DeepSeek raises $7.4B in record China AI funding round

Chinese AI lab DeepSeek closed its first funding round at 50 billion yuan ($7.4 billion) under an unusual deal structure, marking the largest single AI capital raise in China’s history. The round catapults DeepSeek into the top tier of global AI labs by war chest size, directly challenging the funding dominance of US players like OpenAI and Anthropic. The deal’s structure, involving multiple state-linked investors and convertible instruments, signals Beijing’s willingness to deploy patient capital to bypass US export controls on advanced chips. DeepSeek has been developing large language models that rival GPT-4-class performance using a fraction of the compute, a claim that has drawn intense scrutiny from US intelligence agencies. The funding comes as the Trump administration tightens restrictions on AI model exports, most recently freezing Anthropic’s Mythos and Fable AI models for foreign distribution. DeepSeek’s raise is not just a corporate milestone; it is a direct response to the US chip blockade, proving that Chinese AI labs can still attract massive capital even when denied access to Nvidia’s H100 and B200 processors. Why this matters now: the $7.4 billion bet tests whether China can build frontier AI without cutting-edge US hardware, and the outcome will reshape the global balance of AI power.

Where the $7.4 billion came from

DeepSeek’s funding round is notable not just for its size but for its structure. The deal involved multiple state-backed Chinese investment funds, including entities linked to the Ministry of Science and Technology and the China Internet Investment Fund, according to sources familiar with the terms. The round was structured as a mix of equity and convertible bonds, giving investors downside protection while allowing DeepSeek to avoid a valuation that would trigger US sanctions review. The unusual structure reflects the constraints Chinese AI labs face: they cannot easily take money from US venture capital firms, and they must navigate Treasury Department rules that penalize any entity providing “material support” to Chinese AI development. By using convertible instruments, DeepSeek’s investors can defer valuation disputes while the lab proves its technology. The round also included participation from sovereign wealth funds and provincial government guidance funds, which typically require a domestic supply chain commitment. DeepSeek has pledged to source at least 70% of its compute from domestic chip suppliers, including Huawei’s Ascend 910B and Cambricon’s MLU370, within 18 months. That commitment is a direct response to US export controls that block Nvidia from selling its highest-performance chips to China. The round closed in late May after six months of negotiations, with the final amount exceeding the initial $5 billion target by nearly 50%.

How the money flows through DeepSeek’s P&L

DeepSeek will burn through its $7.4 billion war chest at an estimated rate of $800 million to $1.2 billion per year, based on its disclosed compute procurement plans and headcount growth. The largest line item is chip procurement: DeepSeek has committed to buying at least 100,000 Huawei Ascend 910B chips over the next two years, at a cost of roughly $2.5 billion, based on current market pricing of $25,000 per chip. The second-largest cost is talent: DeepSeek has been poaching researchers from Baidu, Alibaba, and Tencent with compensation packages that include base salaries of $500,000 to $1 million plus equity, according to recruitment data from Chinese tech job platforms. The company’s headcount has grown from 200 to 1,200 in the past 12 months, and it plans to reach 3,000 by the end of 2027. The remaining capital will fund data center buildout, including a new 500-megawatt facility in Guizhou province that will be powered by hydroelectricity. DeepSeek’s revenue model is still unproven: the lab generates less than $100 million annually from API access and enterprise licensing, meaning it will operate at a negative gross margin for at least three years. The funding round buys DeepSeek time to develop a commercial product, but it also creates pressure to show progress toward profitability. Investors are betting that DeepSeek can replicate the path of ByteDance’s Doubao, which went from zero to 100 million users in 18 months, but the AI lab faces steeper competition from Baidu’s Ernie Bot and Alibaba’s Tongyi Qianwen.

ByteDance, Huawei, and the chip supply chain shuffle

DeepSeek’s funding is part of a broader reordering of China’s AI chip supply chain. ByteDance, the parent company of TikTok, is in talks with Iluvatar CoreX to purchase AI chips for inference workloads, and is also considering a similar deal with Baidu, according to Reuters. Iluvatar CoreX would become ByteDance’s third major domestic GPU supplier after Huawei and Cambricon. The company reported 1 billion yuan ($148 million) in 2025 revenue, with about 90% coming from selling GPUs, and it expects to ship at least 50,000 chips to ByteDance this year, mostly for inference. That order alone would nearly double Iluvatar CoreX’s annual revenue. The deal reflects a strategic shift by Chinese hyperscalers: they are moving away from reliance on Nvidia’s A800 and H800 chips, which were designed to comply with US export controls but have since been banned, and toward domestic alternatives that offer lower performance but guaranteed supply. Huawei’s Ascend 910B is the current leader in this market, with an estimated 60% share of domestic AI chip sales in China, followed by Cambricon at 20% and Iluvatar CoreX at 10%. ByteDance’s diversification is a hedge against further US restrictions: if Washington bans all chip sales to China, ByteDance needs multiple domestic suppliers to keep its Doubao and recommendation algorithms running. The Iluvatar CoreX deal also signals that Chinese chip startups are gaining traction, albeit from a low base. Iluvatar CoreX’s Tiangai series chips offer roughly 60% of the performance of Nvidia’s A100 for inference tasks, according to benchmarks published by the company, but at 40% lower cost.

