Let’s just call a spade a spade. When the US government decided to tighten the screws on exporting high-end AI chips to China, the writing was on the wall. The move was meant to be a technological checkmate, a decisive blow to slow down China’s burgeoning AI ambitions. But if you thought for a second that tech giants like Alibaba and ByteDance would simply pack up their algorithms and go home, you haven’t been paying attention to this industry for the last two decades.
This isn’t a story about submission; it’s a story about circumvention. The silent, frantic arms race for computational power has simply moved into the shadows. What we’re seeing now is a masterclass in AI chip restrictions evasion, a fascinating and, frankly, entirely predictable response to a policy that underestimated the sheer force of market incentives and national pride. The game hasn’t stopped; the playing field has just got a whole lot bigger.
The Great Tech Decoupling or Just a Detour?
The US strategy was, on the surface, quite logical. AI models, particularly the gigantic Large Language Models everyone is chasing, are ferociously hungry for processing power. The most powerful processors—the digital shovels in this AI gold rush—are designed and controlled by American firms, most notably NVIDIA. By restricting access to top-tier chips like the H100 and even its less powerful, China-specific cousins like the H20, Washington aimed to starve Chinese AI development of its most critical nutrient.
It seemed a neat and tidy solution. Control the chokepoint, control the industry. But this approach views the world as a static map, not the fluid, interconnected network it truly is. A restriction in one place simply creates a powerful vacuum elsewhere. And capital, much like water, always finds a way to flow around an obstacle.
The Offshore Shuffle: If You Can’t Bring the Chips to the Data, Bring the Data to the Chips
So, what do you do when the best tools are banned in your country? You don’t use them in your country. According to a recent report in the Financial Times, this is exactly the playbook being run by China’s largest technology firms. They are engaging in large-scale offshore model training, effectively renting time on powerful NVIDIA chips located in data centres across Southeast Asia.
Think of it like this: the US built a high wall around China to keep NVIDIA’s super-powered tools from getting in. Instead of trying to smash through the wall, companies like Alibaba and ByteDance have simply “air-dropped” their data and engineers into neutral territories where those tools are readily available. They are building their AI models abroad and then bringing the finished product back home. This manoeuvre highlights the emergence of a new, sprawling geopolitical AI infrastructure that is far more complex than simple national borders would suggest.
According to the report, which was also covered by Yahoo Finance, this trend accelerated significantly after the US moved to restrict sales of the H20 chip. It’s a classic case of the market adapting. The demand didn’t vanish; it just migrated.
– The Strategy: Chinese tech firms are leasing capacity from foreign-owned data centres in places like Singapore.
– The Execution: They send their data and AI training workloads to these offshore facilities.
– The Result: The models are trained on unrestricted, top-of-the-line NVIDIA hardware, completely sidestepping US export controls.
It’s a clever, if costly, workaround. But when you’re in a race to build the next generation of artificial intelligence, “costly” is a relative term. The real cost would be falling behind permanently.
Plan B: The Home-Grown Rebellion
This offshore strategy is a brilliant short-to-medium-term solution, but it’s not a permanent one. Relying on foreign data centres introduces new risks, from data security to the whims of other governments who might one day face pressure from the US. The real end-game for China is self-sufficiency.
This is where the second part of the strategy comes into play, a two-pronged internal effort focused on stockpiling and innovation. We’re seeing some firms, like the AI start-up DeepSeek, play a shrewder game. The Financial Times notes that DeepSeek was savvy enough to accumulate a large inventory of NVIDIA chips before the most stringent bans came into effect. This gives them a domestic hardware advantage and a runway to continue development without immediately needing to go abroad.
But stockpiles eventually run out. The truly strategic move is the pivot towards NVIDIA chip alternatives. And the name on everyone’s lips is Huawei. Once a pariah of the tech world thanks to US sanctions, Huawei has been forced into a corner, and it has come out fighting. The company is now seen as China’s national champion for developing high-end silicon.
The collaboration between firms like DeepSeek and Huawei is the most significant long-term development here. While Huawei’s current Ascend 910B chip is still considered a step behind NVIDIA’s best, it’s a viable competitor. By partnering, Chinese tech companies are not only getting access to domestically produced hardware but are also creating a crucial feedback loop, helping Huawei refine its technology with real-world AI workloads. This isn’t just about finding a substitute; it’s about building a genuine, resilient ecosystem from the ground up.
What Does the Future Look Like?
The US export controls have, ironically, achieved two things Washington likely did not intend. First, they’ve kick-started a massive, well-funded effort within China to build a world-class domestic semiconductor industry, a project that now has the urgency of a national security imperative. Sanctions have proven to be a more effective catalyst for innovation than any five-year plan from Beijing could ever be.
Second, they’ve globalised the AI infrastructure chess-board. The game is no longer just about who can design the best chips in California, but about who controls data centres in Singapore, who can lay subsea fibre-optic cables, and who can offer the most politically stable environment for training AI models.
This campaign of AI chip restrictions evasion is a powerful lesson in unintended consequences. The attempt to create a technological bottleneck has instead created a series of workarounds that may ultimately make China’s tech ecosystem more resilient and independent. The West’s lead in AI is far from guaranteed, and the strategies being deployed today in the data centres of Southeast Asia and the research labs of Shenzhen will define the technological landscape for years to come.
The question is no longer if China will catch up, but what the world will look like when it has two parallel, competing AI ecosystems. So, what do you think? Is this cat-and-mouse game a sustainable strategy for the US, or has it simply accelerated the very outcome it was designed to prevent?
