DeepSeek Grants Huawei Early Access to V4 AI Model, Excluding Nvidia and AMD

DeepSeek Grants Huawei Early Access to V4 AI Model, Excluding Nvidia and AMD

In a move that disrupts long-standing industry protocols, Chinese artificial intelligence startup DeepSeek has provided Huawei with early access to its forthcoming DeepSeek-V4 model, while deliberately excluding U.S. semiconductor leaders Nvidia and AMD from pre-release collaboration. According to Reuters and The Information, this decision marks a deliberate pivot away from the traditional model of AI labs working hand-in-hand with GPU manufacturers to optimize performance on dominant hardware platforms.

Typically, leading AI developers such as OpenAI, Meta, and Google collaborate extensively with Nvidia’s engineering teams months before launching new models. These partnerships ensure that software is fine-tuned for Nvidia’s H100 and Blackwell architectures, maximizing inference speed, energy efficiency, and scalability. But DeepSeek’s decision to bypass this norm — and instead prioritize Huawei’s Ascend AI chips — suggests a broader geopolitical recalibration in the global AI supply chain.

Multiple sources confirm that Huawei received access to DeepSeek-V4 several weeks ahead of the model’s public release, which is expected to coincide with the post-Lunar New Year period. This window allowed Huawei’s engineers to adapt its Ascend 910B and next-generation AI accelerators to better support DeepSeek’s architecture, potentially improving training throughput and reducing latency. In contrast, Nvidia and AMD were not granted any pre-release documentation, benchmarks, or API previews — a stark departure from industry norms that have underpinned AI innovation for the past five years.

The implications extend beyond technical optimization. DeepSeek’s choice reflects China’s accelerating strategy to insulate its AI ecosystem from U.S. export controls and technological sanctions. Since 2022, U.S. restrictions have limited Huawei’s access to advanced semiconductors and AI development tools. Yet, by fostering domestic alliances — such as this one with DeepSeek, one of China’s most promising open-weight AI firms — Beijing is building a self-sustaining pipeline of AI software and hardware.

Analysts suggest this is not an isolated incident. According to The Information, similar partnerships are emerging between other Chinese AI labs and domestic chipmakers like Biren Technology and Moore Threads. These collaborations are designed to create a closed-loop ecosystem: Chinese AI models trained and optimized on Chinese hardware, deployed within Chinese cloud infrastructure, and shielded from Western influence.

For Nvidia and AMD, the exclusion represents both a strategic and commercial challenge. While their GPUs still dominate global AI training markets, the rise of alternative architectures in China could erode their long-term market share. If DeepSeek-V4 gains traction in Asia and among global users seeking non-U.S.-controlled AI, the demand for Ascend-based deployments may grow — particularly in regions wary of U.S. data sovereignty policies.

DeepSeek has not publicly commented on the matter. However, internal documents reviewed by Reuters indicate the company views this decision as a matter of national technological sovereignty. "We are building for the future of AI in China," one developer reportedly wrote in an internal memo. "The old model served the West. We are building our own."

Meanwhile, U.S. officials are monitoring the development closely. The Department of Commerce has not issued any formal statements, but insiders say internal briefings have begun evaluating whether to expand export controls to include AI software optimization tools. The move could further escalate the tech cold war, turning AI model access into a new front in the global semiconductor rivalry.

As the AI landscape fragments along geopolitical lines, DeepSeek’s decision may prove to be a watershed moment — not just for China’s tech ambitions, but for the future of global collaboration in artificial intelligence.