lots of energy and very cheap

It seemed like the United States had the upper hand in the AI ​​race. Having the most advanced chips is undoubtedly an important asset, but there is something even more critical: having energy to power those chips. And if anyone has energy, it’s China.

master move. The control that the US exercises over NVIDIA and other advanced semiconductor manufacturers seemed to make this power a great candidate to win the AI ​​race. However, in this game of geopolitical chess, China has moved a piece that challenges that reality. The Asian giant’s strategic advantage is not in the chips, but in something more fundamental and massive: a colossal and enviable energy supply.

Lots of energy and very cheap. Between 2010 and 2024, China increased its energy production more than the rest of the world combined. Last year alone it generated more than twice as much electricity as the United States, which is saying something. That difference has made OpenAI I already spoke of the “electron gap” (electron gap), and that translates into a brutal cost advantage for data centers: while an operator in Virginia pays between 7 and 9 cents per kW/h, their Chinese counterparts pay 3 cents.

The long term works. China has shown that Your long-term strategy continues to bear fruit. In this case, this energy advantage is not an accident either, but rather the result of state planning that crystallized in the plan of 2021 known as “Data from the East, Calculation from the West”. What they did was take advantage of the vast energy resources of the country’s interior, especially in regions like Lower Mongolia, to power data centers that serve demand in the more populated eastern part of the country. What were once just steppes are now in many cases infinite wind farms and transmission lines that supply energy to more than 100 data centers in operation or under development.

Power makes up for lack of advanced chips. For Chinese companies, access to cheap energy is especially important. In fact, since you cannot match the performance of advanced chips like the H100 with your own chips, what you do is group thousands of your own less advanced chips, taking advantage of the fact that what is “left over” is energy. We have the perfect example in Huawei’s CloudMatrix 384 cluster that makes use of your Ascend chips. It consumes four times more energy, and although that would be an unsustainable waste for the US, for China it is a viable way to compete.

Satya Nadella already warned of the problem. China continues to invest in expanding its network and that electron gap can widen. Morgan Stanely predicts that around 560 billion will be spent until 2030, and Goldman Sachs affirms that in 2030 China will have 400 GW capacitytriple what global data centers will need. The room for maneuver to continue expanding that facet without problems. Meanwhile, some executives like Microsoft’s Satya Nadella already warned weeks ago that it doesn’t matter if the US has the most advanced components when there is no power for so many chips.

Time is in China’s favor. The contrast between both powers is clear. The US has the technology, but its energy expansion is hampered by bureaucracy and insufficient energy transmission capacity. This has precisely made AI companies look for chestnuts with solutions like SMRbut time is on China’s side because they continue to work tirelessly on the development of its own advanced chips of AI and manufacturing technologies latest generation. The longer that race lasts, the more opportunities there will be for the Asian giant to close the component gap.

Image | Antonio Garcia

In Xataka | In the midst of a trade war, there is a battle that China has already won: that the world depends on its new energy