Geopolitics & Sovereign Positioning
Top Line
China's Ministry of State Security has issued a formal warning against using relay services to access foreign AI models, signalling that Beijing views the grey market for tools like ChatGPT as both a security threat and a sovereignty challenge — a move that tightens the information perimeter around Chinese AI infrastructure.
China is executing a coordinated data sovereignty play: the National Data Administration has published a sweeping plan to build state-curated, industry-specific AI training datasets, positioning data supply as a strategic asset in the same tier as compute and models.
Chinese AI models have become the dominant choice for African developers, according to Foreign Policy, handing Beijing a structural foothold in Global South AI adoption that Western export controls have inadvertently accelerated by pricing out American alternatives.
Trump's National Security AI Memo has drawn mixed expert assessments from CFR, with analysts noting it correctly frames AI as a national security priority but leaves unresolved the institutional architecture needed to operationalise that framing — a gap Congress has not yet filled.
A Foreign Policy analysis finds that U.S. efforts to export military AI to middle powers face structural friction: AI labs retain strong views on end-use, middle powers have Chinese and European alternatives, and there is no agreed framework governing technology transfer conditions.
Key Developments
China's MSS Criminalises the AI Grey Market — Tightening the Sovereignty Perimeter
China's Ministry of State Security published a formal notice warning that 'AI relay services' — intermediaries that give Chinese users and developers access to restricted foreign models like OpenAI's GPT series — constitute security risks including data leaks, unauthorised cross-border data transfers, and potential backdoor exposure. This is an enacted regulatory warning with enforcement implications, not merely commentary: the MSS has existing authority to prosecute actors deemed to violate national security laws, and this notice establishes the conceptual predicate for doing so. South China Morning Post
The strategic logic runs in two directions simultaneously. Domestically, the crackdown protects Chinese AI providers — DeepSeek, Qwen, Kimi — from being undercut by grey-market access to frontier Western models, reinforcing the commercial viability of the domestic ecosystem. Externally, it accelerates the bifurcation of the global AI stack by making the Chinese digital environment less permeable to Western model influence. The net effect is to deepen Chinese developers' home-market captivity while reducing the soft-power surface area through which American AI companies might otherwise build usage habits and developer lock-in inside China.
China Moves on Data Sovereignty — Strategic Asset Framing of AI Training Supply
China's National Data Administration released a draft plan to construct a national supply of high-quality, industry-specific AI training datasets, framing data as a core strategic input alongside compute and model architecture. This is a policy document with institutional backing, not yet an enacted statute, but the NDA has regulatory authority to implement it. The timing is deliberate: as global AI developers confront what researchers describe as a looming shortage of high-quality training data, China is moving to ensure its domestic developers are not dependent on foreign data infrastructure or open-web scraping that is increasingly legally contested in Western jurisdictions. South China Morning Post
Simultaneously, China is investing in physical AI infrastructure: the first prefabricated green energy hub for computing facilities was unveiled in Qingdao, designed to provide modular, deployable power supply for data centres. Combined with the data sovereignty push, this represents a full-stack sovereign AI infrastructure strategy — compute, energy, and training data all being brought under state-directed supply chains. South China Morning Post
China's AI Models Are Winning Africa — A Geopolitical Foothold Built Through Linguistic Advantage
A Foreign Policy analysis reports that Chinese AI models have become the dominant choice for African developers, driven primarily by two factors: price competitiveness and coverage of African languages that Western models have systematically underserved. This is not a diplomatic commitment or a Belt-and-Road-style infrastructure loan — it is market adoption driven by genuine product-market fit, which makes it structurally more durable than politically negotiated technology partnerships. Foreign Policy
The geopolitical implications extend beyond market share. Developer ecosystems built on Chinese model APIs create dependency chains: African startups optimising for Chinese model architectures, training on Chinese infrastructure, and building with Chinese toolchains will face switching costs that grow over time. This mirrors the dynamic through which American cloud infrastructure locked in Global South developers in the 2010s — except that U.S. export controls and the high cost of frontier American models have ceded this opportunity. The Rest of World event analysis of the 'AI divide' noted that middle economies see both U.S. and Chinese AI dominance as a constraint, but in practice they are making pragmatic choices — and those choices are trending Chinese. Rest of World
U.S. Military AI Export Strategy Is Structurally Constrained — Middle Powers Have Options
A Foreign Policy analysis of U.S. efforts to export military AI to allied middle powers identifies three compounding constraints: AI laboratories retain strong internal views on weapons-adjacent end-uses and have leverage to condition or refuse government export programmes; middle powers have credible alternatives from China and, increasingly, European sources; and no agreed legal or diplomatic framework governs the terms under which military AI technology transfer should occur. The piece frames this as a 'Pax Silica' problem — Washington cannot simply assume that allied nations will accept U.S. military AI on U.S. terms when competitive alternatives exist. Foreign Policy
