Geopolitics & Sovereign Positioning

Beijing is committing approximately $295 billion to domestic AI data centre buildout with an explicit policy of excluding foreign firms from participation, according to The Diplomat. This is an enacted industrial policy with procurement mechanisms, not a diplomatic aspiration — the exclusion of foreign firms reflects the same logic driving the US CHIPS Act's domestic content requirements, but applied with greater bluntness. The direct beneficiaries are Chinese AI infrastructure companies, whose valuations are expected to rise as guaranteed state demand insulates them from international competition.

The strategic significance extends beyond the investment figure. A domestically controlled AI compute layer — data centres, interconnects, power systems — means China is building sovereignty into the physical stack, not just the model layer. This reduces exposure to any future escalation of US sanctions targeting cloud or compute access, and it creates a captive market for Chinese semiconductor firms to scale production volumes even as they remain behind the leading edge. The dependency Beijing is eliminating is not merely economic; it is the leverage that Western governments have historically assumed they could exercise through infrastructure chokepoints.

The Diplomat identifies three specific technical barriers China must clear to produce functional EUV lithography equipment: light source power, optical precision at scale, and resist chemistry. The article's core analytical contribution is a call for granular monitoring of progress on each barrier individually, rather than treating EUV as a monolithic gap. This matters for policy because Western export control regimes — coordinated through the Wassenaar Arrangement and bilateral US-Netherlands-Japan agreements restricting ASML equipment exports — were designed on the assumption that EUV represents an insurmountable barrier. If China closes one of three sub-problems, the control architecture may need recalibration before the overall barrier falls.

The second-order consequence of sustained EUV denial is already visible: China has industrialised multi-patterning techniques on DUV equipment to produce chips at nodes closer to the leading edge than originally projected. This mirrors historical patterns — the CoCom embargo on Soviet technology in the Cold War accelerated Soviet indigenous development in specific subsectors even as it constrained overall capability. Export controls are confirmed to be creating friction and delay; they are not confirmed to be creating permanent incapacity. The distinction is critical for any realistic assessment of the 2028-2030 competitive window.

Two confirmed US military AI developments this week signal operational maturation rather than continued pilot-phase experimentation. The Pentagon has launched a 'War Force' programme recruiting young two-year technical talent with direct access to policymaking processes, as reported by Defense One. Separately, the Army is deploying AI-enabled autonomous boats for logistics operations across the Pacific, with a senior general explicitly framing the theatre as the proving ground for global applicability — 'if you can work in the Pacific, you can work anywhere,' per Defense One. The Pacific framing is not incidental; it is a direct signal about which contingency is driving operational AI integration timelines.

The talent recruitment initiative addresses a structural vulnerability that has long limited DoD AI implementation: the gap between technical capability in the private sector and institutional capacity within the military to deploy and sustain AI systems. The two-year rotation model is designed to import commercial AI talent without requiring full military careers, acknowledging that the DoD cannot compete on compensation but can offer mission-scale impact. Whether this closes the talent gap or merely papers over it depends on absorption capacity within programme offices — a question that will be answered by implementation, not announcement.

Nvidia is actively recruiting for more than a dozen robotics roles across Beijing, Shanghai, and Shenzhen, focused on embodied intelligence, simulation, and implementation, according to South China Morning Post. The move is legal under current export control architecture — robotics software, simulation platforms, and solutions engineering are not currently controlled at the same level as the most advanced AI chips — but it underscores a fundamental tension in the US approach. Washington has restricted Nvidia's H100 and A100 chip sales to China, yet Nvidia is simultaneously building organisational capacity in China to serve a robotics market where Chinese vendors dominate global shipment volumes.

The physical AI and robotics sector is precisely where the next competitive frontier lies: the integration of AI models with real-world mechanical systems for manufacturing, logistics, and potentially defence applications. China's robotics firms — led by companies like Unitree and UBTECH — are scaling rapidly on domestic hardware. Nvidia's presence in this ecosystem gives it commercial relevance but also creates knowledge transfer risks that current export controls were not designed to address. The Bureau of Industry and Security has not classified robotics simulation and embodied intelligence software under the most restrictive export categories, leaving a meaningful gap between the intent of chip controls and their practical effect on AI capability diffusion.