Who’s winning the AI race?
A look at why frontier models may not decide the victor in this race – ecosystems and energy will.
Duration: 5 Mins
Date: Mar 18, 2026
The US leads in cutting-edge artificial intelligence (AI) models and private investment, but China's strengths in energy infrastructure, open-source development, manufacturing integration, and data policy suggest the race's outcome depends less on today's leaderboard than on which nation builds the most resilient and scalable ecosystem.
AI has the potential to rewire economic growth and reshape geopolitics, earning comparisons to the nuclear and space races that defined the 20th century. While the US may hold the lead today, this is a marathon, not a sprint – and the finish line remains distant.
Who's winning the US-China AI race?
Are frontier models a distraction? In our latest Macro Bytes, we dig deeper into why energy, infrastructure, and ecosystems – not just algorithms – may decide who ultimately crosses the line first.
Models and CapEx
US dominance … for now
US companies occupy the frontier of large language models (LLMs), and export controls on advanced chips are designed to keep them there. American tech giants have spent more than $865 billion on AI infrastructure since 2022, with another $610 billion planned for 2026 alone.1 By comparison, Chinese tech firms have invested roughly $67 billion over the same period (Chart 1).1
Chart 1. US tech firm spending dwarfs that of Chinese firms
Yet these figures may understate China's true spend. Unlike the US, where hyperscalers like Microsoft and Google drive capital expenditure (CapEx), China's buildout is distributed across state-backed telecoms and a broader array of firms, making the total harder to track.
US firms control an estimated 70% of global AI compute; Chinese companies hold just 10%.2 Export restrictions on Nvidia's most advanced chips reinforce this gap, though reports suggest Chinese firms – including DeepSeek – have trained models on smuggled or repurposed top-tier hardware.
But while US companies race toward artificial general intelligence (AGI) with closed models, China's open-source LLMs are closing the gap (Table 1) and may achieve wider global adoption, especially in markets where cost and accessibility matter more than cutting-edge performance.
Table 1. The US and China dominate LLMs, judged from reasoning, knowledge, math, and coding tests
China's electro-state advantage
Data centers are ravenous for electricity. The International Energy Agency estimates US data centers consumed 183 terawatt-hours in 2024 – over 4% of total US consumption – and McKinsey projects that figure could exceed 600 TWh by 2030.3,4
Electricity prices are rising (up 6.3% year-on-year in January 2026), and public backlash is building.5 More than half of surveyed Americans blame data centers for higher power costs, and President Trump has warned on Truth Social that tech companies "must pay their own way."6
After two decades of minimal growth, US power consumption is hitting an inflection point, with demand growth expected to nearly double to 2% annually by 2030, driven largely by data centers.3
China, by contrast, is rapidly outbuilding demand. Since 2022, it has added 1,100 GW of new grid capacity – nearly the size of the entire US grid – with 420 GW added in 2025 alone, much of it solar and wind (Chart 2).7
Chart 2. China has rapidly increased its grid capacity
This expansion positions China to become the world's first electro-state, cementing a structural cost advantage (Chart 3): Chinese industrial electricity prices are roughly half those paid by large US firms.
Chart 3. Chinese firms have a substantial energy cost advantage over their US counterparts
If energy becomes the bottleneck – as Nvidia CEO Jensen Huang has warned – China’s edge could prove decisive.8
Chips and workarounds
A five-to-seven-year gap
China's semiconductor industry remains five to seven years behind the US, despite heavy subsidies and encouragement from Beijing.9 The Trump administration's recent decision to allow Nvidia to sell its H200 chip to China offers only marginal help; the H200 is already two generations behind Nvidia's cutting edge Rubin architecture.
Yet export controls have proven less effective than intended. Reports indicate DeepSeek trained its latest model using advanced Nvidia chips, likely acquired through gray-market channels. And Chinese companies are innovating around the restrictions: Huawei's strategy of clustering millions of lower-capacity chips – dubbed "swarms beat the titan" – has mitigated some of the technological disadvantage.10
Still, China is unlikely to manufacture frontier-level semiconductors domestically anytime soon, limiting its ability to compete at the absolute cutting edge of model training.
China's AI+ strategy
Integration over innovation
While US firms chase AGI, China is pursuing a fundamentally different strategy. Its AI+ framework focuses on deploying current AI tools across its vast manufacturing ecosystem to drive productivity gains today, rather than betting on a distant technological breakthrough.
This approach leverages China's strengths: a dominant position in global manufacturing (exporting far more than the US), more industrial robots per worker, and aggressive integration of AI into future industries like robotics, biomanufacturing, and quantum computing.
Beijing is also leading on data policy. Its 2021 Data Security Law and Personal Information Protection Law have created frameworks for standardizing, sharing, and trading data – including public datasets and government-backed marketplaces.11 While China's regulations have similarities to the EU's GDPR, the country remains less constrained by privacy concerns than the US, potentially accelerating development of world models that use vast datasets to train AI for autonomous driving, robotics, and physical-world applications.
The Chinese Communist Party has streamlined model approval processes, and trusted firms no longer face thousands of political safety checks during training. Meanwhile, state financing through the People's Bank of China's relending programs ensures Chinese firms receive far larger financial subsidies than their US counterparts.
AI as a strategic lever
The Trump administration's AI action plan emphasizes cutting regulatory barriers, building energy and computing infrastructure, and exporting American AI globally – including integrating it across defense and federal agencies. Interestingly, the plan also encourages a shift toward open-source models for easier adoption, at odds with the closed-model focus of most US tech giants.
The US is also wielding AI as a geopolitical lever, particularly in the Gulf. Countries in the Middle East and North Africa have been required to dump Chinese technology in exchange for access to frontier US AI capabilities, with large sovereign wealth funds and abundant energy positioning the region as a key US technology partner.
Yet not all US policy points in the same direction. The administration's culture war on elite universities, a new $100,000 price tag on H-1B visas, and political pressure to lower electricity costs ahead of mid-term elections all risk constraining the talent pipeline and energy supply necessary for continued AI leadership.
Who's ahead? Depends how you score
Judging the AI race in real time is difficult. The US leads on frontier model performance and private-sector investment. China leads on energy cost and capacity, open-source model development, manufacturing integration, robot density, STEM PhD output, and public enthusiasm for AI adoption.
In the most extreme scenario, achieving AGI could lock in enduring economic and military advantages. But if the race is won not by a single breakthrough but by the scale and resilience of the ecosystem – spanning energy, data, manufacturing, talent, and adoption – then the outcome remains far from certain.
