It’s clear that very soon, maybe even later this year, we’ll be producing more chips than we can turn on — except for China.
Elon Musk, at World economic forum 2026
When Musk made that remark, his core argument was that the biggest bottleneck to AI advancement is not computing hardware but a shortage of electrical power.
For the last few years, the global conversation about artificial intelligence has revolved around one issue: semiconductors. Policymakers across the world have discussed silicon supremacy in their planning documents, as though the future of the AI industry would depend solely on it. NVIDIA became the first chip company to reach $5 trillion valuation. It has now become the gold standard of the industry.
But, nowadays, there seems to be another challenge in front of AI companies. The launch of DeepSeek in 2025 has changed the assumption that was becoming conventional wisdom.
Today, nations, companies and investors are realizing that they have less need to stockpile GPUs, battle for chip supply and chase the next hardware breakthrough. DeepSeek shows that algorithmic efficiency can partially compensate for hardware constraints through optimizations.
Unlike the other AI models that use the entire model for every query, Deep Seek activates only the relevant parts, improving performance per unit of compute.
Whenever an AI model responds to a query, generates a word, answer, image, or video, it performs billions of calculations. But those calculations run on electricity. For example, one ChatGPT query used 10 times as many resources as a Google search query. Generating an AI image requires as much power as charging your smartphone.
Now multiply by hundreds of millions of queries every day, and the math will be terrifying. So, the biggest constraint in the AI industry today is not computing but power. As AI systems grow larger and more capable, electricity supply is emerging as a critical constraint, and the pressure is already showing in the United States.
Washington focused on semiconductors and Beijing on grid
In the US, data centers will account for 38% of the growth in electricity demand between 2024 and 2030, though just 6% in China, according to Bloomberg NEF projections. Data centers will command almost 7% of total US power demand by 2030, compared with 2% in China.
By 2035, US data center power demand is projected to reach 106 gigawatts. To put that in perspective, the United States operates the largest nuclear fleet on the planet. But its total output is 97 gigawatts, which still wouldn’t be enough to meet the demands. A data centre can be built in as little as 18 months, while bringing a new power supply takes three times longer. In 2024, the US built 888 miles of transmission lines, when it needs about 5000 miles every year.
In March 2025 CSIS reported that electricity supply had become the biggest bottleneck of US companies. This can be confirmed from the fact that AI data centers are driving up electricity costs in the US. In data center-heavy states like Virginia, power prices have risen by as much as 267% over the past five years. According to IEA predications, the global data-center electricity demand could rise from current estimates of 415 terawatt hours in 2024 to approx. 945 TWh by 2030, with AI being the single largest driver of this growth.
In 2024, China generated over 10,000 TWh of electricity, more than double any other country on the earth. In the next five years, China could add more than 3.4 terawatts of new generating capacity, vastly exceeding expected additions in the United States. This is equally important because AI deployment ultimately depends on access to reliable and affordable electricity. So, today, the AI race has moved from computational efficiency to energy abundance.
AI race will have a clean energy advantage
Today, in the West, government and energy companies are still focused on protecting traditional industries like oil and natural gas. China meanwhile has invested heavily in new technology such as solar panels, batteries and wind power not to reduce pollution but as industries to create job, drive innovation, and strengthen the economy. In short, Western leaders viewed solar panels, batteries, and wind turbines as climate tools; the Chinese, as industrial tools.
Today, China possesses approximately 430 gigawatts of hydropower, 550-600 GW of Wind Power, and 850-900 GW of Solar Power, apart from 1,150 GW of coal-fired generation. These clean energy assets provide China with a stable energy foundation capable of supporting any industrial expansion, electrification, or AI infrastructure challenge.
The future of AI race will not by determined by the fastest chips or the best algorithms but the one who has better power capacity to run those models. China has recognized this way before the rest of the world. Over the past two decades, the country has invested massively in clean energy and cemented its status as the world’s undisputed clean energy superpower, spending more on renewables than the rest of the world combined.
But the United States has extraordinary strengths: top universities, deep capital markets, an innovation culture, global dominance in the semiconductor industry, entrepreneurial dynamism, and the topmost AI-influential companies in the world. As the landscape evolves, AI is no longer remain a software competition but an infrastructure competition. In infrastructure, expertise and scale matter.
The upcoming decade will reveal whether the decisive factor in the AI race will be silicon or electricity. In case AI becomes a competition over electrical power, then the most important geopolitical story of the next decade may or may not be happening in Silicon Valley.
