chips from Huawei, and is developing its own inference chip, the hardware that runs trained models to answer requests, so it relies less on both Nvidia and Huawei.

Earlier reporting already described DeepSeek’s models as optimised to run on Huawei chips, positioning the lab as a key player in China’s push for AI self-sufficiency. State investment funds, including the chip-focused ones named in the May and June reports, fit that national goal.

What it means for developers using DeepSeek models

For builders, the main question is whether more money, a state shareholder with voting rights and a planned IPO change how DeepSeek releases models. On the record, the answer so far is no. Liang wants DeepSeek to keep releasing open models with freely available weights, and says the core product will stay free while he asks investors for as much as $15 billion. DeepSeek’s models remain open-weight, according to the May reporting.

A few practical points follow from the sources:

• Open weights are a stated intention, not a contract. The commitment comes from the founder’s reported wishes. Restructuring for an IPO and new investor rights could put pressure on that over time; teams that depend on DeepSeek weights should keep local copies of versions they rely on.
• V4 models are still previews. V4-Pro and V4-Flash were released in April as previews, so behaviour and availability may change before final versions.
• Hardware shifts may affect serving. A move toward Huawei chips and a custom inference chip could change how DeepSeek runs its own hosted service, though the sources say nothing about effects on speed, pricing or regions.
• Governance is founder-led, with one state exception. Liang reportedly keeps significant control, but a state-backed fund holds voting rights. Teams with compliance requirements around state-linked vendors may want to factor this in.

Open questions and what to watch next

Several things remain unconfirmed. DeepSeek has not commented on the round, and the reported size has moved quickly, from a $7.5 billion target to a possible $15 billion. The exact split between private companies and state-owned investors in the new round is not shown in the sources used here, nor whether any new investor gets voting rights or lock-up exemptions like NAIIF’s.

Watch for:

• A final closing figure and valuation for the new round, and the investor list behind it.
• Whether state-backed funds take a larger share than the nearly 1bn yuan NAIIF reportedly put in during June, and on what terms.
• The restructuring ahead of the planned early-2027 IPO.
• Final, non-preview releases of V4-Pro and V4-Flash, and whether they ship with open weights.
• Progress on the Huawei-based data center and DeepSeek’s own inference chip.