Microsoft, AWS and Google Rethink China Supply Chains

Microsoft, Amazon Web Services and Google are now working on even more predictable platforms as part of continuing efforts to engineer out the risk of entanglements with China that they have long wanted to avoid. The moves are wide-ranging and happening fast, according to industry sources interviewed by Nikkei Asia, as companies react to geopolitical risk, trade restrictions and an unforgiving clock as they race to build AI infrastructure that works at global scale.

Why the Big Three Are Moving Supply Chains Now

Concentration risk is now a boardroom issue. “Beyond high-performance computing, export controls from the U.S. Commerce Department on advanced chips and networking components have significantly complicated sourcing,” adds Burney. Meanwhile, procurement teams recall factory shutdowns and logistics bottlenecks of recent years with the benefit of hindsight—single-country supply chain exposure exacerbated disruption. The priority is now resilience: more suppliers in more places, even if the near-term bill is higher.

There is also a compliance aspect. Regulations around critical infrastructure and national security—such as restrictions for some communications and surveillance components—drive cloud providers to inspect sub-tier parts that might have been overlooked. That includes power supplies, cables, printed circuit boards and chassis metalwork that have historically come from clusters in China.

Microsoft’s 80 Percent Goal for Non-China Sourcing

Microsoft is aiming for a major shift in reshoring and a “China-plus-many” strategy by mid-decade. The company has asked partners to ready capacity outside China so that up to 80% of components for its Surface PCs, tablets, and data center gear come from elsewhere. The directive covers assemblies for future generations of server platforms, and even part of Xbox production.

In practical terms, that has led original design manufacturers — companies with names such as Quanta, Wiwynn, Foxconn, Pegatron and Inventec — to expedite expansions in places like Thailand, Vietnam, Mexico and Taiwan. Some have already announced new server campuses or begun construction on them, with Thailand one of the preferred locations due to breeding-ground industrial parks that cater specifically to IT hardware and easy access to components that are coalesced in existing ecosystems near those places.

AWS and Google Change Server Sourcing Strategies

With AWS, the emphasis is on AI data centers’ backbone: high-layer-count PCBs, backplanes and power delivery networks. AWS has examined reducing orders from a long-established Chinese supplier of PCBs, Key Akao Electronics (Dongguan), and is qualifying alternatives in other Asian markets, sources tell Nikkei Asia. Requalification is strenuous — with the thermal and signal integrity requirements driven by GPU-dense racks there’s not much room at all for material or process variation.

Google, for its part, is reportedly pushing suppliers to jack up server production in Thailand (where it’s wrangled numerous partners for components and final assembly).

The shift dovetails with prior moves to spread consumer hardware production more broadly among geographies, such as relocating some of the assembly of its Pixel phones to Vietnam and India, and marks a deeper bet on Southeast Asia for supplying cloud infrastructure.

New Hubs in Asia and the Americas Emerge

Near-term winners are Vietnam and Thailand. Company filings and local investment boards signal new server lines, metal chassis shops and cable-harness suppliers are clustering around Bangkok and the Eastern Economic Corridor while board assembly and enclosure-making remain concentrated in northern Vietnam. In Malaysia, it is gaining in test and measurement, power electronics and back-end semiconductor stages.

In North America, Mexico’s growing border assembly of heavy racks and power systems for U.S. data centers could prove more than just a cost-effective alternative. Shorter transit times, the USMCA framework and access to skilled labor are attracting final integration interest, even if numerous sub-components continue to originate from Asia.

The Engineering and Regulatory Hurdles Ahead

Complex electronics are more difficult to move than flipping a switch. Multi-layer PCBs need special materials and plating chemistries, with yields only getting better after months of refining these. Tolerances for high-density racks are unforgiving, and power shelves need to pass rigorous safety and electromagnetic compliance testing. Each shift in plant or supplier requires qualification rides, reliability runs and firmware twiddles that ripple through schedules.

China has an edge in density: a network of sub-suppliers for screws, heat sinks and cables that are all reachable by car. Building that stack in new places just takes time and money. Analysts and supply chain managers repeatedly caution that the first couple of waves of diversification may bear mid- to high-single-digit cost premiums until economies of scale take over.

Implications for AI Buildouts and Cloud Timelines

The timing is delicate. Cloud vendors are doing what they can to bring on GPU clusters, liquid cooling and uprated power distribution. Dell’Oro Group foresees double-digit growth in data center capex into the middle of this decade led by AI’s outsized share. A hiccup in server boards, power supplies or rack integration can slow deployments as customers wait in line for training and inference capacity.

On the plus side, a wider base of suppliers can reduce systemic risk and increase bargaining power over time. It could also drive further standardization in next-generation open server designs, enabling easier dual-sourcing without compromising performance.

The Policy Backdrop Shaping Tech Supply Chains

The drumbeat is trade policy as always. The U.S. has tightened export controls on advanced semiconductors and lithography, while still maintaining tariffs in place for a variety of electronics inputs. Europe is moving ahead with supply chain due diligence rules, and India’s production-linked incentives are wooing global OEMs. Meanwhile, the World Bank estimates that China contributes about a third of global manufacturing value-added — which shows why decoupling is hard and partial at best.

What to Watch Next in Global Cloud Hardware

Key signals include the speed at which server motherboards and power systems are qualified outside of China, the portion of final assembly that’s transferred to Thailand and Mexico by 2026, and an indication that cost premiums narrow as factories ramp. Pay attention to supplier disclosures from the big ODMs, capacity news tied to AI server lines, and any salves or incentives that push the economics back towards—or further away from—China.

The headline, however, is clear: the hardware heart of the cloud is being rewired. And the pace and accuracy at which Microsoft, AWS, and Google make this pivot will determine the speed — and cost — of AI infrastructure for years to come.