The Three Gorges dam appears to have been vulnerable on seperate occassions. Due to excessive rainfall. Dependancy is a risk to the Chinese society. Solar and wind, gas and coal turbines and nuclear power may mitigate this risk.
You make a good point. Relying too heavily on any single source creates vulnerabilities, which is also why China has been expanding solar, wind, nuclear, gas, and coal capacity alongside hydro. The Three Gorges Dam itself is a good example of China’s willingness to invest decades ahead for long-term energy diversification and security.
On point here. The west Asia conundrum is Exhibit A in this regard. Especially, the troubling downstream effects across the world that will only worsen over the next 6 quarters, perhaps. There is much at stake, and at a civilization level for some, who can least afford the coming circumstances.
There was some early American engineering involvement decades before construction, but the actual dam was overwhelmingly a Chinese state project built by Chinese SOEs. Foreign firms supplied equipment and expertise, but calling it “an international consortium led by the United States” overstates it quite a bit.
The energy thesis is correct, and the demand-driver chart is the clearest single image of the asymmetry available in the independent press. Data centers consuming nearly the entire margin of American incremental electricity demand while Chinese industry, EVs, and manufacturing absorb far more within a vastly larger expansion cycle tells you less about AI ambition than about the substrate into which AI is being deployed. The United States is not bolting an AI economy together. It is fastening an AI annex onto a stagnant industrial base and calling it a strategy. Whether a nation can dominate a technology whose physical requirements it has structurally decided not to meet is a question neither government at the Beijing summit was asked to answer.
China did not build 3,900 gigawatts of installed capacity because its population chose to save and invest at civilizational scale. It built that capacity by designing a financial system that transferred income away from households and toward state-directed capital formation for three consecutive decades. The Three Gorges Dam is not merely the world's largest hydroelectric installation. It is a monument to suppressed wages, and the electricity it generates was paid for by people who never received the bill. That deferred invoice has arrived in the form of a property market collapse that no trade framework can reach and no summit communiqué can acknowledge. At what point does the analyst reading Chinese energy capacity as strategic advantage reckon with the savings architecture that produced it, and decide whether the asset and the liability belong on the same ledger?
The American side of the chart achieves a different species of absurdity. Oracle's remaining performance obligations reached $523 billion while its operating cash flow runs at roughly $21 billion annually against a capital expenditure requirement approaching $50 billion. The company announced a plan to raise $45 to $50 billion in calendar 2026, including an at-the-market equity program, to fund capacity it has already promised to deliver. SoftBank completed a $22.5 billion investment in OpenAI while racing to source the capital through asset sales and margin loans backed by Arm shares. When the funding structure for civilizational infrastructure depends on collateral-sensitive borrowing against assets whose value the buildout itself is inflating, how does the ledger balance if the grid delays a single major site and capital markets reprice the risk simultaneously?
There is a layer this framing does not reach. In January 2026, procurement officers from Google, Microsoft, and OpenAI were living in long-stay hotels near Samsung and SK Hynix facilities outside Seoul, waiting for allocation meetings that would determine whether their employers received sufficient memory to power the next generation of infrastructure. The Americans had crossed the Pacific. The Koreans poured tea, reviewed spreadsheets, and sorted the wounded. Three corporations control ninety-two percent of global DRAM supply and have already committed the majority of their high-bandwidth memory output through multi-year agreements. DDR5 contract prices moved 298 percent in a single quarter. The wrist tags have already been distributed, and neither flag flying over the Great Hall last week belongs to the institution holding the pen.
If the decisive constraint in the AI race is sustained electricity at national scale, and the constraint beneath that is memory allocation at oligopoly discretion, what were the two most powerful governments on earth actually deciding in Beijing — and for whom?
Renewables aren’t reliable. Not without a battery infrastructure that would cost $1.2 Trillion at least and requires REE. Natural gas we have in the world’s greatest abundance.
Nuclear we can do again- and are - and we have the world’s largest reserves of coal. We also have enormous oil reserves.
