- What Exactly Are the Giants Buying?
- Why Have Chinese Technologies Become the “Best Choice”?
- Technological Pragmatism Outweighs Sanctions
- Hide-And-Seek: How Sanctions Are Bypassed
- Conclusion
While headlines are filled with stories about trade wars and export bans on chips to China, the opposite process is quietly unfolding behind the scenes in Silicon Valley. The biggest players—from Google to Elon Musk’s xAI—are increasingly turning to Chinese suppliers for critical infrastructure components.
It turns out that building the “most powerful supercomputer in the world” is now practically impossible without China’s participation.
What Exactly Are the Giants Buying?
Contrary to popular belief, AI infrastructure is not just Nvidia GPUs. It consists of massive data centers where hardware must be cooled, powered, and able to exchange data at near-cosmic speeds. This is precisely where China has become indispensable.
1. Liquid cooling systems
Modern chips such as Nvidia’s Blackwell consume more than 1 kW of power each, effectively turning server racks into hot plates. Chinese companies (for example, Asetek and Huawei divisions) offer some of the most advanced direct-to-chip liquid cooling systems. They cost 30–40% less than Western equivalents while delivering higher energy efficiency.
2. Optical components and photonics
Connecting thousands of GPUs into a single AI cluster requires high-speed optical modules (800G and 1.6T). Chinese market leaders like Innolight control a significant share of these supplies. Without them, internal cluster bandwidth drops sharply and AI training can drag on for months.
3. Rare earths and power electronics
China dominates the supply of critical materials such as gallium, germanium, and antimony. These are essential for power transistors and energy converters used in the power supplies of Google and Tesla servers.
Why Have Chinese Technologies Become the “Best Choice”?
Price is only part of the story. By 2025–2026, three key factors became decisive.
1. Scalability
Chinese factories can deliver volumes that no other country can match. When Musk needs to assemble a 100,000-chip cluster within months, only Chinese suppliers can ship miles of fiber optic cable on time.
2. Energy focus
China is investing heavily in data center power infrastructure. Its smart grid solutions help U.S. companies optimize enormous electricity bills and stabilize energy consumption at hyperscale facilities.
3. Open-source culture
The success of the DeepSeek-V3 model in early 2025 shocked Silicon Valley. It proved that Chinese engineers can train competitive models at a fraction of the cost through algorithmic efficiency—significantly cheaper than OpenAI or Google.
Technological Pragmatism Outweighs Sanctions
Western hardware vendors are facing a “perfect storm.” Industry leaders—Samsung, SK hynix, and Micron—have prioritized AI over consumer markets, shifting production lines toward high-margin HBM memory for accelerators, where profits are five times higher.
As a result, shortages of conventional DDR4/DDR5 memory have pushed prices up by 60–150%, prompting major American brands to test chips from China’s ChangXin Memory Technologies (CXMT) for the first time.
Another example is power electronics. AI infrastructure requires enormous power capacity, and China’s Innoscience has stepped into the spotlight. The company has already become Nvidia’s only Chinese supplier for its new 800V architecture. Google has followed suit, integrating Innoscience components into its data center power modules. The advantage is straightforward: Chinese suppliers offer a full voltage range (15–1200V), at lower prices, and—most importantly—in the volumes available right now. Western alternatives simply cannot keep up with the scaling speed of Google’s cloud farms.
Elon Musk, who publicly criticizes dependence on China, has quietly deepened partnerships with Chinese tech leaders. Reports suggest that teams from SpaceX and Tesla have visited major Chinese solar manufacturers such as JinkoSolar, GCL Group, TCL Zhonghuan, and Jingsheng Electromechanical. Rumors indicate SpaceX has already signed contracts for equipment for HJT production lines.
In late January, Musk announced plans to build 100–200 GW of solar capacity within three years to power xAI data centers and Starlink’s space infrastructure. Orbital applications require lightweight, high-efficiency, radiation-resistant panels—exactly what HJT and perovskite tandem technologies provide. China controls more than 80% of global solar panel production. Building that capacity quickly in the U.S. is simply unrealistic: the equipment, expertise, and supply chains are not there.
Hide-And-Seek: How Sanctions Are Bypassed
The policy of “technological containment” increasingly resembles trying to stop a waterfall with bare hands. Companies use several strategies to avoid regulatory scrutiny (BIS).
Gray hubs
Components are purchased through intermediaries in the UAE, Vietnam, or Malaysia. After “final assembly” or simple relabeling, the products legally enter the U.S. as Malaysian goods.
Cloud loopholes
American firms rent computing capacity in data centers located in neutral countries that are fully equipped with Chinese hardware. Formally, they’re buying a service rather than equipment.
Subsidiary exemptions
Many Western companies operate manufacturing sites in China that are allowed to procure components for “internal use,” after which the resulting technologies and developments are transferred back to headquarters.
Conclusion
Dependence cuts both ways. While Washington tries to restrict China’s access to the “brains” (advanced chips), American tech giants have discovered that without China’s “body” (the infrastructure) those brains simply cannot function.
