When DeepSeek released its latest artificial intelligence model last month, the announcement included a technical detail that fundamentally alters the geopolitical landscape of the semiconductor industry. The Chinese startup confirmed that its advanced models are now fully compatible with Huawei’s Ascend line of AI processors. This development represents a critical milestone in Beijing’s long-standing quest for technological self-sufficiency and significantly weakens the United States’ leverage through its export control regime.
The integration of DeepSeek’s software with Huawei’s hardware is not merely a technical achievement; it is a strategic breakthrough. For years, the U.S. strategy of containment has relied on the assumption that Chinese AI developers could not remain globally competitive without access to Nvidia’s industry-standard chips and its ubiquitous CUDA software ecosystem. According to The New York Times, DeepSeek’s successful pivot to domestic hardware demonstrates that this assumption is no longer valid.
Breaking the CUDA Monopoly
Nvidia’s dominance in the AI sector rests on two pillars: the raw computational power of its graphics processing units (GPUs) and the entrenched position of its CUDA software platform. CUDA allows developers to write code that efficiently utilizes the parallel processing capabilities of Nvidia’s chips. Because the vast majority of AI research and development over the past decade has been conducted using CUDA, transitioning to a different hardware architecture requires a massive investment of time and engineering resources.
DeepSeek undertook this exact challenge. The company spent months rewriting its core code to function natively on Huawei’s Compute Architecture for Neural Networks (CANN), the domestic equivalent to CUDA. By achieving day-zero adaptation for its V4 models on Huawei’s newest Ascend chips, DeepSeek proved that the software moat surrounding American hardware can be breached. This technical feat provides a blueprint for other Chinese AI firms seeking to decouple from the U.S. supply chain.
(Related: DeepSeek V4 Goes Fully Open Source; Huawei Confirms Ascend Chip Compatibility in Separate Statement)
The Hardware Reality
The hardware enabling this transition is Huawei’s Ascend series, specifically the Ascend 910C and the newer Ascend 920. These chips are fabricated by Semiconductor Manufacturing International Corporation (SMIC) using older deep ultraviolet (DUV) lithography equipment, as U.S. export controls have blocked China’s access to the extreme ultraviolet (EUV) machines required for the most advanced manufacturing nodes.
Despite these manufacturing constraints, the Ascend chips are proving capable of handling the massive computational workloads required for training and running foundational AI models. While the Ascend 910C delivers roughly 60 percent of the inference performance of Nvidia’s H100, the sheer volume of chips Huawei is producing, targeting 1.6 million Ascend dies this year, allows Chinese developers to compensate for individual chip performance through massive parallelization. The fact that DeepSeek, which operates some of the most sophisticated models in China, finds this hardware sufficient is a testament to the viability of the domestic ecosystem.
(Related: Beijing Put Huawei’s Secret Chip Lab on National TV Two Days Before Trump Arrived)
Diminishing Returns of Export Controls
The successful pairing of DeepSeek’s models with Huawei’s chips highlights the diminishing returns of the U.S. export control strategy. The restrictions were designed to freeze China’s AI capabilities by denying access to the necessary hardware. Instead, the controls have acted as a powerful catalyst, forcing the consolidation of the Chinese tech sector around domestic champions.
As President Trump arrives in Beijing for a state summit, the U.S. delegation faces a fundamentally altered negotiating environment. The leverage derived from controlling the supply of advanced semiconductors is eroding. The Chinese government has actively encouraged this transition, directing state-owned enterprises and private firms alike to prioritize domestic hardware. The result is a parallel technology ecosystem that is increasingly insulated from American policy decisions. The DeepSeek-Huawei partnership is the clearest evidence yet that China’s drive for AI independence is not just a policy goal, but an operational reality.
The Broader Ecosystem Shift
DeepSeek’s pivot to Huawei hardware is not an isolated event; it is the most visible example of a broader ecosystem shift underway across China’s AI industry. Dozens of Chinese AI companies, from large state-owned enterprises to venture-backed startups, are actively rewriting their software stacks to achieve compatibility with domestic hardware. This coordinated transition is being actively encouraged by the Chinese government, which has made domestic hardware adoption a condition for certain state procurement contracts and research funding.
The transition is also being facilitated by a growing body of open-source tools and libraries that abstract the differences between Huawei’s CANN framework and Nvidia’s CUDA. Chinese academic institutions and research labs are contributing to this ecosystem, creating a virtuous cycle in which the more developers adopt domestic hardware, the more robust the supporting software infrastructure becomes. This community-driven development mirrors the early growth of the CUDA ecosystem in the United States, suggesting that China is replicating the playbook that made Nvidia dominant in the first place.
Implications for U.S. Policy
The DeepSeek-Huawei integration presents a fundamental challenge for U.S. policymakers. The export control strategy was predicated on the assumption that Chinese AI development was inextricably dependent on American hardware. That assumption has been invalidated. The U.S. now faces a strategic choice: continue tightening restrictions in the hope of slowing China’s progress, or pursue a more nuanced approach that preserves some commercial engagement while protecting the most sensitive technologies.
The evidence from DeepSeek’s success suggests that continued escalation may be counterproductive. Every new restriction provides additional motivation for Chinese companies to invest in domestic alternatives, accelerating the very self-sufficiency that the U.S. is trying to prevent. As the two leaders meet in Beijing, the DeepSeek-Huawei story serves as a powerful reminder that in the AI race, software ingenuity can overcome hardware constraints, and that the most effective long-term strategy may be to maintain the leading edge of innovation rather than deny access to existing technology.