DeepSeek has upended AI industry conventions by denying Nvidia and AMD early access to its upcoming V4 flagship model while granting Huawei and Chinese chipmakers weeks to optimize their hardware. This unprecedented move signals China’s determination to build AI infrastructure independent of American technology.
Breaking Industry Norms
AI developers typically share pre-release versions of major models with all leading chipmakers—regardless of nationality—to ensure software runs efficiently on widely used hardware. According to Reuters, DeepSeek previously worked closely with Nvidia’s technical staff for optimization.
Not anymore. For the highly anticipated V4 model expected around Lunar New Year, DeepSeek provided zero access to U.S. chipmakers while giving Huawei, Cambricon, and other domestic suppliers a multi-week head start.
This departure from standard practice allows Chinese chipmakers to fine-tune their processors for V4’s architecture before the model’s public release, potentially giving them a significant performance advantage when V4 launches.
What We Know About DeepSeek V4
According to multiple reports, DeepSeek V4 will be the company’s first native multimodal model, capable of generating pictures, video, and text. The Financial Times reports V4 is optimized primarily for coding and long-context software engineering tasks, with internal tests suggesting it could outperform ChatGPT and Claude on these workloads.
The model is expected to feature a trillion-parameter architecture with a one-million-token context window—architectural capabilities that demand tight hardware-software integration to run efficiently. By granting early access to Huawei and other Chinese chipmakers, DeepSeek ensures its flagship model runs optimally on domestic silicon.
Export Control Allegations
The timing is particularly charged. A senior Trump administration official told Reuters that DeepSeek’s latest model was trained using Nvidia’s most advanced Blackwell chip on a cluster located in mainland China—a potential violation of U.S. export controls.
The official further alleged that DeepSeek may attempt to remove technical indicators revealing its use of American chips, planning to publicly claim Huawei hardware was used instead. If true, this would represent an attempt to circumvent export restrictions while obscuring the actual technology dependencies.
Nvidia just secured limited U.S. government approval to export H200 chips to China, but zero units have actually shipped as of late February. The company explicitly excluded China data center revenue from its Q1 outlook.
Strategic Implications
PC Gamer notes this marks the first time a major Chinese AI lab has deliberately locked American chipmakers out of its pre-release pipeline. Industry analysts see it as part of a broader Chinese government strategy to disadvantage U.S. hardware and software in China.
Ben Bajarin, CEO of research firm Creative Strategies, downplayed immediate commercial impact: “Most enterprises are not running DeepSeek, which serves as a benchmarking model more than anything else.” He added that new AI coding tools are compressing optimization timelines “from months to weeks.”
However, Bajarin acknowledged the move likely reflects deliberate Chinese government policy “to try to keep US hardware and models disadvantaged” within China’s AI ecosystem.
Building a Parallel Software Stack
The strategic calculus extends beyond one model release. By optimizing V4 for Huawei’s Ascend chips and other domestic hardware, DeepSeek is accelerating development of a parallel AI software ecosystem that reduces long-term dependence on U.S. technology.
China’s data center infrastructure is increasingly designed to run on domestically optimized hardware and software stacks. If DeepSeek V4 performs well on Huawei chips, it validates the viability of Chinese AI infrastructure independent of Nvidia.
According to Business Standard, this approach mirrors what U.S. tech companies experienced with China’s semiconductor development—initial struggles followed by rapid improvement through focused investment and optimization.
Huawei’s AI Chip Challenges
The Financial Times previously reported that DeepSeek initially attempted to train its R2 model using Huawei’s Ascend AI chips but encountered repeated failures and performance issues. The problems stemmed from stability concerns, slow chip-to-chip interconnect speeds, and immature software tooling.
Ultimately, DeepSeek reverted to Nvidia hardware for training while relegating Huawei chips to inference tasks—running trained models rather than training them from scratch. The V4 optimization partnership suggests DeepSeek is working to make Huawei chips viable for more demanding workloads.
Open Source Dominance
DeepSeek’s influence extends far beyond China. According to reports, DeepSeek models have been downloaded over 75 million times on Hugging Face since the company emerged in January 2025. Among models released in the past year, downloads for Chinese models have surpassed those from any other country on the platform.
This open-source dominance helps explain U.S. government concerns. Chinese AI labs are competing directly with American companies like OpenAI, Anthropic, and Google while distributing their models freely, potentially undercutting Western commercial efforts.
What’s Next
DeepSeek V4 is expected to launch imminently, likely within days. The model represents the biggest challenge yet from open-weight models to proprietary systems, with potential to reshape competitive dynamics in AI development.
Whether V4 delivers on its ambitious performance claims—particularly when running on Chinese hardware—will signal how quickly China’s domestic AI ecosystem is maturing. If Huawei’s Ascend chips can run V4 efficiently, it validates China’s multi-year investment in indigenous AI infrastructure.
For Nvidia and AMD, the immediate commercial impact may be limited. But the precedent concerns them: if more Chinese AI labs follow DeepSeek’s lead, American chipmakers could find themselves increasingly marginalized in the world’s largest tech market.
The U.S.-China AI competition is fragmenting into parallel ecosystems with different hardware, different software, and different optimization priorities. DeepSeek V4’s selective access policy is one more step in that divergence—a signal that the era of globally integrated AI development is giving way to something more fractured and geopolitically charged.
Read more tech related articles here.
Leave a Reply