Summary
A recent report said DeepSeek has begun work on a purpose-built semiconductor for AI computing, targeting inference tasks rather than model training. The announcement coincided with a 1.6% drop in Nvidia shares in premarket trading by 06:59 ET (10:59 GMT).
Market reaction
Nvidia stock slid 1.6% in premarket activity by 06:59 ET (10:59 GMT) after publication of the report describing DeepSeek's chip plans. The move in Nvidia shares followed news that DeepSeek is pursuing its own silicon, a shift that could alter the startup's dependence on external chip suppliers.
Technical focus: inference, not training
The semiconductors under development at DeepSeek are intended for inference work - the stage when an already-trained AI model generates answers for users - rather than for training models from scratch. That product focus aligns the project with the fastest-growing segment of AI compute demand as workloads increasingly move from building models to executing them for end users. Inference chips can often be optimized to cost less and consume less power than general-purpose graphics processors.
Project timeline and sourcing
Although the initiative remains in its early phases, the work has been underway for roughly a year. DeepSeek has been holding talks with third-party chip-design firms, foundries and memory suppliers as part of the effort. The company has also stepped up hiring of chip-design engineers in recent months, conducting recruitment quietly rather than by posting openings on public job boards, according to the report.
Strategic implications
Developing an in-house chip would represent a meaningful strategic change for DeepSeek. The startup has been promoted within China as a national AI success story since it drew swift global attention last year with the release of two AI models noted for their efficiency - releases that rattled technology hubs abroad. Creating proprietary silicon could reduce DeepSeek's reliance on major suppliers and place it among a growing group of AI developers that have moved to design their own hardware.
Barriers and constraints
Despite the potential benefits, the report underscores that producing a competitive AI chip generally requires years and substantial capital. DeepSeek also faces constraints specific to China: U.S. export restrictions prevent Chinese chip designers from using the most advanced overseas foundries, and separate U.S. curbs have limited access to high-bandwidth memory - a component the report described as essential to inference chips.
Outlook
At present, the project is nascent and faces both technical and geopolitical hurdles. If DeepSeek continues advancing its chip program, it may increase competitive pressure on existing hardware suppliers. The company’s quiet hiring and supplier discussions suggest a deliberate, early-stage push into chip design rather than an announced, near-term commercial rollout.