Downstream effects on hyperscalers, fabs, and enterprise buyers

DeepSeek’s funding and ByteDance’s chip diversification are creating ripple effects across the global AI infrastructure market. The most immediate impact is on data center construction in China: DeepSeek’s 500-megawatt facility in Guizhou is one of at least a dozen new AI data centers planned by Chinese labs and hyperscalers, representing a combined 5 gigawatts of new capacity by 2028. That demand is straining China’s domestic chip fabrication capacity. SMIC, China’s largest foundry, is running at over 95% utilization for its 7nm and 14nm nodes, and it has warned customers of lead times extending to 12 months. The chip shortage is pushing Chinese AI labs to consider alternative architectures, including analog computing and optical interconnects, which could reduce reliance on advanced lithography. Meanwhile, the US side of the market is seeing a different dynamic: SpaceX announced it is leasing AI infrastructure to Google for $920 million per month over 32 months, a deal that will bring $30 billion in revenue to SpaceX’s AI business. Google owns roughly 4.9% of SpaceX, worth over $100 billion as of Friday’s close, and SpaceX has become the cloud provider to Google. The deal is a stunning reversal of the rift between Elon Musk and Larry Page, who fell out over AI safety a decade ago. Now their companies are intertwined in a way that gives Google access to SpaceX’s Starlink-based edge computing network for its Gemini Enterprise customers. The SpaceX-Google deal also highlights the growing importance of non-traditional cloud providers: hyperscalers like AWS, Azure, and Google Cloud are increasingly leasing capacity from satellite operators, nuclear power plants, and even oil rigs to meet AI compute demand.

Policy signal: what the funding round says about US-China AI strategy

DeepSeek’s $7.4 billion raise is the clearest signal yet that China is treating AI as a national security priority on par with nuclear weapons and semiconductors. The involvement of state-backed funds and the unusual convertible structure indicate that Beijing is willing to absorb short-term financial losses to build a domestic AI ecosystem that can operate independently of US technology. The funding round also reveals the limits of US export controls: while Washington has blocked Nvidia from selling H100 and B200 chips to China, it has not prevented Chinese labs from raising massive capital and buying domestic alternatives. The US response has been to tighten controls on AI model exports, as seen in the Trump administration’s freeze on Anthropic’s Mythos and Fable AI models. That freeze has created an immediate opportunity for rivals: Cohere, the Canadian AI lab, reported “huge inbound” from international customers after the US block, according to Bloomberg. Cohere’s models are now being evaluated by European and Middle Eastern governments that previously relied on Anthropic. The policy dynamic is creating a bifurcated global AI market: one bloc centered on US models and chips, and another centered on Chinese models and domestic hardware. DeepSeek’s funding is a bet that the Chinese bloc can achieve parity within five years, even if it requires 2x to 3x more capital to compensate for less efficient chips. The next test will come when DeepSeek releases its next-generation model, which it claims will match GPT-5 performance using only Huawei chips.

The funding round also reshapes the competitive landscape for US AI labs. DeepSeek’s $7.4 billion war chest is roughly equal to the combined venture capital raised by Anthropic and Cohere in 2025, and it exceeds the total funding of all European AI labs combined. US labs now face a choice: they can either raise even larger rounds to maintain their compute advantage, or they can focus on software efficiency to close the gap. The market is already voting with capital: OpenAI is reportedly seeking a $15 billion round at a $300 billion valuation, while Anthropic is in talks for a $5 billion round after the model freeze. The arms race is accelerating, and DeepSeek’s record funding is the opening salvo in a new phase of the competition. The question is whether China’s strategy of massive state-backed investment can overcome the hardware gap, or whether US export controls will eventually starve Chinese labs of the compute they need to reach frontier capability. The answer will determine not just the future of AI, but the balance of technological power between the world’s two largest economies.