Two Crackdowns, One Message: AI Is Now Geopolitics
In the span of a few days, US China AI tensions over artificial intelligence have escalated on two fronts. In Washington, the State Department issued a directive to diplomats worldwide, warning allies about AI technology theft and the alleged distillation of US proprietary models by Chinese firms, notably DeepSeek, Moonshot AI and MiniMax. The memo calls out what it describes as “industrial-scale” efforts to extract capabilities from American systems, and urges partners to avoid AI models derived from such practices. Beijing has rejected the accusations as baseless and insists it protects intellectual property. Almost simultaneously, Beijing’s National Development and Reform Commission shocked markets by blocking Meta’s proposed acquisition of agentic AI startup Manus, citing national rules on foreign investment and technology leakage. Together, these moves signal a hardening stance on both sides: AI is no longer just a commercial race, but a tightly controlled strategic asset.
DeepSeek and the New US Narrative on AI Technology Theft
The latest US campaign singles out DeepSeek as emblematic of rising AI geopolitical risks. According to a State Department cable, American diplomats are instructed to warn foreign counterparts about the risks of using AI models allegedly distilled from US proprietary systems. Distillation, a common technique for training smaller, cheaper models using outputs from larger ones, is portrayed by Washington as a vector for AI technology theft and for stripping away safety and neutrality features. DeepSeek has drawn attention after releasing a low-cost model and teasing a new system tailored for Huawei chips, underscoring China’s drive to reduce dependence on US hardware. The White House’s science adviser, Michael Kratsios, has framed these activities as “industrial-scale” exploitation of American innovation, promising tighter cooperation with US AI firms to detect and punish offenders. Beijing, via its embassy in Washington, firmly denies wrongdoing and decries what it calls unjustified suppression of Chinese companies.
China Blocks Meta–Manus: Guarding Agentic AI from Foreign Control
If Washington is trying to stop sensitive AI from flowing out, Beijing is now just as determined to keep it from being bought up. China’s state planner has prohibited foreign investment in Manus, effectively blocking Meta’s planned US$2 billion (approx. RM7.91 billion) acquisition of the Singapore‑incorporated startup. Manus shot to prominence after unveiling what it described as the world’s first “general AI agent”, capable of completing complex tasks on a user’s behalf. Although registered in Singapore, its core R&D teams were based in Beijing and Wuhan, and the founders only shifted headquarters and key staff to Singapore in 2025. Chinese regulators feared this restructuring would become a blueprint for shifting homegrown breakthroughs offshore to be acquired by US tech giants. The decision to unwind a deal largely completed — Manus staff had already joined Meta and investors had been paid — sends a chilling signal: Chinese AI regulation will follow the technology’s roots, not just its legal address, especially in sensitive agentic AI.
Mirror Strategies: Outbound Controls vs Inbound Barriers
Taken together, the DeepSeek accusations and the Meta Manus deal block look like mirror-image strategies. The US is focused on closing outbound channels: restricting how its AI models can be accessed, copied or distilled abroad, and warning partners against using Chinese stacks that may embed repurposed American technology. The goal is to defend an eroding performance lead and preserve control over standards, security features and downstream applications. China, by contrast, is tightening inbound barriers. By blocking Meta’s Manus acquisition and warning local AI champions off American capital, Beijing is prioritising sovereignty over integration. It aims to prevent strategic capabilities, especially in agentic AI and core algorithms, from ending up under foreign corporate control. The likely result is a more fragmented global AI landscape, where cross‑border M&A faces higher political risk, due diligence extends into export‑control and security reviews, and both sides double down on domestic champions and alternative suppliers.
Implications for Global and Malaysian Users of US or Chinese AI Stacks
For developers and enterprises in Southeast Asia and beyond, these US China AI tensions are no longer abstract. They reshape vendor risk, compliance obligations and technical strategy. Companies that rely heavily on a single US or Chinese provider now face higher exposure to sanctions, export controls, data‑access limits and sudden regulatory shocks to AI M&A. Using Chinese models accused of AI technology theft could trigger scrutiny from US partners or regulators; relying solely on US hyperscalers may invite political risk in markets closer to Beijing. Practical responses include multi‑cloud and multi‑stack architectures, clearer contractual guarantees on data handling and model provenance, and internal assessments of AI geopolitical risks. Regional firms should map which workloads truly need US or Chinese frontier models and which can be shifted to neutral or local providers. While tighter controls may slow some forms of cross‑border innovation, they will also create openings for ASEAN and other third‑country players to offer compliant, politically safer alternatives.