A structural shift in the economics of artificial intelligence is under way. China is no longer simply catching up with the West — it is redefining the competitive framework entirely. For LGPS trustees and investment officers still holding underweight positions in Chinese equities, this is not a story to monitor from the sidelines argues Steven Luk, CEO of Fountain Cap.
The race has changed — and China is winning on efficiency
Global AI competition is no longer converging toward a single frontier. It is bifurcating — and the Chinese system is emerging as the more scalable and commercially viable model. The performance gap between US and Chinese frontier AI models has narrowed to approximately 2.7% in 2025 according to Stanford’s 2026 AI Index, despite a staggering disparity in capital deployed: the United States invested $285.9 billion versus China’s $12.4 billion.
That asymmetry is the point. China is not winning by outspending. It is winning by out-optimising. The competitive variable in AI is shifting from raw model size to tokens per watt and cost per unit of intelligence — metrics on which China’s structural advantages in power infrastructure, manufacturing integration, and end-to-end system design are increasingly decisive.
On real-world coding benchmarks (SWE-Bench Verified, the industry’s gold standard), Chinese models achieve 80.2% versus the US benchmark of 80.8% — near-parity performance at roughly one-tenth to one-twentieth the inference cost: approximately $0.30 per million input tokens versus $5 for comparable US models.
Scale and adoption: a compounding structural advantage
What makes this structural rather than cyclical is the scale of real-world deployment. Chinese AI models are already processing more than 5.2 trillion tokens per week, vs ca. 3 trillion for US models, reflecting significantly higher workload intensity and commercial penetration.
Open-model platforms have generated more than 100,000 derivative models, and policy-driven rollout programmes are embedding AI across tens of thousands of Chinese enterprises. This creates a compounding feedback loop: scale drives usage, usage generates proprietary data, and data improves models. The result is a self-reinforcing system that is advancing technologically whilst simultaneously widening its cost advantage.
US vs China AI: Key Metrics at a Glance
| Metric | United States | China |
| AI Investment (2025) | $285.9bn | $12.4bn |
| Frontier performance gap | Leading | ~2.7% behind |
| Inference cost (per 1M tokens) | ~$5 (Claude Opus) | ~$0.30 (MiniMax / GLM) |
| Coding benchmark (SWE-Bench) | 80.8% | 80.2% |
| Weekly token usage | ~2.9T | 4.1T → 5.2T |
| Open ecosystem derivatives | Centralised | 100,000+ models |
Source: FountainCap Research & Investment, May 2026
The LGPS underallocation risk: tactical and strategic
For LGPS funds, the underallocation to China represents a dual risk — both tactical and strategic.
Tactically, most UK pension funds remain significantly below common emerging-market benchmark weights for China. As Chinese equities reprice to reflect genuine AI earnings power and improving domestic fundamentals, the opportunity cost of maintaining low allocations will compound. Funds invested in global tech and AI-linked equities are already exposed to the risk that US valuations — built on the assumption of sustained AI leadership — are vulnerable to compression as the cost-performance argument tilts eastward.
Strategically, the more consequential risk is missing the long-term structural repositioning. China is not simply a lower-cost provider of today’s AI technology. It is building the infrastructure — physical, institutional, and ecosystem — to become the default provider of applied AI across large parts of the global economy. Open platforms, policy-driven enterprise rollouts, and a self-improving data flywheel are creating durable competitive moats that are distinctly different in character from those underpinning current US AI valuations.
A question of reframing, not risk-taking
The instinct to treat China as a risk factor rather than an opportunity is understandable given geopolitical headlines. But framing is everything in long-term portfolio construction.
U.S. – China relations is less the “Thucydides Trap” of inevitable confrontation as mentioned by President Xi during his meeting with President Trump earlier in the year, and more the “Kindleberger Trap” of partial leadership vacuums — a world in which China acts as a geopolitical linchpin, facilitating dialogue and anchoring economic relationships, rather than triggering abrupt confrontation.
In that environment, Chinese markets offer something increasingly rare: a combination of structural growth, valuation discipline, and genuine innovation at scale.
For LGPS funds reviewing their strategic asset allocation, the China question is no longer simply about emerging-market diversification. It is about whether your portfolio is positioned for where the AI economy is heading, not where it has been.
