The US has paused a major move to blacklist China's DeepSeek and over 100 other companies, even though officials already approved them for trade restrictions. This delay signals how tangled the US-China AI competition has become, moving far beyond simple chip export controls into software ecosystems, national security concerns, and the risk of escalating geopolitical tensions.

DeepSeek shocked the tech world in January 2025 by releasing low-cost AI models that challenged the assumption that only American giants like OpenAI and Google could build cutting-edge artificial intelligence. According to Reuters reporting cited in recent analysis, US officials later accused DeepSeek of supporting China's military and attempting to access restricted US technology through shell companies. An interagency US committee approved DeepSeek, Chinese memory chipmaker CXMT, and more than 100 other firms for addition to the Commerce Department's Entity List last year, but the Trump administration has not yet published those additions.

The Entity List matters because it fundamentally changes what US companies can sell. Firms on the list typically need a US government license before they can receive certain American goods, software, or technology. This doesn't automatically ban every interaction, but it makes deals significantly harder and more expensive to navigate.

Why Is the US Hesitating on This Blacklist?

The delay appears to reflect Washington's attempt to avoid escalating US-China tech tensions at a moment when relations are already strained. Reuters frames the pause as part of a wider strategy under the Trump administration, which may want to avoid making US-China relations worse. The politics now look harder than the paperwork. It's a classic dilemma: the US can move aggressively and risk a bigger clash with Beijing, or it can slow down and face criticism that sensitive US technology may still reach firms it already considers dangerous.

This hesitation matters because DeepSeek sits at the center of three overlapping concerns for US policymakers:

• AI Capability: DeepSeek's models are competitive on price and performance, forcing US companies to rethink their business models and proving that frontier AI development isn't exclusively an American domain.
• Chip Access: DeepSeek's rise depends on access to advanced semiconductors and computing infrastructure, which is why restrictions on companies like CXMT matter so much to US strategy.
• National Security: US officials worry that AI tools could support Chinese military intelligence or defense systems, and that restricted technology might move through third-party routes or shell companies.

US officials have already warned allies about alleged AI model theft involving Chinese firms, including DeepSeek. In April, the US State Department instructed diplomats to raise concerns about "distillation," a technique where one AI model uses the outputs of another model to train a cheaper system. OpenAI has accused DeepSeek of trying to copy its models, while DeepSeek has denied intentional misuse of synthetic data.

How Is China Building an Alternative to Nvidia's Dominance?

The real competition extends far beyond individual companies. China is systematically building an alternative software and hardware ecosystem to challenge Nvidia's near-total dominance of AI infrastructure. Nvidia controls roughly half the world's installed AI chip stock and as much as two-thirds of installed computing capacity, giving it extraordinary pricing power with profit margins between 70 and 80 percent.

Nvidia's durability rests not just on superior chips but on CUDA, its proprietary software platform that has become the de-facto standard for AI development. CUDA creates powerful network effects: more developers write code optimized for CUDA, which attracts more hardware makers to support it, which attracts more developers. This self-reinforcing cycle makes switching costs extremely high for anyone trying to use alternative hardware.

China's Huawei is building a direct parallel strategy, layer by layer:

• Hardware Alternative: Huawei's Ascend AI processors remain technologically inferior to Nvidia's best chips, but demand within China is strong because of subsidies, cheap energy, procurement preferences, and US export controls that create a captive market.
• Software Ecosystem: Huawei developed CANN (Compute Architecture for Neural Networks) and MindSpore, an open-source machine learning framework optimized for Ascend processors. In August 2025, Huawei open-sourced the CANN toolkit, directly striking at CUDA's proprietary model.
• Developer Bridge: Huawei created torch_npu, a backend plugin that lets standard PyTorch code run on Ascend processors. Since most AI developers write in PyTorch rather than directly in CUDA, this dramatically lowers the single biggest switching cost for developers considering alternatives to Nvidia.

Notably, Huawei collaborates with DeepSeek, one of China's leading AI labs. DeepSeek's models have been engineered for compatibility with both Nvidia and Huawei processors. This is strategically clever: it lowers barriers for developers already familiar with Nvidia ecosystems to try Ascend hardware, while widening the addressable market for Huawei chips. DeepSeek is rapidly becoming one of the most popular AI models globally, and unlike some earlier closed efforts by US labs, leveraging popular open-weight models as a bridge appears to be accelerating Huawei's software ecosystem momentum.

The early evidence suggests that the software gap, like the hardware gap, is no longer fixed or unbridgeable. An open-sourced toolkit with a state-backed contributor pipeline, falling switching costs through PyTorch compatibility, flagship open-weight models running on Ascend, and a protected domestic market large enough to sustain the ecosystem through its immature phase represent real progress that didn't exist in meaningful form two years ago.

What Do Businesses Outside the US and China Need to Know?

This geopolitical competition has immediate practical implications for companies far beyond Washington and Beijing. Many businesses now use AI tools, cloud platforms, APIs, and imported hardware without thinking much about geopolitical risk. A company in South Africa, India, or Europe may not buy chips directly from DeepSeek or CXMT, but it may still depend on global AI services, cloud vendors, developer tools, and devices shaped by these restrictions.

The key questions businesses should ask themselves include:

• Service Continuity: Which AI tools can your business safely use, and could a service change overnight because of US export rules or sanctions?
• Data Location: Where does your data go when you use cloud-based AI services, and what happens if geopolitical tensions affect access?
• Procurement Risk: Do your procurement teams check sanctions and export-control exposure before adopting new AI platforms or hardware?
• Cost vs. Risk: The cheapest API or fastest tool may carry hidden legal or reputational risk that outweighs the savings.

The DeepSeek blacklist delay becomes bigger than one company because it reveals the fundamental tension in US strategy. If Washington names DeepSeek and CXMT, US suppliers will face tighter limits on what they can sell to Chinese firms. If it keeps waiting, critics will argue that the US is weakening its own controls. Either way, the AI race now runs through export law, diplomacy, and national security.

What Does This Mean for the Broader AI Competition?

Nobel Prize-winning economist Daron Acemoglu has argued that the current AI discourse often misses the real stakes. He estimates that AI will deliver only about 0.55 percent in total factor productivity gains over the next decade, far below Wall Street's euphoric projections, and that only about 5 percent of tasks will be profitably automated in the near term. More importantly, he stressed that the US needs genuine global governance for AI, including cooperation with China, which he says is ahead of the US in integrating AI into manufacturing, robotics, and commerce, even as it lags on large language models.

"I think that US-China collaboration would be so beneficial. We need global governance for AI. We also need the AI race not to get out of control," Acemoglu stated.

Daron Acemoglu, MIT economist and 2024 Nobel Prize winner

The hesitation over DeepSeek reflects a deeper strategic question: whether the US can maintain its AI dominance through export controls and regulatory pressure, or whether it needs to engage in the kind of international cooperation that Acemoglu advocates. The delay suggests that policymakers recognize the limits of a purely restrictive approach, but have not yet settled on an alternative strategy.

For now, the uncertainty itself is the story. DeepSeek remains off the Entity List, Chinese companies continue developing alternatives to US technology, and businesses worldwide must navigate an AI landscape where geopolitical risk is becoming as important as technical capability.