Geopolitics & Sovereign Positioning
Top Line
China's chipmaking sector posted a 2,579% profit surge in H1 2026, confirming that US export controls are accelerating domestic alternatives rather than suppressing Chinese AI compute capacity.
Anthropic CEO Dario Amodei has publicly called on Washington to tighten chip export bans and crack down on model distillation, framing both as prerequisites for preserving a US military-relevant AI lead — a direct intervention by a leading frontier lab into export control policy.
The open-weight model front is becoming a second theatre of US-China AI competition: Chinese models like Kimi K3 are now freely downloadable globally, reaching users and infrastructure that US export controls on hardware cannot touch.
China's focus on 'supernodes' — integrating hundreds of domestically produced chips into unified compute clusters — represents a structural architectural response to chip-level sanctions, potentially circumventing the per-chip performance gap.
Taiwan's military remains in the early stages of AI adoption despite producing the world's most advanced semiconductors, creating an asymmetric vulnerability as the PLA accelerates AI-enabled military capability development.
Key Developments
China's Domestic Chip Ecosystem Demonstrates Export Control Resistance
China's major chip manufacturers recorded profit growth of 2,579.5% in the first half of 2026, according to China's National Bureau of Statistics. The NBS attributed the surge directly to AI-driven compute demand — a signal that domestic chipmakers are capturing investment and revenue that would previously have flowed to foreign suppliers. South China Morning Post This is not a lagging indicator of pre-control inventory drawdown; it reflects structural demand being met by domestic supply chains.
The parallel showcase at the World Artificial Intelligence Conference of 'supernode' architectures — designs from Huawei and Biren Technology that network hundreds or thousands of domestically produced chips into unified compute clusters — provides architectural context for those profit numbers. South China Morning Post Rather than attempting chip-for-chip parity with Nvidia, Chinese developers are engineering system-level workarounds that aggregate lower-performance hardware into frontier-scale compute. Alibaba's simultaneous extraction of windfall returns from stakes in both CXMT and Zhipu AI illustrates how China's major tech platforms are integrated across the chip-to-model stack in ways US export controls cannot cleanly sever. South China Morning Post
Anthropic Pushes for Tighter Controls as Open-Weight Models Expose Export Control Boundaries
Dario Amodei published a public call for Washington to tighten chip export bans and restrict Chinese model distillation, describing both as essential to preventing Beijing from developing frontier AI for military and surveillance applications. South China Morning Post His framing is strategically significant: by centering the argument on military use and domestic surveillance rather than economic competition, Amodei is attempting to reframe the export control debate on national security rather than commercial grounds, making restriction politically easier to sustain.
The timing is directly linked to the open-weight pressure. Moonshot AI's release of Kimi K3 — including full weights and infrastructure tooling — has triggered explicit debate in Silicon Valley about whether bans on Chinese open-source models are warranted. South China Morning Post The Atlantic Council has framed this as a governance gap: export controls built around data centres cannot reach the laptop, and the answer it proposes is competitive open-model development by Western labs rather than prohibition. Atlantic Council These two positions — Amodei's call for tighter restriction versus the Atlantic Council's competitive response strategy — define the current policy fault line in Washington.
DeepSeek's Strategic Posture and the Chinese Open-Weight Commercial Transition
Leaked comments from DeepSeek founder Liang Wenfeng reveal a deliberate strategy of maintaining low commercial visibility while pursuing frontier capability development — a posture that complicates Western threat assessment and policy targeting. South China Morning Post DeepSeek's backing by High-Flyer Quant gives it a patient capital structure insulated from the commercial revenue pressures that typically constrain AI labs, making it an outlier in the landscape.
Goldman Sachs has flagged a potential strategic inflection for the Chinese open-weight ecosystem more broadly: Chinese AI developers including Moonshot and Zhipu AI may shift toward 'paid weights' commercial licensing, charging cloud platforms to host their models. South China Morning Post If this transition occurs, it would transform the current open-weight global distribution from a pure geopolitical soft-power instrument into a revenue-generating infrastructure dependency — cloud providers globally would be paying licensing fees to Chinese AI developers, creating a commercial lock-in dynamic that mirrors the strategic dependency concerns historically raised about US cloud dominance.
