THREAT ASSESSMENT: China’s Remote Access to U.S. Cloud Compute Undermines AI Export Controls
Chinese entities are accessing U.S.-origin AI chips through third-party cloud infrastructure, bypassing export controls that restrict physical transfers. If remote compute access remains unregulated, the cost of frontier AI development could shift decisively toward entities with access to U.S. cloud ecosystems.
**Bottom Line Up Front:** The unrestricted ability of Chinese entities to access U.S.-origin advanced computing via cloud services in third-party countries and within the United States represents a critical loophole in national security policy, undermining semiconductor export controls and accelerating China’s AI capabilities—potentially eroding U.S. military, intelligence, and commercial advantages.
**Threat Identification:** Chinese state-linked firms—including Alibaba, ByteDance, Tencent, and INF Tech—are legally leasing high-end U.S.-made AI chips (e.g., Nvidia Blackwell and B200) through cloud infrastructure located in Southeast Asia and potentially within U.S. data centers. This remote access allows them to train frontier AI models without violating current export control laws, effectively circumventing restrictions on physical chip transfers^1^. At least eleven Chinese state-linked entities have exploited this channel, and U.S. hyperscalers have committed over $21 billion to AI infrastructure in Malaysia, Indonesia, and Thailand, much of which serves long-term Chinese contracts^2^.
**Probability Assessment:** High probability within 1–2 years. Cloud-based access is already occurring at scale and is expected to grow as Southeast Asia’s data center market expands to over $30 billion by 2030^3^. Without legislative intervention, this trend will accelerate, particularly as China advances domestic AI applications in military targeting, intelligence analysis, and commercial platforms. The revised Remote Access Security Act (RASA) has cleared the House but faces Senate delays, suggesting a narrow window for policy action before the capability gap widens^4^.
**Impact Analysis:** The consequences are strategic and multidimensional:
- **National Security:** Adversaries gain access to the same compute used in Pentagon AI projects, including automated targeting systems^5^.
- **Commercial:** U.S. firms’ investments in AI model development could be undercut if Chinese competitors replicate frontier models at lower cost using U.S. infrastructure^6^.
- **Diplomatic:** Unilateral U.S. cloud controls risk alienating Southeast Asian allies and incentivizing regional decoupling from U.S. tech standards^7^.
- **Intelligence:** While current access may provide U.S. surveillance opportunities under FISA, long-term reliance on this intelligence benefit risks ceding technological initiative^8^.
**Recommended Actions:**
1. Enact the revised Remote Access Security Act (RASA) with targeted scope: redefine “exports” to include remote compute access and grant BIS authority to enforce end-user verification^9^.
2. Implement threshold-based controls—apply strict know-your-customer (KYC) rules only when compute usage exceeds levels required for frontier AI training.
3. Exempt non-AI commercial workloads (e.g., streaming, SaaS) to minimize collateral impact on global cloud users.
4. Coordinate with Southeast Asian partners to establish joint compliance frameworks, reducing friction and preserving U.S. influence in regional digital ecosystems.
5. Maintain conditional licensing as a diplomatic lever by reserving the right to suspend cloud access in exchange for strategic concessions (e.g., rare earth access).
**Confidence Matrix:**
- **Threat Existence:** High confidence — Multiple verified cases (e.g., INF Tech, Tencent) and public financial commitments^10^.
- **Impact Severity:** High confidence — AI compute parity directly enables military and economic competition^11^.
- **Policy Efficacy:** Medium confidence — Enforcement feasibility depends on KYC scalability and foreign provider cooperation^12^.
- **Retaliation Risk:** High confidence — China has precedent in rare earth export restrictions and could escalate in response^13^.
---
^1^ Carnegie Endowment for International Peace, “The Geopolitical Debates Over Controlling Cloud Compute,” 2026
^2^ Ibid.
^3^ Ibid.
^4^ Ibid.
^5^ Ibid.
^6^ Ibid.
^7^ Ibid.
^8^ Ibid.
^9^ Ibid.
^10^ Ibid.
^11^ Ibid.
^12^ U.S. Government Accountability Office assessments cited in source
^13^ Carnegie Endowment for International Peace, 2026
Published June 8, 2026