DeepSeek’s Chip Push Targets 50% Huawei-Dominated $50B AI Market

For supply chain executives, DeepSeek’s move is a case study in strategic insourcing. The AI startup, long dependent on external chip suppliers, is now designing its own silicon and engaging a web of foundry and memory partners — a pivot that could redefine procurement risk and supplier diversification in the race for AI infrastructure. With US export controls reshaping the global semiconductor map, every link in the chip supply chain faces disruption, and DeepSeek’s quiet hiring and private partnership talks signal an early, high-stakes reconfiguration.

DeepSeek, the Hangzhou-based AI startup that stunned the world with its highly efficient language models, is now making a dramatic strategic pivot into semiconductor development. According to three sources familiar with the matter, the company is designing its own AI chip — focused exclusively on inference, the stage where trained models generate responses for users, rather than training new models. The effort, which began about a year ago, would mark a major departure for a firm celebrated for algorithm breakthroughs over hardware commercialization, and it directly challenges the incumbents on which it has long depended: Nvidia and Huawei.

“

A US export ban on Nvidia's most advanced processors has effectively handed Huawei roughly half of the $50 billion domestic AI chip market, with DeepSeek as a prominent customer.

The move comes amid a transforming landscape for AI chips in China. A US export ban on Nvidia's most advanced processors has effectively handed Huawei roughly half of the $50 billion domestic AI chip market, with DeepSeek as a prominent customer. Yet Huawei's dominance is already under siege: Alibaba and Baidu are developing their own AI silicon, fragmenting a market that was once nearly monolithic. DeepSeek entering this race, even at an early stage, suggests the company sees chip sovereignty as the next frontier — a way to insulate its model development from geopolitical shocks and supply bottlenecks. The chip is inference-focused, which aligns with DeepSeek's business model: its models are already trained and widely used; now the cost and speed of serving them become paramount.

From a supply-chain perspective, the quiet ramp-up is telling. DeepSeek has begun privately hiring chip-design engineers without public job postings, indicating a deliberate, low-profile strategy likely aimed at minimizing scrutiny from Washington and Silicon Valley competitors. More significantly, the company is actively reaching out to external partners — chip-design houses, foundries, and memory companies. This points to a fabless model: DeepSeek will design but outsource fabrication, potentially tapping China's increasingly capable foundry ecosystem, possibly including SMIC or others, though no specific partners were named. The need for memory suppliers underscores that inference chips require high-bandwidth memory (HBM), a segment also constrained by export controls, adding another layer of supply-chain complexity.

The timing is critical. With the US continuing to tighten chip controls — in 2025, a group including the EU, Japan, and Netherlands intensified curbs on chipmaking gear — Chinese AI firms are under pressure to secure alternative supply lines. DeepSeek's inference chip could reduce its reliance on Nvidia's sanctioned A100/H100 equivalents and Huawei's Ascend chips, but it also introduces new dependencies: reliable foundry access, advanced packaging, and HBM procurement. If the chip reaches production, it would reshape not just DeepSeek's cost structure but also the competitive dynamics for AI hardware in China, potentially lowering costs for the broader AI ecosystem. However, chip design and manufacturing are notoriously slow and capital-intensive; the project's early stage and private hiring suggest a multi-year path, during which Huawei, Nvidia, and domestic rivals will continue evolving their offerings.