China Is Winning the Race for the Engines of the Future

February 2026 will be remembered as the month the tech world woke up to a new reality. While many in the West were still debating the SaaS-to-AI shift and phase-2 of the AI bubble (from AI excitement to AI economics), China quietly and decisively ended years of US dominance in three interconnected areas that will define the next decade: humanoid robots, industrial AI, and energy.

This isn’t about a single breakthrough, but a convergence of strategic victories that reveal a fundamental shift in the global landscape. Let's break down what happened.

1. The Rise of the Humanoids

At this year's Chinese New Year Gala the real stars weren't singers or actors, but an army of humanoid robots from Unitree performing kung fu routines. For those who read my latest book, you know how I write about kung fu and (Fosbury) backflips as two metaphors to explain how China is turning the innovation model upside down.

When those two are displayed simltanously on China’s main annual televised event, watched by hundreds of millions, you can imagine my excitement. What made the moment even more striking for me was the contrast with just one year prior, when the same company's robots had shuffled through a wobbly folk dance. This time, the precision, complexity, and fluidity of their movements were of a completely different order, nearly indistinguishable from the human martial artists performing alongside them. The message was clear: China is no longer a follower in robotics. It is poised to become the global leader.

While Elon Musk now is shifting his attention from Tesla to the Optimus humanoid, he no longer enjoys the massive head start he had with Electrical Vehicles. The race for embodied AI is on, and China is the dominant force. In China, domestic robotics and EV companies are all investing heavily in humanoid robotics, creating a powerful ecosystem of innovation and manufacturing.

In 2025, China accounted for over 85% of the 15,000 humanoid robot installations worldwide, with Tesla accounting for just 4.7%. This dominance extends to the entire supply chain, with China controlling an estimated 63% of the global supply chain for humanoid robotics, from high-torque actuators to the rare earth magnets that power them. This vertical integration allows Chinese companies to produce robots at a fraction of the cost of their Western counterparts. Unitree, one of the stars of the CNY Gala, advertises its G1 humanoid for just $13,500 on its website. Optimus is expected to sell between $20,000 and $30,000 and mass availability is likely to be 2027: that is at twice the price and at least two years behind Chinese vendors.

This raises a critical question: can the US still lead in embodied AI when it has already lost the manufacturing and supply chain advantage? I believe the answer is no. For the first time in his career, Elon Musk finds himself in the unfamiliar position of trying to catch up rather than defining the race.

But what if the US wins the AI race and uses that lead to dominate embodied AI? The token war suggests otherwise.

2. China’s AI Efficiency Trumps US Scale

The second major shift in February was the revelation that Chinese AI models, for the first time since ChatGPT came to live, surpassed their US counterparts in a key metric: token usage. The OpenRouter leaderboard, which tracks the most popular models on the platform, tells a stunning story. As of late February 2026, Chinese AI models hold four of the top five spots, dominating token consumption. This is the real trend few are watching. Combine that fact with the performance of Chinese open-source models that now match or even exceed US models in many domains, and one starts to wonder why so few are paying attention to China’s AI global adoption race versus America's AI global infrastructure investment race. It’s almost ironic to notice that in order to win the global AI market, the US is building massive infrastructure, while China is mainly building the best of applications. The world upside down.

Source: OpenRouter, Feb 2026 - monthly view

Chinese models now account for 61% of token consumption among the platform's top ten models. How is this possible when US Big Tech outspends China by 3.5 to 1 on both CapEx and cloud revenue? The answer lies in a staggering divergence between scale and efficiency.

As the chart above shows, while the US maintains a 3.5x lead in spending, China has seen a 300x surge in daily token consumption in just 18 months. My favorite Chinese tech blogger Poe Zhao from Hello China Tech reveals what is going on. This is the result of a relentless focus on efficiency. Chinese models like DeepSeek use a MoE architecture to deliver performance at a fraction of the cost — roughly 5% of OpenAI's o1 model. This has given rise to a phenomenon called "Token Export": developers worldwide are rerouting their AI workloads to Chinese data centers, bypassing US services to take advantage of the dramatic cost savings.

This is where the true nature of the AI race is revealed. Today, 84% of the world's 8.1 billion people have never used AI. While most of us feel that with OpenAI's close to 1 billion users that the US has monopolised AI, but reality is that the commercial (read paid services) market is almost entirely untapped. Only 0,3% of the world’s population is actually paying the 20 USD to get access to AI. China's strategy is different: high-efficiency, low-cost, open-source models that are built for solid real-world applicability are perfectly positioned to capture the vast, still emerging global AI market. Is this the end of Silicon Valley’s playbook: conquer the Western markets first and others will follow soon after?

Source: Design Bootcamp/Medium, Feb. 2026

But won't the whole world ultimately run any efficient, open-source models on US hyperscalers? Won't the best American open-source models match the Chinese ones anyway, given the capital, computing power, and talent available? Perhaps. But here is the uncomfortable truth: if the world runs its AI workloads on whoever offers the cheapest tokens, the winner is not only determined by model quality. It is primarily determined by the cost of electricity. Token globalization is, at its core, electricity globalization. The ability to generate and process AI tokens at scale and low cost is directly tied to a nation's energy capacity. And on that front, the numbers speak for themselves.

3. The Energy Race

This is where the third, and perhaps most decisive, of China’s victories comes into play. In 2025, China invested $500 billion in its domestic energy system, adding a staggering 540 GW of new power capacity. That is more than eight times the mere 63 GW added by the United States.

Of those 540 GW, over 430 GW came from wind and solar alone. This is energy that is cheap, clean, and effectively inexhaustible, in stark contrast to the gas-fired and coal-dependent grid that still powers much of the United States. And while the US still operates the world's largest nuclear fleet today, China is building 29 new reactors to be on track to surpass US nuclear capacity by 2030. This massive and growing energy advantage provides the foundation for China’s dominance in both AI and robotics. Abundant energy at a cost roughly 40% lower than in the US allows China to power the vast data centers needed for AI training and inference. More info in the first topic of this newsletter (below).

In March 2026, China will unveil its 15th Five-Year Plan, widely expected to double down on clean energy targets, accelerate nuclear deployment, and embed AI-driven efficiency across its industrial base. I can’t wait to write next month’s newsletter.

The New Reality

February 2026 was a wake-up call. The victories in humanoid robots, AI tokens, and energy are not isolated events. They are three expressions of the same underlying strategy: build the engines of growth first, dominate the supply chain, lower the cost, and let the global markets come to you.