Take a step back. Strip away the tickers, and the filings describe the fourth great network buildout of the industrial era. Railroads for the movement of goods, electrification for power, fiber for information. Each consumed a scandalous share of national capital.104 Each was called a bubble mid-construction, sometimes correctly as measured by the builders’ equity and never correctly as measured by the infrastructure’s use. And each forced finance to change its instruments: railroad accounting invented depreciation, utility regulation invented the rate base.105 The instruments always change last, after the steel is already in the ground.
But this time the substrate is computation, and the difference from every predecessor is that it is being built pre-sold, by the most profitable companies in history, off order books visible in audited filings. If Part I’s “line” continues even in decelerated form, the defining corporate species of 2030 is something markets have never priced: a utility’s asset base, a railroad’s capital intensity, software margins on the services running above the metal. There is no comp for that, and no historical multiple to reach for. I know the market will construct one, because it always does, and the repricing has visibly begun at the edges, in credit desks running EV-per-megawatt math and in “contracted gigawatts” migrating quietly into sell-side models.106 The market is not stupid. It is mid-realization. This essay is simply the argument for finishing the thought.
The metamorphosis is silent because it speaks in watts, backlogs, and depreciation schedules, a language the ticker does not display. But the filings are public, the contracts are signed, the reactors are restarting, and the staircase keeps adding steps. The instruments will catch up. They always do, one buildout too late for the builders and right on time for whoever read the gigawatts first.
I believe at the end of this, common intelligence will exist. Common intelligence will be a utility that flows through every household, airport, hospital, factory, university, bank, power grid, government, military and so on. Just like how water, electricity, and internet flows through each of those.
Common intelligence will be embedded in global infrastructure. It will simply become so ubiquitous that we will stop thinking about it as AI and instead infrastructure. The significance of this isn’t that we achieve AGI or Superintelligence. It is that common intelligence becomes a fundamental input to economic activity and output. Just like railroads did, just like electricity did, and just like the internet did. Electricity and the internet transformed the world because it did not conform to a single machine, it was an input to every machine.
I believe that intelligence will follow a similar path. And when intelligence becomes a utility, the infrastructure required to produce, transmit, and deliver it becomes equally fundamental. Compute, power, networks, memory, and data centers cease to be merely technology infrastructure; they become the physical foundation of an economy of common intelligence. Thank you for reading.
Notes
- Railroads absorbed roughly 15–20% of United States gross capital formation in peak decades of the nineteenth century. See Robert Fogel, Railroads and American Economic Growth (Baltimore: Johns Hopkins Press, 1964), and Alfred D. Chandler Jr., The Visible Hand (Cambridge, MA: Harvard University Press, 1977). ↩
- Depreciation accounting was formalised through United States railroad practice and the Interstate Commerce Commission’s uniform systems of accounts from 1907. The utility rate base derives from Smyth v. Ames, 169 U.S. 466 (1898), and was recast on an “end result” basis in Federal Power Commission v. Hope Natural Gas Co., 320 U.S. 591 (1944). ↩
- Enterprise value per contracted megawatt and “contracted gigawatts” now appear routinely in sell-side and credit work on CoreWeave, Nebius and Oracle. Goldman Sachs’ own AI capital expenditure model is constructed bottom-up from gigawatts, dollars per kW for new power and dollars per MW for data centres rather than from revenue multiples. ↩
Thank you to Dr. Anthony Watson of the University of Cambridge and The Kinkaid School for the continuous support with this paper and with my goals, and for being a formative figure in my life.
Thank you to everyone who has supported and helped me throughout this process. Special thanks to Mr. Arvind Khattar for his invaluable guidance and encouragement.
And to my friends, Martin Lim, Jonathan Quek, and Skye Flecker — thank you for your support, conversations, and friendship along the way.
This work is dedicated to Mr. Robert Ludwig, whose guidance, generosity, and belief in me have shaped this work and my thinking.