DeepSeek's next-generation V4 artificial intelligence model remains unreleased after months of anticipation, with global tech experts watching closely to determine whether China has successfully built advanced AI without relying on restricted US semiconductor technology. The Chinese AI startup, which shocked US tech markets in early 2025 with its affordable R1 chatbot, now faces a critical test: can it train and deploy a state-of-the-art model using domestic Chinese chips instead of Nvidia's powerful processors that the US has banned from export to China .

DeepSeek's journey illustrates the real-world impact of US export controls on the global AI race. The company started in 2023 as a side project of a hedge fund with access to powerful Nvidia processors. When it launched its R1 deep-reasoning chatbot in January 2025, the model performed comparably to US rivals while costing significantly less to run, prompting President Donald Trump to call it a "wake-up call" for American firms . Now, the question surrounding V4 is not just whether it will be impressive, but how it was built.

What Chips Is DeepSeek Actually Using for V4?

Tech news outlet The Information reported that DeepSeek's V4 can run on the latest chips made by China's Huawei, marking a potential milestone in China's effort to reduce dependence on US technology. The report cited five people with direct knowledge of large orders for Huawei chips made in preparation for the DeepSeek launch by major Chinese tech companies including Alibaba, ByteDance, and Tencent . If accurate, this shift would represent significant progress in China's semiconductor self-sufficiency strategy.

However, the transition from Nvidia to domestic chips is not straightforward. According to AI analysts, moving to Huawei silicon would require "substantial re-engineering" of DeepSeek's models.

"That transition can slow development cycles and introduce performance trade-offs, especially for V4, a model expected to be state-of-the-art," explained Wei Sun, Principal AI Analyst at Counterpoint Research.

Wei Sun, Principal AI Analyst at Counterpoint Research

This engineering challenge may explain why V4 has not yet launched despite weeks of industry anticipation.

The stakes of this technological shift extend beyond one company.

"If they have successfully trained V4 entirely on Huawei silicon, it signals a material shift in the geopolitical tech landscape," noted Stephen Wu, founder of the Carthage Capital fund.

Stephen Wu, Founder of Carthage Capital

Such a development would demonstrate that US export restrictions, while slowing Chinese AI progress, have not stopped it entirely.

Are There Allegations of Smuggling Nvidia Chips to China?

Beyond the official narrative about Huawei chips, some reports allege that DeepSeek may have circumvented US export controls through less transparent means. According to allegations, DeepSeek trained V4 using thousands of Nvidia's top-end Blackwell chips that were dismantled in third countries and smuggled to China . These claims highlight the cat-and-mouse game between US regulators and companies seeking to access restricted technology.

Nvidia has publicly disputed these smuggling allegations. The company told The Information that it had not seen evidence of such activity and that "such smuggling seems farfetched" . However, the very existence of these allegations underscores the difficulty of enforcing export controls in a globalized tech supply chain where components can be disassembled, routed through intermediaries, and reassembled elsewhere.

Training advanced AI models requires enormous amounts of computing power. The computational demands for training are far greater than for inference, which is the process of running a model to generate responses to user queries. This distinction matters because it affects which chips are most critical to restrict and where enforcement efforts should focus.

How to Understand China's AI Chip Independence Strategy

• Domestic Chip Development: Chinese companies like Huawei are racing to develop processors that can rival Nvidia's capabilities, reducing reliance on US exports and creating alternatives for training large language models.
• Multi-Company Coordination: Tech giants including Alibaba, ByteDance, and Tencent are reportedly placing large orders for Huawei chips, suggesting a coordinated national strategy to build AI infrastructure on domestic silicon.
• Open-Source Model Distribution: By releasing AI models as open-source software, Chinese companies can distribute their technology globally without needing to export physical chips, circumventing some export control restrictions.
• Regional Market Expansion: DeepSeek's customizable AI tools have been widely adopted in China and are also popular in emerging markets such as Southeast Asia and the Middle East, building a user base independent of US markets.

Other Chinese AI companies are pursuing similar strategies. Zhipu, another Chinese AI startup, unveiled an image generator in January that it said had been entirely trained on Huawei chips . Additionally, Alibaba announced this week that it would open a new data center for AI training and inference in southern China, powered by 10,000 of its own chips and operated by China Telecom . These parallel efforts suggest a broader Chinese strategy to build AI capabilities without dependence on US technology.

The broader implications of DeepSeek V4's eventual release will extend far beyond one company's product roadmap.

"It's important to know because at one level, it is a signal of China's AI self-sufficiency trajectory," stated Wei Sun.

Wei Sun, Principal AI Analyst at Counterpoint Research

The model's performance, architecture, and the chips used to train it will provide concrete evidence of how effective US export controls have been and how quickly China can develop alternatives.

Stephen Wu predicted that when V4 finally launches, it could again shake US tech valuations. "I expect the upcoming DeepSeek V4 release will not just be a software update; it will be a highly capable, open-source model that handles massive context windows at a fraction of the cost," he said . A multimodal model capable of generating text, images, and video at low cost could disrupt markets dominated by US companies and accelerate the adoption of Chinese AI tools in emerging markets.

Stephen Wu

The ongoing wait for DeepSeek V4 reflects deeper tensions in the global AI race. US export controls aim to preserve American technological leadership by restricting access to the most powerful chips. However, these restrictions also create incentives for other countries to develop alternatives, potentially accelerating a fragmentation of the global AI ecosystem into competing regional systems. The question of which chips power V4 will help determine whether those restrictions are working as intended or whether they are simply reshaping the competitive landscape without stopping Chinese progress.