This matters against the backdrop of Trump's National Security AI Memo, which CFR analysts assess as correctly identifying AI as a national security priority but leaving the institutional architecture underspecified — no clear export promotion framework, no adjudication mechanism for conflicts between lab preferences and national security objectives, and no congressional mandate to fill those gaps. The combination of a voluntarist domestic policy framework and a competitive export environment means U.S. military AI diffusion to allies will be slower and more contested than Washington's strategic posture assumes. CFR
Chinese AI Capital Markets Consolidation — Domestic Ecosystem Matures Under Competitive Pressure
DeepSeek's V4 model release has triggered a cascade of API price cuts across China's domestic AI market, with Xiaomi slashing MiMo-V2.5 API costs by up to 99 percent. This deflationary dynamic is simultaneously a sign of ecosystem health — competitive pressure driving efficiency — and a commercial threat to smaller Chinese AI companies whose revenue models depend on API monetisation. Moonshot AI's bid for a $30 billion valuation in a $1-2 billion funding round, a 50 percent premium over its previous round, suggests investor appetite remains strong for frontier model developers even as the commoditisation of mid-tier models accelerates. South China Morning Post
Alibaba's restructuring into 'Token Foundry' — consolidating Tongyi Lab and Future Life Lab under direct CEO oversight — signals that China's largest technology conglomerate is centralising AI strategy at the executive level, treating model development as a core business function rather than a research subsidiary. The inclusion of MiniMax and Zhipu in the Hang Seng Tech Index formalises Chinese pure-play AI companies as investable assets within Hong Kong's benchmark, potentially channelling significant passive capital into the sector. South China Morning Post
Signals & Trends
The Global South AI Alignment Gap Is Closing — Against Washington's Interests
The combination of Chinese linguistic AI investment in Africa, competitive API pricing, and U.S. export control regimes that raise the cost of American frontier models for price-sensitive markets is producing a quiet but durable reorientation of Global South developer ecosystems toward Chinese AI infrastructure. This is not primarily a diplomatic or Belt-and-Road dynamic — it is a product-market-fit dynamic, which makes it harder to reverse through aid programmes or diplomatic pressure. The Rest of World analysis of the 'AI divide' captures the emerging consensus among non-aligned technology communities: both U.S. and Chinese AI dominance is seen as a constraint, but absent a credible third option, pragmatic adoption follows price and capability signals. Washington's current posture — export controls that restrict access without providing a subsidised alternative — is effectively leaving the field to Chinese providers in the markets that will determine the digital infrastructure norms of the next generation of internet users.
AI Governance Multilateralism Is Fragmenting Before It Consolidates
Two parallel governance processes — the UN's new Scientific Panel on AI and the China-UK bilateral AI dialogue signalled by Foreign Secretary Cooper's Shenzhen visit — represent different theories of how international AI governance will be constructed. The UN panel, as CFR analysis notes, faces the structural tension between scientific independence and political legitimacy that has historically neutered similar bodies; it could provide a shared evidentiary baseline for policymaking, but only if major powers accept findings that constrain their own behaviour, which neither the U.S. nor China has demonstrated willingness to do. The bilateral China-UK track is softer but potentially more actionable — the article framing that dialogue as focused on 'putting people first' is aspirational language, but the diplomatic contact itself is notable given broader UK-China tensions. The risk is that governance multilateralism fragments into a set of overlapping bilateral and regional frameworks that collectively fail to address the cross-border coordination problems — AI-enabled disinformation, autonomous weapons proliferation, data flows — that require genuinely multilateral solutions.
China Is Building a Self-Reinforcing Sovereign AI Stack — Compute, Energy, Data, and Security
The convergence of separate Chinese policy announcements this week — the prefabricated green computing hub in Qingdao, the NDA's national AI data supply plan, the MSS warning against foreign model relay services, and Alibaba's internal AI consolidation — should be read as a coherent stack-building strategy rather than disconnected initiatives. Beijing is simultaneously securing the energy infrastructure for AI compute, curating the training data supply, restricting foreign model access, and consolidating domestic model development under commercially viable corporate structures. The strategic goal is a domestic AI ecosystem that is self-sufficient at each layer — not merely competitive at the model level. This is a longer-duration project than U.S. export controls were designed to address, and its success would mean that even if controls remain effective at restricting advanced chip access, China's AI development trajectory would be sustained by architectural efficiency gains and sovereign data advantages rather than raw compute parity.
Explore Other Categories
Read detailed analysis in other strategic domains