Geopolitics & Sovereign Positioning
Top Line
China's two largest chip foundries, SMIC and Hua Hong, posted triple-digit quarterly profit growth driven by surging domestic demand for AI chips outside US export control reach, signalling that sanctions are accelerating China's domestic semiconductor ecosystem rather than constraining it.
Beijing is simultaneously advancing on two fronts — model capability and governance architecture — with Chinese AI models now matching or exceeding Western frontier models on cybersecurity benchmarks while China pushes its WAICO framework to set international AI rules.
Open-weight Chinese AI models, including Alibaba's Qwen series and DeepSeek variants, are gaining commercial adoption across European enterprises, creating a structural tension between Brussels' technological sovereignty goals and the cost-competitiveness of Chinese alternatives.
South Korea has announced an $880 billion AI investment ambition anchored in its chipmaking base, positioning itself as a swing-state actor between US alliance commitments and the broader AI infrastructure buildout.
Guangdong province's strategic framework agreement with Alibaba — covering compute, AI models, and semiconductors — illustrates how China is fusing provincial industrial policy with national AI champions to create integrated regional AI stacks.
Key Developments
Export Controls Are Backfiring: China's Foundry Boom Demonstrates Sanctions Displacement Effect
SMIC and Hua Hong Grace Semiconductor reported net profit surges of 261.7 percent and 385.9 percent year-on-year respectively in Q2 2026, driven explicitly by demand for domestic AI chips that fall outside US export control parameters. SMIC's CEO Zhao Haijun confirmed on an earnings call that orders are 'far exceeding previous expectations' and the company is evaluating capacity additions at existing sites. This is not a lagging indicator — it is a real-time measure of how the US sanctions regime has created a captive domestic market for Chinese foundries that previously competed on price against TSMC and Samsung. South China Morning Post
The strategic logic is straightforward: by restricting Chinese access to advanced-node chips, Washington has forced Chinese AI developers to optimise for mature-node architectures, which SMIC and Hua Hong can manufacture at scale. The mature-node chip shortage triggered by AI infrastructure demand — noted in SMIC's investor communications — means Chinese foundries are now capacity-constrained and investing in expansion, not struggling for customers. The second-order consequence US policymakers did not fully anticipate is that this dynamic is funding the very R&D investment in advanced manufacturing that export controls were designed to prevent. South China Morning Post
China's Dual-Track AI Strategy: Model Supremacy and Governance Standard-Setting Simultaneously
Chatham House analysis of China's WAICO (World AI Cooperation Organisation) initiative and the Kimi K3 model release frames Beijing's approach as explicitly dual-track: achieve frontier model capability while simultaneously positioning Chinese governance frameworks as the global default. The strategic logic is that whichever power sets AI governance norms — on safety standards, content moderation, data sovereignty, and liability — will shape the terms on which other nations deploy AI, regardless of which country's models they run. China's 2017 Next Generation AI Development Plan set a 2030 deadline for global AI leadership; WAICO is the institutional vehicle for translating model capability into rule-making authority. Chatham House
This is reinforced by the cybersecurity dimension. Zhipu's GLM-5.3 model reportedly achieved 84.5 percent on the CyberGym benchmark, exceeding Anthropic's Mythos 5 at 83.8 percent and OpenAI's comparable offering. DeepSeek's V4-Pro-0813 update similarly drew attention for cybersecurity performance despite underwhelming general benchmarks. The concentration of Chinese model advancement in offensive and defensive cybersecurity capabilities is not coincidental — it reflects PLA and state security procurement priorities feeding into civilian AI development cycles. South China Morning Post
Chinese Open-Weight Models in Europe: Sovereignty Paradox Creates Strategic Vulnerability
European enterprises are adopting open-weight Chinese models — including Alibaba's Qwen series — at increasing rates, partly on the argument that running models on local servers provides more operational control than using proprietary US cloud services. The cited logic from Germany-based consultants is that locally-deployed Chinese open-weight models reduce API dependency and data transmission to foreign servers, addressing one dimension of technological sovereignty. Brussels has expressed trepidation but has not enacted any binding restrictions on deploying Chinese open-weight models, leaving the decision to individual firms and member states. South China Morning Post