A knock on effect of Epic Fury - just luck I’m sure- is the Petrodollar is now American petroleum.
Ahem. Lucky again.
As a side note Diplomacy is not to place to seek veracity, nor political maneuvers.
But we are adding enormous power infrastructure precisely to support AI data center buildout and we need enormous power for industry across the board. That is of primary importance to the nation.
China refusing the H200 chips is the most important signal in this piece and I think it deserves to sit alongside the energy data rather than near the end. A country that turns down access to the worlds most advanced AI hardware is telling you it believes domestic alternatives are close enough that the near-term cost of delay is worth the long-term independence. Thats a confidence signal about China's semiconductor timeline. And if you combine that with the energy numbers youve laid out here, both constraints that limited China's AI ambitions, chips and electricity, resolve within the same window. The US advantage in frontier AI has been sustained by two moats simultaneously, hardware access and energy scarcity abroad. If China reaches chip parity while holding a 2.6x installed capacity advantage, both moats drain at once and the window in which the US holds meaningful AI superiority turns out to be measured in quarters rather than decades. the energy infrastructure you describe is the slow variable that was already in place. the chip refusal is the fast variable telling you the convergence is closer than Washington is pricing.
Thank you for your comment, and very good point. The H200 refusal is probably the clearest signal that Beijing believes the dependency window is closing faster than the West assumes. When you combine semiconductor convergence with China’s enormous power advantage, the question is how long the current US lead realistically will last?
Great article. Energy, not computer chips, is the real deciding factor for AI. That is probably also why 24-year-old AI prodigy Leopold Aschenbrenner (who grew $225 million to $5.5 billion in a year) is heavily invested in Bloom Energy (BE), a company that provides offline energy generators to AI companies. That investment essentially confirms your article’s point. Our grid-supplied energy (other than China’s) is not enough to support our AI centers. Thus, offline augmentation it is until we catch up.
The Three Gorges dam appears to have been vulnerable on seperate occassions. Due to excessive rainfall. Dependancy is a risk to the Chinese society. Solar and wind, gas and coal turbines and nuclear power may mitigate this risk.
You make a good point. Relying too heavily on any single source creates vulnerabilities, which is also why China has been expanding solar, wind, nuclear, gas, and coal capacity alongside hydro. The Three Gorges Dam itself is a good example of China’s willingness to invest decades ahead for long-term energy diversification and security.
On point here. The west Asia conundrum is Exhibit A in this regard. Especially, the troubling downstream effects across the world that will only worsen over the next 6 quarters, perhaps. There is much at stake, and at a civilization level for some, who can least afford the coming circumstances.
Appreciate you underscoring the impact of LT planning!,
Thank you, Cornelia. It’s especially important when it comes to energy infrastructure. I’m glad you enjoyed it!
The Three Gorges Dam was a pipe dream until it was actually constructed by an international consortium led by the United States. China took notes.
There was some early American engineering involvement decades before construction, but the actual dam was overwhelmingly a Chinese state project built by Chinese SOEs. Foreign firms supplied equipment and expertise, but calling it “an international consortium led by the United States” overstates it quite a bit.
The energy thesis is correct, and the demand-driver chart is the clearest single image of the asymmetry available in the independent press. Data centers consuming nearly the entire margin of American incremental electricity demand while Chinese industry, EVs, and manufacturing absorb far more within a vastly larger expansion cycle tells you less about AI ambition than about the substrate into which AI is being deployed. The United States is not bolting an AI economy together. It is fastening an AI annex onto a stagnant industrial base and calling it a strategy. Whether a nation can dominate a technology whose physical requirements it has structurally decided not to meet is a question neither government at the Beijing summit was asked to answer.
China did not build 3,900 gigawatts of installed capacity because its population chose to save and invest at civilizational scale. It built that capacity by designing a financial system that transferred income away from households and toward state-directed capital formation for three consecutive decades. The Three Gorges Dam is not merely the world's largest hydroelectric installation. It is a monument to suppressed wages, and the electricity it generates was paid for by people who never received the bill. That deferred invoice has arrived in the form of a property market collapse that no trade framework can reach and no summit communiqué can acknowledge. At what point does the analyst reading Chinese energy capacity as strategic advantage reckon with the savings architecture that produced it, and decide whether the asset and the liability belong on the same ledger?
