Geopolitics & Sovereign Positioning
Top Line
China activated its first 10,000-card AI cluster built entirely with Huawei Ascend 910C chips in Shenzhen, demonstrating tangible progress in domestic AI infrastructure despite US export controls.
South Korean memory chip makers Samsung and SK Hynix increased China fab investments by 67.5% in 2025 to meet global AI demand, showing US export restrictions have not severed South Korea's manufacturing dependency on Chinese facilities.
Iran designated US tech companies including AWS, Google, Microsoft, and Nvidia as legitimate military targets following drone strikes on AWS data centers in UAE and Bahrain, establishing AI cloud infrastructure as active front-line military assets.
Chinese GPU makers Biren and Iluvatar CoreX posted triple-digit revenue growth but continued losses, indicating China's domestic AI chip ecosystem is scaling commercially but remains pre-profitability and vulnerable.
Key Developments
China demonstrates progress on domestic AI compute despite export controls
Shenzhen activated China's first 10,000-card intelligent computing cluster using Huawei's Ascend 910C chips with 11,000 petaflops capacity, according to South China Morning Post. This represents China's largest domestic AI infrastructure deployment to date and signals Huawei has achieved sufficient yield and reliability to deploy its advanced chips at scale. The facility's activation comes three years after US export controls targeted China's access to cutting-edge AI chips.
Separately, Chinese GPU makers Biren Technology and Iluvatar CoreX reported revenue growth of 207.2% and similar triple-digit increases respectively in 2025, though both remain unprofitable, per SCMP. Biren's revenue reached 1.03 billion yuan ($149 million), beating estimates. The earnings reports, their first since January listings, provide rare transparency into China's domestic GPU ecosystem and suggest domestic customers are deploying these chips commercially despite performance gaps versus Nvidia.
South Korea deepens China manufacturing ties despite US pressure
Samsung Electronics increased investment in its Xian, China chip plant by 67.5% to 465.4 billion won ($308.8 million) in 2025, with SK Hynix similarly expanding its Chinese operations to address global AI memory shortages, according to SCMP. The investments underscore that despite US export controls targeting Chinese AI capabilities, South Korean manufacturers view their Chinese fabs as essential capacity for meeting surging global AI demand.
This creates a strategic contradiction: South Korea is a US treaty ally and part of the Chip 4 alliance meant to coordinate semiconductor policy toward China, yet its leading companies are increasing dependence on Chinese manufacturing facilities. South Korea also faces supply chain vulnerability from Middle East instability, importing 77% of its naphtha—a critical petrochemical for chip production—from that region, per SCMP.
AI cloud infrastructure becomes kinetic military target
Iran conducted drone strikes against Amazon Web Services data centers in UAE and Bahrain on March 1 following Israeli and US attacks, causing structural damage and service disruptions, and designated major US tech companies including AWS, Google, Microsoft, Nvidia, Oracle, and Palantir as legitimate military targets, according to SCMP. The strikes represent the first documented kinetic attacks specifically targeting commercial AI cloud infrastructure based on its dual-use military value.
This marks a doctrinal shift where concentrated cloud infrastructure—essential for AI training and inference—is treated as valid military targets during conflict. The targeting rationale appears based on these companies' Israeli business relationships and provision of cloud services to defense and intelligence customers. For countries hosting major cloud facilities, this creates new considerations about infrastructure vulnerability that traditional cybersecurity frameworks do not address, as explored in Atlantic Council analysis of cloud security gaps.
US releases AI policy framework amid institutional fragmentation
The White House released a National Policy Framework for AI, analysed by Georgetown CSET, outlining policy priorities though prospects for near-term legislative action remain unclear. The framework arrives as key US cybersecurity institutions have been weakened or dissolved, according to Atlantic Council assessment, creating gaps in governance capacity precisely as AI raises security stakes.
Separately, Apple accidentally deployed its Apple Intelligence features in mainland China on Tuesday before swiftly withdrawing them—a release that violated Chinese requirements for AI security evaluations, algorithm filings, and data protection approvals, per SCMP. Industry experts warn the incident could expose Apple to regulatory penalties and demonstrates the compliance complexity of deploying AI features across divergent regulatory regimes.
Signals & Trends
AI sovereignty strategies diverging between infrastructure ownership and manufacturing capacity
The pattern across these developments shows countries pursuing AI sovereignty through different vectors—China building domestic compute clusters and chip design capability, South Korea maintaining manufacturing capacity regardless of location, and nations like those in ASEAN potentially caught between dependencies. The Iran strikes and Apple compliance failure suggest a third dimension: regulatory sovereignty over AI deployment within borders. These three forms of sovereignty—infrastructure, manufacturing, and regulatory—are increasingly recognized as separable rather than bundled, allowing countries to optimize for different strategic positions rather than pursuing full-stack independence. This creates more complex alliance dynamics than simple US-China bipolar framing suggests.
Physical concentration of AI infrastructure creating escalation pathways
Iran's targeting of AWS facilities and NATO analysis of AI in algorithmic warfare, per Atlantic Council, indicate military planners now view concentrated AI compute as force multipliers worth neutralizing. The economic logic of AI—which rewards massive concentrated facilities for training efficiency—directly conflicts with military logic favoring distributed resilient systems. This creates an unresolved tension: the countries and companies making the largest AI capability investments are simultaneously creating the most attractive strategic targets. Neither commercial nor military institutions have adapted doctrine or architecture to resolve this trade-off, suggesting current AI infrastructure buildout could create brittle rather than robust strategic capabilities.
Global South positioning as resistance leverage point rather than pure recipients
The Rest of World reporting on AI infrastructure resistance in Chile, Mexico, Kenya, and the Philippines shows communities successfully challenging data center and digital labor deployments based on environmental and social costs. This resistance differs from great power competition dynamics but affects the same infrastructure—if communities can delay or block facility construction, they hold de facto veto power over where AI capacity gets built. For countries seeking to reduce dependency on US or Chinese AI infrastructure, this community resistance becomes strategically relevant: it's easier to build sovereign alternatives when competing powers face local opposition. The pattern suggests AI infrastructure negotiations increasingly involve three parties—great powers, host governments, and local communities—rather than bilateral state-to-state arrangements.
Explore Other Categories
Read detailed analysis in other strategic domains