A deal unwound, a message delivered
China’s latest order forcing Meta to unwind its acquisition of Manus shows its intent to firewall its cutting-edge AI tech. Such post-deal reversals are rare, which makes this intervention particularly striking. By targeting a completed transaction, Chinese authorities are making it clear that cross-border deals in sensitive sectors like AI will face scrutiny even after the fact.
The Manus case is revealing on multiple fronts. The startup had already begun shifting its operational base to Singapore, a move designed to attract foreign capital while sidestepping domestic restrictions. That strategy has now been explicitly challenged. The message is that relocating legal structures or headquarters will not shield companies from oversight if their underlying assets are deemed strategically important.
The restrictions placed on Manus executives from leaving China further underline the seriousness of the response. Combined with the unwinding order, they reflect a broader effort to retain control over AI talent and intellectual property. For China, the concern is about preventing a transfer of capabilities that could strengthen foreign competitors.
Meta acquired Manus to boost its capabilities in AI agents, tools that can execute more complex tasks than chatbots with minimal human intervention, as per a Reuters report. Manus was hailed early last year by state media and commentators as China’s next DeepSeek after releasing what it said was the world’s first general AI agent. Manus does not produce its own AI model, but an agent framework that operates on top of existing Western large-language models. Alfredo Montufar-Helu, a managing director at Ankura China Advisors, told Reuters that AI has become central to strategic competition between the world’s two largest economies, with controls that were once focused on semiconductors now extending into AI. “China is saying we will prevent foreign acquisition of assets we consider important for national security — and AI is now clearly one of them,” he said, adding that it also signals to firms that relocating overseas will not prevent scrutiny.
Also Read | In blocking Meta-Manus’ $2 billion AI deal, China is trying to stem brain drain
The US escalates beyond chips
On the other side, the US is widening the scope of its defensive measures. A recent White House policy memo shows that model distillation, a technique used to replicate or approximate the performance of advanced AI systems, may soon be treated as a form of intellectual property theft. This marks a shift from hardware-focused controls toward a more expansive view of technological protection.
The implications are significant. Export restrictions on advanced semiconductors were once the central tool for maintaining an edge. Now, the focus is expanding to include algorithms, data access, and even how models interact through APIs. Intelligence-sharing with private firms and the possibility of blacklisting entities suspected of extracting capabilities point to a more coordinated and aggressive posture.
Uncertainty around chip exports adds another layer of tension. Although there was earlier approval for certain high-end chip sales to China, shipments have yet to materialise. This ambiguity reflects the balancing act between commercial interests and national security concerns, a tension that is becoming harder to manage as competition intensifies.
Also Read | China to curb US investment in tech companies: Report
DeepSeek and the collapse of old assumptions
Much of the current anxiety in the US stems from a deeper shift in the economics of AI development. China’s progress, exemplified by systems like DeepSeek, has challenged the assumption that only massive capital expenditure and access to cutting-edge hardware can produce frontier models.
The ability to achieve competitive performance at lower cost has strategic implications. It weakens the effectiveness of export controls that rely on limiting access to expensive infrastructure. If advanced capabilities can be replicated or approximated efficiently, the barriers to entry become less formidable.
This is why techniques like distillation have become so contentious. For US firms, they represent a potential pathway for rivals to close the gap. For Chinese developers, they are a practical response to constraints in hardware access. The disagreement is not just technical, it is fundamentally about what constitutes fair competition in a rapidly evolving field.
Narrowing gaps, shifting strengths
Despite the escalating tensions, the balance of capabilities remains complex. The US still leads in frontier model performance, particularly in reasoning and multimodal tasks. However, the gap is shrinking. Benchmark comparisons show that differences in performance between leading systems are now marginal rather than overwhelming. At the same time, China holds advantages in scale.
The recent Stanford University Institute for Human-Centered Artificial Intelligence 2026 AI Index report confirms that the US still leads in frontier model performance, particularly in reasoning-intensive benchmarks and multimodal systems. However, the gap has narrowed significantly in areas like natural language understanding and code generation. The gap between the top AI bots in the US and China is shrinking in Arena scores, a metric that shows relative performances of large language models. In May 2023, the U.S.’s top model, OpenAI’s GPT-4, led with more than 1,300 Arena points compared with China’s fewer than 1,000. By March 2026, the gap shrank from more than 300 to just 39 points. Anthropic’s Claude Opus 4.6 was leading China’s Dola-Seed 2.0 by just 2.7% points.
More striking is China’s dominance in scale metrics. It produces a larger share of global AI publications and patents, indicating both breadth and depth of research activity. In industrial deployment, China leads in robotics integration, particularly in manufacturing and logistics. The report also highlights a surge in domestic AI startups, supported by state funding and regional innovation clusters.
Yet the data also reveals constraints. Chinese models tend to lag slightly in reliability and alignment, partly due to stricter content controls that limit training diversity. Access to cutting-edge chips remains uneven, forcing firms to innovate around hardware bottlenecks. This is where techniques like distillation and optimisation become critical for the Chinese.
Talent and energy, the hidden battlegrounds
Beyond algorithms and policy, two structural factors are becoming decisive variables: talent and energy. The US continues to benefit from its ability to attract global expertise through universities and a dynamic innovation ecosystem. However, China is rapidly expanding its domestic pipeline of AI engineers while encouraging overseas researchers to return.
The scale of this effort matters. AI development is no longer a boutique activity led by small elite teams. It requires large, coordinated workforces capable of iterating quickly and deploying systems at scale. In this context, the ability to mobilise talent domestically is as important as attracting it internationally.
Energy is emerging as an equally critical constraint. Advanced AI systems demand enormous computational resources, translating into significant electricity consumption. The US faces bottlenecks related to grid capacity and regulatory hurdles for expanding data centre infrastructure. China’s ability to build and scale energy systems more rapidly gives it a potential long-term advantage that is often overlooked.
From competition to confrontation
These developments point to a fundamental shift in the nature of the AI race. It is no longer just about who can build the best model. It is about who controls the ecosystems that make those models possible, from chips and data to talent and infrastructure.
The Manus case and the evolving US policy response illustrate how economic activity is being reshaped by strategic concerns. Corporate decisions, investment flows and even research techniques are increasingly subject to geopolitical scrutiny. The boundary between commercial competition and national security is dissolving.
The phrase “AI war” may still be metaphorical, but the actions on both sides are becoming more direct and consequential. As the gap in capabilities narrows and the stakes rise, the likelihood of further escalation grows. What lies ahead is a struggle to define the rules of the next era of global power.
