Geopolitics & Sovereign Positioning

Huawei's announcement of the Tau (τ) Scaling Law — presented by semiconductor division president He Tingbo — claims a pathway to chips equivalent to 1.4nm process performance by 2031, without requiring the extreme ultraviolet lithography equipment that ASML supplies exclusively to non-Chinese fabs. The core proposition is a paradigm shift: rather than measuring progress by transistor node size, Tau scaling prioritises system-level data throughput and interconnect efficiency. As South China Morning Post reports, Huawei frames this as a 'new principle' guiding both chip and software co-evolution — a direct response to the ceiling imposed by US-led export controls on advanced fabrication tooling.

Independent analysts are divided on whether this constitutes a genuine technical breakthrough or a rebranding exercise designed to manage market expectations and sustain investor and government confidence in Huawei's semiconductor roadmap. South China Morning Post's technical assessment notes that architectural innovation — stacking, chiplet integration, memory bandwidth optimisation — can yield meaningful real-world performance gains even without leading-edge nodes. The critical question for Western policymakers is whether the controls are delaying Huawei by years or merely redirecting Chinese semiconductor engineering toward an alternative trajectory that achieves comparable outcomes via different means. The geopolitical risk is that a successful Tau paradigm, if validated, would erode the foundational premise of US export control strategy: that denying advanced lithography equipment is sufficient to cap Chinese AI chip capability.

Chinese AI startup funding reached 110 billion yuan ($16.2 billion) in Q1 2026 alone — a 185% year-on-year increase concentrated in large language models and embodied AI, including humanoid robotics. This is not speculative venture activity: the composition of investment, skewed toward robotics and LLMs with direct defence and industrial dual-use applications, reflects state-aligned capital directing resources toward areas where self-sufficiency is deemed strategically essential. South China Morning Post attributes this surge partly to the 'growing optimism over the country's technology ecosystem' following DeepSeek's global recognition — a dynamic where Chinese model capability benchmarks have become a domestic capital mobilisation tool.

The simultaneous moves by Xiaomi — building proprietary AI chips and models for its smartphone and EV hardware ecosystem — and China's launch of a national humanoid robot digital ID registry indicate that the investment surge is being channelled into vertically integrated AI stacks, not merely application-layer software. This is the strategic pattern that most concerns Western analysts: China building end-to-end AI capability from foundation models through custom silicon to physical-world deployment in robotics and EVs, reducing each dependency layer systematically. The US export control regime targets the silicon layer effectively; it has limited reach over software development, model training methodologies, or robotics system integration.

DeepSeek's decision to make permanent a 75% price cut on V4 Pro — already ranked top globally on intelligence-per-dollar metrics — is not primarily a commercial decision. It is a market access strategy with geopolitical logic. At price points far below OpenAI and Anthropic's flagship offerings, DeepSeek becomes the default frontier AI option for governments, enterprises, and developers in cost-sensitive markets across Southeast Asia, Africa, Latin America, and South Asia. South China Morning Post confirms the model's ranking as world-leading on cost-efficiency benchmarks, which translates directly into adoption decisions by institutions without US-tier AI budgets.

The strategic implication is a bifurcation of the global AI market: US-aligned premium providers serving wealthy economies and enterprise clients with compliance requirements, and Chinese providers capturing the volume market through price leadership. This mirrors the Huawei 5G playbook in telecommunications — where cost advantages in developing markets translated into infrastructure dominance and the data flows and vendor dependencies that accompany it. For Western AI policy, the challenge is that export controls and sanctions have no reach over API pricing decisions. The only countermeasure is subsidy or differential pricing programmes for allied and developing-country markets, which the US has not yet deployed at scale.

Despite the export control regime and intensifying political rhetoric on both sides, Rest of World documents that Chinese and American AI industries remain structurally intertwined — through co-authored research, shared academic networks, diaspora talent flows, and a common technical culture built on open-source frameworks like PyTorch and Hugging Face. This entanglement is not a vulnerability to be managed so much as a structural feature of how global AI knowledge has accumulated: the foundational research base is genuinely international, making clean technological decoupling conceptually difficult and practically expensive.

The policy tension this creates is acute. The US export control apparatus is designed around hardware chokepoints — advanced chips, lithography equipment, EDA software — precisely because software and research knowledge are too diffuse to control effectively. But the Manus episode illustrates Beijing's parallel logic: blocking Meta's acquisition of a Chinese-founded AI startup suggests China is now applying the same national security framing to inbound foreign capital in AI that Washington applies to outbound technology transfer. Both governments are simultaneously maintaining research entanglement at the technical level while erecting barriers at the commercial and capital layer — a contradiction that neither side has fully resolved.