My piece on AI and energy, entitled "The Power and the Glory", got a lot of great feedback. It also predated DeepSeek's big reveal. So here are a few supplementary thoughts, two months later.
Given that very few predicted in 2019, that AI would be here in 2025, do not expect consensus will be right on where AI will be by middle of next decade, including energy requirements.
Microsoft's total capital may be over $500 billion, but it's net equity is only half of that amount.
Entering into capital leases does not go against the entire balance sheet, only against the equity. In essence, a capital lease is a form of debt. The investors in that debt are not interested in "how much you are worth" without first subtracting your debt. And, in the case of Microsoft, those liabilities (of various kinds) include $243 billion that DOES show up on its balance sheet PLUS the present value of any operating leases (for buildings, equipment, and perhaps Nvidia chips).
The operating leases are not disclosed on the balance sheet; one needs to read all the footnotes in the 10-K to figure those out.
All of which supports Michael's conclusion, that these companies are not going to invest TRILLIONS in new AI data centers. They will invest hundreds of billions of dollars, and that WILL result in increased electricity consumption. In some balancing areas, that will be very challenging -- Virginia at the top of the news.
BUT, even an increase from 4% to 9% of US power consumption is only the amount of load growth that the industry kept up with EVERY YEAR in the 1960's and early 70's.
Perhaps Michael's most important point, in that historical context, is that the supply chains to support electric utility growth are not in place. The generation supply chain is fine -- the world will install 10X the renewable generation required by new AI loads each year. We need those renewables to back out fossil, but generation is not the challenge. Transmission and distribution supply chains, however, are a mess and do need to grow back to previous capacities.
Given that very few predicted in 2019, that AI would be here in 2025, do not expect consensus will be right on where AI will be by middle of next decade, including energy requirements.
Fast changing world.
Microsoft's total capital may be over $500 billion, but it's net equity is only half of that amount.
Entering into capital leases does not go against the entire balance sheet, only against the equity. In essence, a capital lease is a form of debt. The investors in that debt are not interested in "how much you are worth" without first subtracting your debt. And, in the case of Microsoft, those liabilities (of various kinds) include $243 billion that DOES show up on its balance sheet PLUS the present value of any operating leases (for buildings, equipment, and perhaps Nvidia chips).
The operating leases are not disclosed on the balance sheet; one needs to read all the footnotes in the 10-K to figure those out.
All of which supports Michael's conclusion, that these companies are not going to invest TRILLIONS in new AI data centers. They will invest hundreds of billions of dollars, and that WILL result in increased electricity consumption. In some balancing areas, that will be very challenging -- Virginia at the top of the news.
BUT, even an increase from 4% to 9% of US power consumption is only the amount of load growth that the industry kept up with EVERY YEAR in the 1960's and early 70's.
Perhaps Michael's most important point, in that historical context, is that the supply chains to support electric utility growth are not in place. The generation supply chain is fine -- the world will install 10X the renewable generation required by new AI loads each year. We need those renewables to back out fossil, but generation is not the challenge. Transmission and distribution supply chains, however, are a mess and do need to grow back to previous capacities.