DeepSeek V4 Drops Next Week — Trillion-Parameter Multimodal Model Excludes Nvidia From Early Access
DeepSeek is about to make waves again. The Chinese AI lab plans to release V4, its next-generation multimodal large language model, next week — and it's making a deliberate geopolitical statement by giving Huawei and Cambricon early access while excluding Nvidia and AMD entirely.
What V4 Brings
DeepSeek V4 is a massive upgrade from the R1 reasoning model that turned heads in January 2025. The new model features native support for image, video, and text generation — making it truly multimodal rather than text-only. The preview version, codenamed "sealion-lite," has entered closed internal testing with a context window expanded to 1 million tokens, enabling analysis of extremely long documents and large code libraries.
Reports suggest V4 may be a trillion-parameter scale model, which would put it in the same league as the largest models from OpenAI and Google — but built at a fraction of the cost, consistent with DeepSeek's reputation for efficiency.
The Nvidia Snub
In a break from industry standard practices, DeepSeek has prioritized providing early access to domestic Chinese chip suppliers including Huawei, without providing test versions to Nvidia and AMD. AI developers typically share pre-release versions of major models with chip manufacturers to ensure software runs efficiently on their hardware. DeepSeek previously worked closely with Nvidia technical personnel, making this exclusion a significant departure.
The move signals DeepSeek's commitment to reducing reliance on U.S.-made chips and strengthening China's domestic semiconductor ecosystem. By optimizing V4 specifically for Huawei and Cambricon hardware, DeepSeek could significantly boost demand for Chinese-made AI chips.
Strategic Timing
The release is timed ahead of China's annual "Two Sessions" parliamentary meetings beginning March 4 — a politically significant moment that will maximize attention on China's AI capabilities. DeepSeek plans to publish a brief technical note alongside V4, with a more comprehensive engineering report following about a month later.
The Bottom Line
DeepSeek's V4 isn't just a model release — it's a declaration of technological independence. By deliberately excluding Nvidia and AMD from early access and optimizing for Chinese chips, DeepSeek is accelerating the AI industry's split along geopolitical lines. The fact that they might achieve trillion-parameter scale on non-Nvidia hardware would be a remarkable engineering achievement — and a nightmare for U.S. chip export policy.
Source: Pandaily, Financial Times