Geopolitics & Sovereign Positioning

Executives at the 2026 HKEX conference assessed China as approximately 100 days behind the US on frontier model capabilities — a remarkably compressed gap given the scale of export controls on advanced compute — while simultaneously leading in consumer AI application penetration. Alibaba Cloud's Chi Zhang framed the compute deficit as the primary remaining structural disadvantage, but the deployment lead suggests Chinese firms have found ways to optimise for real-world use at scale despite hardware constraints. The assessment comes with a significant caveat: the same experts warning of China's progress also flagged Chinese AI firms as increasingly overvalued, suggesting capital markets may be pricing in expectations that outrun current fundamentals. South China Morning Post

The 100-day framing is analytically significant because it implies that US export controls have not produced the 2-3 generation lag that policymakers originally anticipated. If accurate, it suggests controls are delaying rather than decoupling Chinese AI development, and that Chinese firms have partially compensated through algorithmic efficiency, stockpiled legacy chips, and domestic semiconductor alternatives. The deployment advantage in consumer applications matters geopolitically because it shapes which AI ecosystems developing markets adopt — China's edge in affordable, integrated AI products gives it a commercial influence vector that pure model benchmarks don't capture.

Both US and Chinese research communities are actively pursuing recursive self-improvement — the capacity for AI systems to autonomously redesign and enhance their own architectures — with both governments now treating it as a potential decisive strategic inflection point. The theoretical logic is stark: the first actor to achieve stable RSI would compound capability advantages at a pace that renders all current gaps irrelevant, potentially within months. Chinese state media and research institutions have begun discussing RSI publicly, signalling that it has moved from academic speculation into active strategic planning. South China Morning Post

The RSI competition introduces a structural instability into current deterrence frameworks. Unlike conventional military capability, which accumulates gradually and is observable, an RSI breakthrough could be rapid, initially covert, and strategically decisive before adversaries can respond. This creates strong incentives for preemptive action — either accelerating domestic programs or taking steps to disrupt a competitor perceived as close to a breakthrough. The framing of RSI as a national security priority by both Washington and Beijing elevates AI development from an economic competition into a domain with direct strategic stability implications.

A War on the Rocks analysis makes the case that ASML's monopoly on EUV lithography systems and TSMC's concentration of advanced semiconductor manufacturing constitute functional deterrents to conflict over Taiwan — not through military threat but through the credible promise of mutual economic catastrophe. The argument is that any military action disrupting TSMC's operations would deny the aggressor access to the chips underpinning its own AI and defence modernisation programs, creating a self-defeating dynamic. The Dutch government's control over ASML export licenses adds a European dimension to this chokepoint architecture. War on the Rocks

The analysis has direct policy implications for how Western governments should manage the deliberate diversification of semiconductor manufacturing. Programs to build fabs in the US, Europe, Japan, and elsewhere — while strategically sound for resilience — potentially erode the Taiwan-specific deterrent value of geographic concentration. If TSMC's advanced production is successfully replicated elsewhere before conflict materialises, the chokepoint deterrence mechanism weakens precisely as it was intended to be most useful. This creates a tension between resilience strategy and deterrence strategy that policymakers have not publicly resolved.

Memory module maker Biwin has signed a two-year, $1.86 billion supply agreement for enterprise-grade flash memory chips — a contract larger than the company's annual revenue — covering Q3 2026 through Q2 2028. The supplier was not disclosed in the exchange filing, a detail that may reflect sensitivity around sourcing relationships under the current sanctions environment. The deal structure — locked volume and locked price — signals that Biwin and its counterparts are treating supply security as a higher priority than price optimisation, accepting cost and balance sheet risk to guarantee access to components critical for AI server and data centre builds. South China Morning Post

This pattern of pre-positioning is visible across the Chinese AI hardware supply chain. The MLCC market shows a parallel dynamic: Japanese and Korean leaders shifting upmarket creates an opening for Chinese component manufacturers in lower-tier segments, gradually building domestic industrial capacity in AI-adjacent hardware. South China Morning Post Taken together, these moves suggest a coordinated strategy — partly market-driven, partly policy-directed — to reduce dependency on foreign components across the full AI hardware stack over a multi-year horizon.

Apple's WWDC 2026 announcement that the new Gemini-powered Siri will not be available in China — pending regulatory approvals that remain uncertain — is the latest and most consumer-visible instance of the structural split between US and Chinese AI product ecosystems. The exclusion is not purely commercial: deploying a Google-powered AI assistant in China would require navigating data localisation requirements, algorithmic transparency mandates, and content control obligations that are incompatible with the current product architecture. The result is that Chinese iPhone users will receive a functionally inferior version of the flagship AI feature, accelerating incentives to shift to domestic alternatives. South China Morning Post

This bifurcation has compounding effects. It reinforces the commercial case for Chinese AI platform providers — Baidu, Alibaba, ByteDance — as the default AI infrastructure for Chinese consumers and enterprises. It also shapes what developing markets receive: countries choosing between US and Chinese-stack devices are increasingly choosing between different AI capability sets, not just different hardware brands. The regulatory divergence is now so deep that a unified global AI consumer product is effectively impossible for a US firm operating in both markets simultaneously.