AI This line’s funding logic is shifting gears.
In the first half of the year, grabbing chips meant you could lie back and win. Now, focusing only on compute power makes it very hard to capture excess returns. Meanwhile, the tech giants’ capital expenditures this year are still surging, but the money is starting to look for an exit further upstream.
The reason isn’t complicated: when chips are the bottleneck, valuation goes to compute power. But once data centers start landing one by one, the real bottlenecks become electricity and water.
Chips can be expanded with more money. Electricity and water cannot. Site selection, power-grid interconnection, and cooling-water sources are all hard constraints measured year by year.
So whoever takes over next is likely to be the “water sellers” like energy, power equipment, and cooling.
This logic is exactly the same as in mining: in many cases, the ones who ultimately make money are not the miners—they’re the ones selling electricity.
In the first half of the year, grabbing chips meant you could lie back and win. Now, focusing only on compute power makes it very hard to capture excess returns. Meanwhile, the tech giants’ capital expenditures this year are still surging, but the money is starting to look for an exit further upstream.
The reason isn’t complicated: when chips are the bottleneck, valuation goes to compute power. But once data centers start landing one by one, the real bottlenecks become electricity and water.
Chips can be expanded with more money. Electricity and water cannot. Site selection, power-grid interconnection, and cooling-water sources are all hard constraints measured year by year.
So whoever takes over next is likely to be the “water sellers” like energy, power equipment, and cooling.
This logic is exactly the same as in mining: in many cases, the ones who ultimately make money are not the miners—they’re the ones selling electricity.