AI and International Competition — CHIPS Act, Export Controls and the EU AI Act── Three policy axes — semiconductor subsidies, export controls and AI regulation — are moving simultaneously across major economies, creating direct constraints on pharmaceutical supply chains and cross-border data flows

01AI industrial policy operates on three axes: semiconductor subsidies, export controls and usage regulation

In August 2022, the United States enacted the CHIPS and Science Act, committing $52.7 billion in subsidies and tax credits for semiconductor manufacturing. In October of the same year, the Bureau of Industry and Security (BIS) launched export controls on advanced semiconductors bound for China. In August 2024, the EU AI Act entered into force, requiring conformity assessments for high-risk AI systems. In May 2025, Japan enacted its AI Promotion Act.

Why are these policies moving simultaneously yet in different directions? Every major economy faces the same question: how to capture AI's economic benefits while containing national security risks. But the answers diverge depending on whether a country manufactures semiconductors, develops frontier AI models, or controls the scale and cross-border freedom of its data.

02The United States uses subsidies and export controls in tandem to reshape semiconductor supply

Of the CHIPS Act's $52.7 billion, $39 billion is earmarked for domestic fabrication facilities. By the end of 2024, major awards had been finalised: $6.6 billion for TSMC's Arizona fab, $7.86 billion for Intel, $4.7 billion for Samsung and $6.165 billion for Micron. Over $36 billion in total was confirmed by November 2025, funding multiple advanced semiconductor manufacturing sites on US soil.

The other axis is export controls. Since October 2022, BIS has progressively restricted exports of advanced AI chips and semiconductor manufacturing equipment to China. In December 2024, it added 140 entities to the Entity List and extended controls to high-bandwidth memory (HBM) and advanced packaging equipment. In January 2025, the Biden administration published an AI Diffusion Framework, but the Trump administration's BIS rescinded it in May 2025, citing concerns that the rule would hinder US innovation and complicate diplomacy with allied nations.

Subsidies pull the supply side into the United States; export controls cut the demand side away from China. The two policies are two faces of the same strategy.

03The EU leads on usage regulation while Japan prioritises adoption through a penalty-free law

The EU AI Act entered into force in August 2024 and classifies AI systems into four risk tiers. AI applications deemed to pose "unacceptable risk," such as social credit scoring, were banned from February 2025. The conformity assessment deadline for high-risk AI was originally August 2026 but was pushed to December 2027 by the Omnibus amendment approved in June 2026. Penalties reach up to EUR 35 million or 7% of global annual turnover.

Japan's AI Promotion Act took the opposite approach. Passed by the Diet on 28 May 2025 and effective from 4 June, the Act carries no penalties and prohibits no AI uses. Its purpose is to promote AI research, development and adoption. An AI Strategic Headquarters chaired by the Prime Minister was established in September 2025. In December 2025, the Cabinet approved an AI Basic Plan setting the goal of making Japan the world's "most AI-friendly country." METI allocated approximately JPY 1.23 trillion — roughly $8 billion — for AI and semiconductors in fiscal year 2026.

The EU seeks trust through regulatory precision; Japan seeks investment through regulatory lightness. Both start from the position of AI user nations rather than AI developer nations, yet their policy directions are opposite.

04China is advancing domestic chip production and data governance simultaneously under export restrictions

China is responding to US export controls on two fronts: domestic AI chip manufacturing and cross-border data regulation.

Huawei is scaling production of the Ascend 910B using SMIC's 7nm process and has announced plans to double 7nm capacity in 2026. The US Commerce Department estimated Huawei's 2025 AI chip output at approximately 200,000 units, while Huawei itself targets 600,000. Yields reportedly remain well below industry norms, but the build-out of production infrastructure is proceeding.

On cross-border data transfers, the Cyberspace Administration of China (CAC) published relaxation measures in March 2024, permitting general data to flow across borders by default. However, data from five sectors — including pharmaceuticals and AI — remains on a negative list subject to security assessments. National safety standards for cross-border data processing were published in September 2025 and take effect in March 2026.

05Pharmaceutical supply chains are subject to all three policy axes

Pharmaceutical companies use AI across target identification, clinical trial design and analysis, manufacturing process optimisation and post-market safety surveillance. These activities span multiple countries, which means all three policy axes — semiconductor subsidies, export controls and AI usage regulation — impose direct constraints.

Semiconductor subsidies affect where compute infrastructure sits. The CHIPS Act expansion of US-based manufacturing strengthens the stability of AI compute available through US cloud providers. Export controls, meanwhile, cap the performance of AI chips available at research sites in China. Companies with clinical development operations in China face limits on the scale of AI models they can run locally.

AI usage regulation shapes the documentation and conformity assessment requirements for clinical trial analyses that use AI. Medical AI classified as high-risk under the EU AI Act must have conformity assessment infrastructure in place by December 2027. Japan's AI Promotion Act imposes no equivalent obligation, which means AI tools developed in Japan require additional work before deployment in the EU.

06Cross-border data rules directly constrain AI-based analysis in multinational clinical trials

Multinational clinical trials aggregate patient data from trial sites across countries for centralised statistical analysis. AI-based analysis follows the same structure. But cross-border data transfer rules constrain that aggregation.

China's negative list subjects pharmaceutical data to security assessment. Patient data collected at Chinese trial sites cannot be sent to an overseas AI model for analysis without CAC approval. Under the GDPR, transfers of personal data outside the EU require an adequacy decision or standard contractual clauses. Japan's APPI conditions cross-border transfers on whether the destination country provides an equivalent level of protection, or on the data subject's consent.

When these regimes overlap, the most efficient approach — pooling all data in one cloud and running one AI model — becomes impractical. Federated learning, which distributes model training across sites while keeping data local, has attracted attention as an alternative. But whether regulators will accept federated learning outputs as valid statistical analysis for regulatory submissions is not yet established.

07Pharmaceutical companies respond by distributing compute and securing data transfer terms in advance

With three policy axes moving in different directions, pharmaceutical companies have three practical responses.

First, distribute compute across multiple geographies. Place AI workloads in cloud environments across the US, EU and Asia so that a policy change in one region does not halt research and development globally. The CHIPS Act expansion of US semiconductor manufacturing adds stability to US-based compute environments.

Second, secure cross-border data transfer terms at the contracting stage. For clinical trial data in China, patient data in the EU and safety data in Japan, prepare transfer mechanisms compliant with each jurisdiction's rules before trials begin. Maintain technical readiness for alternatives — federated learning, data anonymisation — in case regulations tighten.

Third, build internal capability to evaluate and select AI models. Because AI regulation differs across jurisdictions, a tool approved in one country may not be directly usable in another. The ability to assess model performance, regulatory compliance and portability in-house accelerates response to policy changes.