AI Became a "National Strategic Resource"
Until 2022, AI was a private-competition area, but post-ChatGPT, governments position "AI as a national strategic resource on par with nuclear weapons and semiconductors." Export controls, subsidies, and data-center attraction are becoming central topics of diplomatic negotiation.
US Strategy
Export Controls
- 2022: ban on H100/A100 export to China
- 2023: regulation-compliant H800/A800 also banned
- 2024: low-performance versions like H20 also targeted, HBM export also regulated
- 2025: debate on export controls for AI-model weights/services
Entity List
SMIC, Huawei, DeepSeek etc. added to the US Entity List, restricting access to US products/technology. OpenAI also restricts API access from mainland China.
Subsidies/Industrial Policy
- CHIPS Act (2022): $52B
- Stargate concept (OpenAI + Oracle + SoftBank): hundreds-of-billions-of-dollars scale
China's Response
Domestic Frontier Models
- DeepSeek V3 / R1: surprised the world with low-cost training
- Qwen 3 / 4 (Alibaba): world-top-class as open-weight
- GLM (Zhipu): enterprise-oriented
- Kimi (Moonshot): long-context specialized
- Doubao (ByteDance): consumer-oriented
Multimodal
- Kling (Kuaishou): on par with Sora in video generation
- Vidu (Shengshu)
- Hailuo (MiniMax)
Domestic Semiconductors
Huawei Ascend, Cambricon, Biren as domestic AI chips. SMIC mass-produces 7nm, developing 5nm.
Foreign-Capital Regulation/M&A Review
In April 2026, China's National Development and Reform Commission (NDRC) blocked Meta's $2B acquisition of agent-type AI startup Manus. Through a "foreign-investment security review" it ordered the deal cancelled. A new regulatory risk has emerged for foreign capital entering/acquiring Chinese AI firms.
Europe's Strategy
EU AI Act (Leading by Regulation)
The world's first comprehensive AI regulation. A "lead by values" strategy. Unable to win US-China competition on tech power, it secures influence by rule-formation.
Mistral AI
A France-origin open-weight-model provider. A symbolic EU presence.
Sovereign-AI Policy
- France: a EUR 50B AI investment plan
- Germany: investment in Aleph Alpha
- EU: GAIA-X, AI Innovation Fund
Japan's Situation
Domestic Players
- KDDI / Anthropic partnership: Claude deployment
- NTT tsuzumi: lightweight, Japanese-specialized
- SoftBank: Stargate participation
- Sakana AI: a Japan-origin research-oriented startup
- Preferred Networks: MN-Core + PLaMo model
- Japanese-specialized like ELYZA, Stockmark, Rinna
Government Strategy
- AI Promotion Bill (2025)
- METI subsidies/data-center attraction
- ABCI (AIST) GPU cluster
- Sovereign-AI debate
The Dilemma
Balancing dependence on foreign AI (OpenAI, Anthropic, Google) and domestic cultivation is the challenge. The realistic answer is a hybrid of using foreign-cloud domestic regions + domestic in specialized fields.
Other Countries
- UAE / Saudi: G42, HUMAIN, oil money into AI
- Singapore: AI Verify, becoming a data-center hub
- Korea: Naver HyperCLOVA X, Samsung Gauss
- Israel: AI21, Run:AI
- India: Krutrim, TCS's AI strategy
Latest move from Beijing: travel-permission regime for top AI researchers
In May 2026, Chinese authorities expanded a regime that requires top AI researchers — including those at private firms like Alibaba and DeepSeek — to obtain official approval before leaving the country (Bloomberg / Straits Times). The stated goals are curbing data leaks, technology transfer, and foreign poaching of names that Western players have already tried to recruit, such as former Qwen lead Junyang Lin. Coming after the NDRC blocked Meta's acquisition of Manus, this looks like "export control at the talent layer", paired with US export controls on chips and model weights. For Japanese buyers evaluating Chinese open-weight models or SaaS, the uncertainty over how internationally mobile those researchers and roadmaps can stay needs to be priced in as a fresh layer of geopolitical risk.
On 2026-07-15, Apple Intelligence was cleared for launch in China paired with Alibaba's Qwen AI (TechCrunch). A rare setup where a US consumer product enters China without any US frontier model behind it — the mirror image of US export controls (per-customer approval on GPT-5.6, the shutdown of Anthropic Fable 5 / Mythos 5 for foreign users). Companies with dominant hardware footprints in China (Apple) preserve market access by shipping on a Chinese-domestic model, materializing the US-China "decouple the models, keep the hardware tied" pattern at the product level.
Summary
AI is shifting from "free global competition" to "national-bloc competition." US-China regulatory conflict is the central axis, Europe rules, Japan and India go their own way. Companies must do cloud/model selection accounting for geopolitical risk.
June 2026 — US government tightening its hold on AI
Trump administration in talks over equity stake in OpenAI (CNBC, TechCrunch). The president has publicly floated deals "where the American people can benefit from the success of AI"; CNBC reports active discussion of an equity stake with OpenAI, with proceeds potentially seeding the Public Wealth Fund OpenAI itself recently proposed as a vehicle for distributing returns directly to citizens. Coupled with last year's 10% stake in struggling Intel, federal equity participation in strategic AI firms is moving from rhetoric into actual industrial policy.
Trump executive order accelerates AI deployment across the US military and intelligence community (IBT). The order integrates AI into defense and intelligence workflows for strategic and operational advantage, building on May's classified DoD AI program (in which OpenAI, Google, Nvidia, and xAI were selected and Anthropic was excluded on supply-chain risk grounds). Federal procurement now becomes an even more direct revenue driver for the frontier labs allowed inside.
Sriram Krishnan, White House senior AI advisor, departs at end of June (TechCrunch). Krishnan was central to the administration's AI Action Plan (data-center build-out over safety guardrails), efforts to preempt state-level AI regulation, and several AI-related executive orders. His successor will shape whether the three axes — equity stakes in AI firms, rapid military/intel AI rollout, and federal pushback on state regulation — tilt further into pure industrial policy.
Police in England and Wales told to halt AI use in court statements (Financial Times). Officers were instructed to stop using AI tools to draft court statements without proper governance and verification — a concrete reminder that Europe's "rules-first" posture is now also reaching front-line operational use, in clear contrast with the US/China "ship first" approach.