The American side of the chart achieves a different species of absurdity. Oracle's remaining performance obligations reached $523 billion while its operating cash flow runs at roughly $21 billion annually against a capital expenditure requirement approaching $50 billion. The company announced a plan to raise $45 to $50 billion in calendar 2026, including an at-the-market equity program, to fund capacity it has already promised to deliver. SoftBank completed a $22.5 billion investment in OpenAI while racing to source the capital through asset sales and margin loans backed by Arm shares. When the funding structure for civilizational infrastructure depends on collateral-sensitive borrowing against assets whose value the buildout itself is inflating, how does the ledger balance if the grid delays a single major site and capital markets reprice the risk simultaneously?
There is a layer this framing does not reach. In January 2026, procurement officers from Google, Microsoft, and OpenAI were living in long-stay hotels near Samsung and SK Hynix facilities outside Seoul, waiting for allocation meetings that would determine whether their employers received sufficient memory to power the next generation of infrastructure. The Americans had crossed the Pacific. The Koreans poured tea, reviewed spreadsheets, and sorted the wounded. Three corporations control ninety-two percent of global DRAM supply and have already committed the majority of their high-bandwidth memory output through multi-year agreements. DDR5 contract prices moved 298 percent in a single quarter. The wrist tags have already been distributed, and neither flag flying over the Great Hall last week belongs to the institution holding the pen.
If the decisive constraint in the AI race is sustained electricity at national scale, and the constraint beneath that is memory allocation at oligopoly discretion, what were the two most powerful governments on earth actually deciding in Beijing — and for whom?
Renewables aren’t reliable. Not without a battery infrastructure that would cost $1.2 Trillion at least and requires REE. Natural gas we have in the world’s greatest abundance.
Nuclear we can do again- and are - and we have the world’s largest reserves of coal. We also have enormous oil reserves.
A knock on effect of Epic Fury - just luck I’m sure- is the Petrodollar is now American petroleum.
Ahem. Lucky again.
As a side note Diplomacy is not to place to seek veracity, nor political maneuvers.
Assertions do not reality make.
But we are adding enormous power infrastructure precisely to support AI data center buildout and we need enormous power for industry across the board. That is of primary importance to the nation.
And my priority in supporting AI.
China refusing the H200 chips is the most important signal in this piece and I think it deserves to sit alongside the energy data rather than near the end. A country that turns down access to the worlds most advanced AI hardware is telling you it believes domestic alternatives are close enough that the near-term cost of delay is worth the long-term independence. Thats a confidence signal about China's semiconductor timeline. And if you combine that with the energy numbers youve laid out here, both constraints that limited China's AI ambitions, chips and electricity, resolve within the same window. The US advantage in frontier AI has been sustained by two moats simultaneously, hardware access and energy scarcity abroad. If China reaches chip parity while holding a 2.6x installed capacity advantage, both moats drain at once and the window in which the US holds meaningful AI superiority turns out to be measured in quarters rather than decades. the energy infrastructure you describe is the slow variable that was already in place. the chip refusal is the fast variable telling you the convergence is closer than Washington is pricing.
Thank you for your comment, and very good point. The H200 refusal is probably the clearest signal that Beijing believes the dependency window is closing faster than the West assumes. When you combine semiconductor convergence with China’s enormous power advantage, the question is how long the current US lead realistically will last?
Excellent analysis!
Kudos.
Great article. Energy, not computer chips, is the real deciding factor for AI. That is probably also why 24-year-old AI prodigy Leopold Aschenbrenner (who grew $225 million to $5.5 billion in a year) is heavily invested in Bloom Energy (BE), a company that provides offline energy generators to AI companies. That investment essentially confirms your article’s point. Our grid-supplied energy (other than China’s) is not enough to support our AI centers. Thus, offline augmentation it is until we catch up.