Taiwan's AI Military Gap Creates Asymmetric Vulnerability Ahead of Any PLA Confrontation
Taiwan produces the world's most advanced semiconductors but remains in the early stages of military AI adoption, according to analysis in War on the Rocks. War on the Rocks The PLA is actively developing and fielding AI-enabled military systems with Taiwan as a primary operational planning scenario, meaning Taiwan's own defence could be the first to contend directly with adversarial military AI at scale. This is a textbook capability paradox: the producer of the foundational technology for AI compute cannot translate that production capacity into its own military AI readiness at the pace required.
The implications extend beyond Taiwan's bilateral deterrence posture. Coalition defence scenarios — including any US or allied intervention — depend on interoperability, and Taiwan's nascent military AI integration creates gaps in joint operational architecture that adversaries can exploit. The semiconductor production facilities themselves remain a strategic liability: any conflict scenario that disables TSMC's fabs removes a critical chokepoint in the global AI supply chain, incentivising both deterrence and coercion calculus by multiple parties simultaneously.
Global South Excluded from AI Governance Frameworks as Western Institutions Define the Rules
An open letter signed by over 200 economists and AI researchers, including 16 Nobel laureates and organised by the Stanford Digital Economy Lab, called for urgent institutional action to govern AI's economic transformation. The strategic problem: four in five signatories come from Western institutions, leaving the majority of the world's population unrepresented in what is framed as a universal governance imperative. South China Morning Post This is not an oversight — it reflects structural exclusion from the expert networks that feed into policy frameworks like the G7 AI principles, the EU AI Act, and US executive orders.
For foreign policy strategists, the governance exclusion has operational consequences. Countries across Africa, Southeast Asia, Latin America, and South Asia that lack voice in Western-designed frameworks are more receptive to Chinese AI infrastructure investment and standards proposals precisely because China presents itself as a fellow non-Western actor. Tencent's accelerating portfolio investment across Chinese AI champions — including a $3 billion round for Kling AI — signals that Chinese platform capital is being systematically deployed to build global-facing AI product capacity that can operate outside Western governance frameworks. South China Morning Post
Signals & Trends
Export Controls Are Reshaping Chinese AI Architecture, Not Stopping It
The confluence of the supernode architectural shift, 2,500% chip profit growth, and DeepSeek's continued frontier development despite hardware restrictions points to a consistent pattern: US export controls are functioning as a forcing function for Chinese architectural innovation rather than a capability ceiling. The strategic logic of controls rested on a chokepoint theory — deny access to leading-edge chips, deny access to frontier compute. Chinese developers and hardware firms are systematically engineering around the chokepoint at the system level. The policy implication is significant: controls calibrated to chip-level specifications require continuous updating to account for system-level aggregation, and the administrative lag between Chinese architectural adaptation and US regulatory response is where Chinese capability development is occurring.
Open-Weight Model Distribution Is Becoming China's Primary AI Soft Power Instrument
The release of Kimi K3 with full weights and infrastructure tooling, the broader pattern of Chinese open-weight releases, and the Goldman Sachs assessment that commercial licensing may follow all point to a deliberate sequencing: build global developer dependency through free distribution, then convert that dependency into commercial leverage. This mirrors China's Belt and Road infrastructure playbook — subsidised entry, followed by terms renegotiation once dependency is established. The critical difference from hardware infrastructure is velocity: software dependency can be established globally within months of a model release, far faster than physical infrastructure build-out. Western policy frameworks have no established regulatory mechanism to govern this vector.
The Military AI Adoption Gap Is Widening Between China and Its Most Likely Adversaries
Taiwan's documented lag in military AI adoption, combined with PLA acceleration, is a specific instance of a broader pattern: the countries most directly in China's strategic crosshairs are also the ones with the most friction in translating civilian AI capability into military readiness. This is partly bureaucratic — democratic procurement and integration processes are slower than authoritarian directive deployment. But it is also structural: Taiwan produces the chips but does not have the system integration expertise at scale; South Korea and Japan have advanced civilian AI sectors but significant constitutional and political constraints on military AI deployment. The PLA's advantage may therefore be less about absolute capability than about the speed of integration into operational doctrine — a gap that is harder to close with technology transfer alone.
Explore Other Categories
Read detailed analysis in other strategic domains