Geopolitics & Sovereign Positioning

The strategic logic underpinning US chip export controls — that denying advanced semiconductors would constrain Chinese AI development — is facing its most serious empirical challenge to date. South China Morning Post reports that Nvidia CEO Jensen Huang's proximity to the Beijing summit reflects the commercial pressure building on Washington: controls have cost US chipmakers billions in addressable revenue while Chinese firms have accelerated domestic substitution. The result is a policy that imposes asymmetric costs — restraining US commercial upside without decisively degrading Chinese capability trajectories.

The Trump administration's options are structurally constrained. Escalation — tightening controls further, expanding the entity list, or extending restrictions to cloud-delivered compute — risks accelerating Chinese self-sufficiency and alienating allies whose firms also bear compliance costs. Relaxation would be read as strategic capitulation and invite domestic political blowback. SMIC founder Richard Chang's public call for China to focus on 'pragmatic breakthroughs' in mature and specialty nodes, rather than chasing 2nm parity, is analytically significant: it suggests Chinese industry is internalising a strategy of supply-chain hardening at accessible technology levels rather than direct frontier competition, which is precisely the outcome that makes export controls less decisive over time. South China Morning Post frames this as a call for 'supply chain security', language that signals institutional rather than purely commercial intent.

Chinese contributors now claim majority share of accepted papers at a leading global AI conference, with mainland China and Hong Kong affiliations accounting for over 51% of accepted submissions versus under 32% for the US, according to data compiled from listed affiliations. South China Morning Post notes this occurred despite active uncertainty over whether Chinese researchers would participate given bilateral tensions and an ongoing NeurIPS dispute. The decision to attend in force — and the research volume to sustain that dominance — reflects institutional confidence that international scientific engagement remains strategically valuable to Beijing regardless of diplomatic friction.

This is not merely a soft-power signal. Conference publication share is a leading indicator of the talent pipeline, research agenda influence, and foundational model capability that translate into commercial and military AI applications over a 3-7 year horizon. A sustained Chinese majority at frontier AI research venues means Beijing is shaping the global knowledge commons from which all actors — including US labs — draw. The parallel finding from the Deep Knowledge Group's Global AI Competitiveness Index, which places China third globally in AI for life sciences behind only the US and UK, South China Morning Post reinforces that Chinese AI capability is broadening into high-value, data-intensive verticals beyond consumer applications.

Japan Airlines has announced a two-year trial deploying humanoid robots from Chinese firms Unitree and UBTech for baggage and cargo handling at Tokyo's Haneda Airport, in partnership with GMO AI and Robotics. South China Morning Post frames this primarily as a response to Japan's acute labour shortage, but the strategic implications extend well beyond workforce economics. This is Chinese AI-embodied hardware operating inside the critical infrastructure of a core US security ally, with a two-year data-generating operational footprint inside one of Asia's busiest international airports.

Japan's acute demographic challenge — among the most severe in the developed world — is creating systematic pull demand for Chinese robotics that market-access restrictions alone cannot easily offset. Unlike software or chips, humanoid robots are a dual-use category where the cost-capability gap between Chinese and US or Japanese alternatives is widening in China's favour. The Unitree and UBTech deployments are commercially driven, but they establish operational precedents, supply chain dependencies, and data collection relationships that are structurally difficult to unwind once embedded. This dynamic will recur across labour-scarce allied economies in Europe and Southeast Asia.

The upcoming Trump-Xi summit in Beijing is placing AI governance at the centre of great-power diplomacy, but as The Diplomat argues, neither government's existing policy architecture was designed for the most consequential dimension of the current competition: the software layer. Hardware controls, entity lists, and investment screening are the available instruments, but they are poorly calibrated to address model weight transfer, API-mediated capability access, open-source proliferation, and the embedding of Chinese AI applications in third-country digital infrastructure. These are the vectors through which influence and dependency are currently being established at scale.

The diplomatic framing matters because it shapes what agreements are even articulable. A summit outcome focused on semiconductor trade — relief for Nvidia, reciprocal concessions on rare earths or market access — would leave the software-layer competition entirely unaddressed. Conversely, any attempt to negotiate AI application restrictions or model sharing norms would require verification mechanisms that do not exist and technical definitions that both sides would contest. The summit is more likely to produce declaratory commitments on AI safety dialogue — building on the 2023 San Francisco and 2024 Geneva precedents — than binding instruments, but even process-level engagement has value in establishing escalation-management channels for AI-enabled military incidents.