The AI Cold War: Will Banning Chips Really Hold China Back?

In January, Chinese AI startup DeepSeek sent shockwaves through Silicon Valley by unveiling a model that allegedly rivals those of US tech juggernauts Google and OpenAI - at a fraction of the price.

In the aftermath of the announcement, many have accused DeepSeek of lying about their use of banned semiconductors, with one player – ScaleAI’s billionaire CEO Alexandr Wang - even alleging that the company has secretly acquired as many as 50,000 NVIDIA H100’s, the premier NVIDIA AI chip and importantly, the chip banned from export to China by the US Government.

Source: X.com @atm_tradingnews

This ban, which the U.S. named the AI Diffusion Export Framework, brought in a sweeping set of rules aimed at restricting China's access to advanced AI computing resources, reflecting the belief that AI dominance will shape global power dynamics.

These controls were unprecedented in scale and complexity, dividing countries into three tiers based on access to AI chips, with strict caps on GPU purchases and a focus on U.S. interests.

Key points of the ban:

1. Tiered System: Tier 1 includes the U.S. and its close allies, Tier 3 encompasses arms-embargoed nations (e.g., China, Russia), and Tier 2 covers all others, with varying restrictions.
   

2. New Licensing Regime: Validated End Users (VEUs) are categorized as Universal (UVEUs, primarily U.S. hyperscalers like Microsoft and Google) or National (NVEUs, including some Tier 2 entities such as Australia’s data centre operator NextDC) and must meet stringent security and compliance standards.
   

3. Chip and Model Restrictions: Export controls apply not just to physical chips but also to large-scale AI model training and model weights.
   

4. Geopolitical Impact: The rules obviously favour U.S. hyperscalers (Microsoft, Google, Amazon) while challenging Tier 2 data centres, and effectively shunning Chinese firms.
   

5. Economic Effects: The restrictions ultimately reshape global AI deployment, with reduced capacity in Tier 2 countries (e.g., Malaysia, Brazil) and consolidation of AI power within Tier 1, particularly in the U.S. and its allies like us.

But Does It Work?

Like any good idea, the reality on the ground often differs greatly from what is theorised in the Oval Office.

Despite stringent U.S. export controls aimed at restricting China's access to advanced semiconductor technology, many believe China has successfully leveraged systemic loopholes, state support, and strategic workarounds to sustain and expand its domestic semiconductor capabilities. These methods include exploiting regulatory gaps, repurposing equipment, and developing indigenous alternatives.

In fact, there are numerous ways Chinese companies can circumvent these bans, lending credence to the idea that DeepSeek may indeed be concealing its true capabilities.

Potential Loopholes:

1. Re-Exporting Through Third Countries: Chinese firms can buy restricted AI chips through friendly, unrestricted countries like Malaysia, UAE, or Singapore, which do not have strict enforcement mechanisms. These chips can then be quietly resold or shipped to China through intermediaries.

   
   

2. Cloud-Based Access to AI Chips: Instead of purchasing physical chips, Chinese companies can rent access to AI compute power from cloud providers based in the U.S. or allied countries. Since cloud computing services are harder to regulate, companies like ByteDance can use offshore data centres to train AI models using U.S.-made GPUs.

   
   

3. Using Shell Companies: Chinese entities could create new companies outside China, registered them in Hong Kong, Singapore, or the Cayman Islands. These companies would appear as independent foreign firms, purchasing AI chips legally and then ship them back to China.

   
   

4. Importing AI Chips as "Consumer Electronics": Some AI-capable GPUs, especially high-end gaming graphics cards (like Nvidia's recently released RTX 5090), are not subject to the same restrictions as enterprise AI chips. Chinese buyers can import gaming GPUs in bulk, then modify them for AI workloads.

   
   

5. Partnering With Universities and Research Institutions: Chinese companies collaborate with foreign universities that legally acquire AI chips for research purposes. Once inside these institutions, the chips can be shared or unofficially transferred to Chinese entities through partnerships, joint ventures, or research collaborations.

   
   

These loopholes allow Chinese firms to work around U.S. restrictions and maintain access to advanced AI chips despite the tightening of export controls.

So What Can Be Done?

Enforcing AI chip export restrictions is incredibly challenging in practice due to the sheer number of loopholes, the complexity of global supply chains, and the difficulty in tracking end-use once chips leave U.S. shores. While the laws are strict on paper, enforcement agencies struggle to monitor every transaction, especially when chips are re-exported using the loopholes above, making it nearly impossible to prevent foreign entities from leveraging U.S. compute power remotely.

Even when violations are identified, holding companies accountable requires international cooperation, and not all governments are willing to comply with U.S. sanctions, particularly in regions where Chinese investment is strong.

With China’s well-documented history of circumventing tech sanctions through state-backed smuggling networks, weak oversight, and strategic acquisitions, the enforcement of these laws becomes a constant game of cat and mouse, requiring rapid regulatory updates and global coordination to remain effective.

Bottom Line

Did DeepSeek lie about their compute? Maybe. Maybe not. Does it even matter?

What we do know for sure is that the AI Cold War is only just beginning.