This creates a structural vulnerability that the 'local deployment' framing obscures. Open-weight models carry their training data biases, embedded knowledge cutoffs, and potential architectural backdoors into European enterprise environments regardless of where inference occurs. The sovereignty calculus being used by European businesses — local compute equals sovereignty — conflates infrastructure control with model provenance control. Alibaba's decision to add commercial licensing restrictions to Qwen3.8-Max for entities with over $50 million in annual revenue also signals that Alibaba is now actively managing its open-weight release strategy for commercial leverage, not simply competing on openness. South China Morning Post
South Korea's $880 Billion AI Bet: Chipmaking Leverage Meets Alliance Complexity
South Korea has announced plans anchored in an $880 billion AI investment framework intended to leverage its existing semiconductor manufacturing base — anchored by Samsung and SK Hynix — into a broader AI infrastructure position. The Diplomat analysis frames this as South Korea seeking to convert its status as an indispensable node in the global chip supply chain into a more assertive AI sovereignty posture. This is a diplomatic commitment at this stage, not an enacted policy with enforcement mechanisms, but the scale signals a government-level determination to avoid being reduced to a component supplier in a US-China bifurcated AI ecosystem. The Diplomat
South Korea's strategic position is inherently complex: it is a US treaty ally and deeply integrated into US-led semiconductor export control arrangements, yet its largest export market is China and its chipmakers supply critical HBM memory to Nvidia. Seoul is attempting to use its chokepoint position in AI hardware — particularly HBM3E production — as leverage to secure preferential terms in both US technology alliance frameworks and access to Chinese markets, rather than choosing a side. The $880 billion figure should be read as a sovereign positioning signal directed at Washington and Beijing simultaneously.
Signals & Trends
China's Provincial AI Stacks Signal a Deliberate Vertical Integration Strategy Below the National Level
The Guangdong-Alibaba strategic framework agreement — covering compute investment, AI models, semiconductor development, and smart public services — is one of multiple such provincial-level deals emerging across China's wealthiest regions. This is not simply industrial policy; it is Beijing constructing vertically integrated AI ecosystems at the provincial level that link sovereign compute infrastructure, domestically developed models, and state procurement into a single stack. The strategic consequence is that China is building AI capability redundancy: even if US pressure disrupts any single national champion, provincial ecosystems have sufficient scale and integration to sustain AI deployment. Foreign policy analysts should track these provincial agreements as leading indicators of China's actual AI deployment capacity, which is likely outpacing Western estimates focused on frontier model benchmarks.
Cybersecurity Benchmarks Are Becoming the Contested Terrain of AI Military Signalling
The convergence of Chinese model releases — Zhipu GLM-5.3, DeepSeek V4-Pro, and implicitly the Kimi K3 positioning — around cybersecurity capability claims is a pattern, not a coincidence. When Chinese AI labs choose to highlight CyberGym performance against Anthropic's Mythos model rather than general reasoning benchmarks, they are communicating capability signals to audiences beyond the developer community. This mirrors the dynamic in hypersonic weapons development where performance claims serve deterrence and diplomatic functions independent of operational deployment. The practical implication for Western intelligence services is that open-source Chinese model releases now function as a dual-use capability disclosure mechanism — releasing genuine offensive cybersecurity capability into global circulation while signalling state-level AI military parity. The absence of an agreed international framework for AI capability disclosures in the military domain means this signalling operates in a norm-free environment.
The Open-Weight Model as Geopolitical Instrument Is Maturing Into a Distinct Strategic Category
Alibaba's introduction of commercial licensing restrictions on Qwen3.8-Max for large enterprises, combined with continued free access for smaller users and researchers, represents the maturation of open-weight model strategy from a competitive tactic against OpenAI into a deliberate geopolitical instrument. The tiered licensing structure maximises adoption breadth — embedding Chinese model architectures, training biases, and knowledge representations across global developer ecosystems — while extracting commercial value from the largest deployers. This is structurally analogous to how Huawei's subsidised infrastructure penetration created dependency before commercial terms were tightened. Western governments have not yet developed a policy response to open-weight model proliferation that distinguishes between genuinely open models and strategically tiered releases designed to maximise geopolitical footprint — the EU's AI Act and US executive orders both have significant gaps in this area.
Explore Other Categories
Read detailed analysis in other strategic domains