DeepSeek's $7.4 Billion Bet on a Proprietary AI Chip: How China's AI Star Plans to Escape Nvidia and Huawei Dependency

Long-term: A proprietary inference chip. The most ambitious element is the development of DeepSeek's own custom chip, designed from the ground up for inference tasks . The company is working with external chip-design partners and has been expanding its internal chip design team, though the project remains at an early planning and development stage . A successful in-house chip would give DeepSeek strategic autonomy, insulating it from supply-chain disruptions and the whims of any single hardware vendor .

Medium-term: The shift to Huawei's Ascend. In the near-to-medium term, DeepSeek has already moved significant inference workloads onto Huawei's Ascend 950 chip series. The company's V4 model, a trillion-parameter mixture-of-experts model, was specifically optimized to run entirely on Huawei's Ascend 950PR hardware . DeepSeek granted early access to its V4 model to domestic suppliers like Huawei, while notably withholding it from U.S. chipmakers including Nvidia and AMD, a significant break from industry norms . This move demonstrated that DeepSeek could deploy production AI at scale without relying on U.S. hardware .

Near-term: Algorithmic efficiency. DeepSeek's models have always emphasized algorithmic and architectural efficiency, reducing the raw computing power necessary for both training and inference . Techniques like mixture-of-experts (MoE), selective activation, and transfer learning allow DeepSeek to deliver frontier-level performance on less-capable chips, making the transition to domestic hardware more feasible . This focus on inference efficiency is a key reason why the gap between Chinese-made AI processors and more robust U.S. chips has narrowed .

Why Independence Matters: The Export Control Squeeze

The driving force behind DeepSeek's hardware push is the cumulative effect of U.S. export controls, which have progressively cut China off from the world's most advanced AI chips. Washington has banned the export of chips as capable as Nvidia's A100 since October 2022, and the restrictions have since tightened to include the Blackwell series and specific licensing requirements for the H20 . A senior Trump administration official even alleged that DeepSeek trained its latest model on Nvidia's most advanced Blackwell chips in violation of U.S. export regulations .

DeepSeek has demonstrated it can achieve frontier-level AI performance despite these constraints through sheer engineering ingenuity . However, reliance on a single dominant supplier — even a domestic one like Huawei — remains a strategic risk. DeepSeek's earlier efforts to train models on Huawei Ascend hardware were reportedly met with persistent failures and delays, forcing the company to revert to Nvidia chips for training while using Huawei's only for inference . This rocky history underscores why DeepSeek wants to eventually control its own destiny.

The $7.4 Billion War Chest

DeepSeek's chip ambitions are backed by its first-ever external fundraising round, which closed in June 2026. The company raised over 50 billion yuan, approximately $7.4 billion, at a valuation exceeding $50 billion, making it China's most valuable AI startup . The round was led by founder Liang Wenfeng, who personally contributed 20 billion yuan (~$2.9 billion), while anchor investors included tech giant Tencent (10 billion yuan) and battery maker CATL (5 billion yuan) . Other participants included JD.com, NetEase, and IDG Capital .

The deal structure was unusual: investors placed their funds into a limited partnership managed by Liang, giving them no direct voting rights in DeepSeek and subjecting them to a five-year lockup . This structure allows Liang to retain strategic control even while raising massive outside capital.

The timing is crucial. Designing, taping out, and manufacturing a custom AI chip is a multi-year endeavor that typically costs hundreds of millions to well over a billion dollars. The $7.4 billion war chest provides the financial firepower to fund that effort while continuing to develop next-generation AI models and build out necessary data center infrastructure .

The Bottom Line

DeepSeek's chip strategy is a textbook example of defensive innovation under pressure. Faced with a hardware environment that grows more constrained by the quarter, the company is diversifying across three fronts: squeezing more performance out of existing hardware through algorithm improvements, shifting current workloads to a domestic partner (Huawei), and building for long-term independence with a custom chip design. The $7.4 billion fundraising round gives DeepSeek the financial runway to see this through, transforming its hardware strategy from a contingency plan into a potentially decisive competitive advantage.