DeepSeek's latest AI model marks a turning point in the US-China technology competition, not because it outperforms American models, but because it's learning to work without American chips. On April 24, the Chinese startup launched its v4 model optimized for Huawei's Ascend processors, a milestone that has alarmed Silicon Valley executives and prompted warnings from the highest levels of US tech leadership.

What Makes DeepSeek's Hardware Partnership So Threatening?

The collaboration between DeepSeek and Huawei represents something more significant than raw AI performance. DeepSeek announced that v4 was designed to use Huawei's chips and its software platform called CANN, marking the first time the startup has officially announced such a deep software-hardware partnership with the Chinese telecommunications giant. This integrated approach matters because it demonstrates how Chinese AI labs are learning to optimize models for domestic hardware, reducing their dependence on American technology.

Nvidia CEO Jensen Huang captured the stakes bluntly. In an April 15 interview, he stated: "The day that DeepSeek comes out on Huawei first, that is a horrible outcome for our nation (the US)." Huang has actively lobbied Washington to allow less powerful Nvidia chips to be sold to China, arguing that continued reliance on American hardware is preferable to complete decoupling.

Jensen Huang

The concern isn't that China has suddenly leapfrogged American AI capabilities. According to benchmark testing, DeepSeek's open-source v4 Pro ranked just behind US closed-source models such as OpenAI's GPT-5.5, Anthropic's Claude Opus, and Google's Gemini 3.1 Pro, placing it roughly seven months behind the frontier. Rather, the threat is architectural: China is building an alternative AI ecosystem that could eventually operate independently of American technology.

How Is China Building Its Own AI Stack?

An AI "stack" refers to the complete set of technologies and tools needed to build, maintain, and use AI models. Traditionally, this stack has been dominated by American companies. Nvidia provides the chips; CUDA, Nvidia's software platform, optimizes those chips for AI workloads; and American cloud providers handle deployment. China is now attempting to replace each layer with domestic alternatives.

Poe Zhao, founder of the Hello China Tech newsletter, explained the significance of this shift: "It shows how Chinese AI labs are learning to turn hardware constraints into a design problem, but it is still far from complete AI stack self-reliance." He noted that the compatibility with Huawei's Ascend 950 series of supercomputers is significant if it translates into stable deployment at scale, adding that "it does not mean Huawei chips can immediately replace Nvidia. The more important signal is model-hardware co-optimisation. Chinese model developers and hardware vendors are trying to build a domestic deployment path under chip constraints".

"It shows how Chinese AI labs are learning to turn hardware constraints into a design problem, but it is still far from complete AI stack self-reliance," said Poe Zhao.

Poe Zhao, Founder of Hello China Tech newsletter

The stakes for this competition are enormous. AI is now recognized as a critical battleground between China and the US. As Washington has sought to throttle China's development in the semiconductor supply chain with export controls, Beijing has invested heavily in its own suppliers to address this perceived weakness.

Steps to Understanding the AI Hardware Competition

• Chip Performance Gap: Nvidia still leads significantly in raw chip performance and its CUDA ecosystem, which allows chips to be optimized for AI workloads. Huawei's CANN software is attempting to replicate this functionality but remains less mature.
• Training vs. Deployment: A Reuters report in February cited an unnamed US official as saying that DeepSeek's latest model was trained on Nvidia's Blackwell chips, which are banned in China, indicating that China still lacks the vast number of powerful domestically made chips needed for frontier model development.
• Distillation Concerns: American firms such as Anthropic and OpenAI have accused Chinese AI firms of using distillation, a technique that allows newer models to learn from the answers of more powerful models, raising questions about the independence of Chinese AI development.

Marina Zhang, an associate professor specializing in technology and geopolitics at the University of Technology Sydney, offered a more nuanced assessment. She stated that "DeepSeek v4 does not prove that Chinese AI models have overtaken the best US closed models across the board. But it does narrow the gap and reinforces China's different competitive logic: near-frontier performance at much lower cost, with stronger open-source diffusion and domestic hardware adaptation".

She

"DeepSeek v4 does not prove that Chinese AI models have overtaken the best US closed models across the board. But it does narrow the gap and reinforces China's different competitive logic: near-frontier performance at much lower cost, with stronger open-source diffusion and domestic hardware adaptation," explained Marina Zhang.

Marina Zhang, Associate Professor at University of Technology Sydney

Chris McGuire, a senior fellow for China and emerging technologies at the Council on Foreign Relations, characterized the release as "largely a status quo release," noting that "US models still lead by about seven months and leading Chinese models remain dependent on US tech." He added that "like all other leading Chinese models, v4 was trained using US chips and on data illicitly distilled from frontier US models. If China fully lost access to US chips and models, not to mention US and allied chipmaking tools, DeepSeek and others would likely fall much farther behind".

Yet Zhang cautioned against dismissing China's progress as merely derivative. She noted that while distillation and access to restricted chips may have played some role in parts of China's AI ecosystem, "DeepSeek's progress also reflects real engineering capability." The key insight is that China can increasingly support competitive AI deployment for many commercial and industrial uses, even if a complete break from the US AI stack remains years away.

Yet Zhang

The most likely outcome, according to Zhang, is not clean decoupling but rather "two increasingly parallel and partially interoperable ecosystems." This means that while the US will likely maintain its lead in frontier AI development, China will build sufficient domestic capability to serve its own market and potentially export AI services to countries outside the US sphere of influence.

For Nvidia and American AI companies, the challenge is not that they face imminent displacement but that they face a future where their technology is no longer the only option. DeepSeek's v4 release, despite its modest performance gains, signals that this future is arriving faster than many in Silicon Valley expected.