• How a software bubble may accidentally help rebuild America's industrial foundations
• America over-invested in proprietary AI while under-investing in the infrastructure AI actually requires
• As intelligence becomes abundant, long-term value migrates from software to the physical infrastructure that deploys it
Soon, we can stop arguing about whether AI is a bubble because, my friends, we're about to find out.
The better question is not whether we're in a bubble, but what happens when it bursts. Will this resemble the dot-com collapse of 2000 or the financial crisis of 2008? How badly will it damage America's technological leadership — and which companies will emerge relatively unscathed?
Let's review what we know.
As with so many technology-driven bubbles before it, AI is not itself the cause of the coming collapse. The blame lies with America's Big Tech companies, which made three remarkable strategic mistakes in rapid succession. Individually, each of these missteps is a category error. Together, they form the bubble trifecta.
The first mistake was an obsessive focus on AI itself while neglecting the less glamorous, but absolutely essential, foundations upon which this revolution depends. Capital flooded into models, chips and data centers. It did not flood into power grids, water infrastructure or communications networks.
The second mistake was subtler, but ultimately even more damaging. America's leading technology companies concentrated extraordinary amounts of capital, engineering talent and political attention on proprietary frontier language models, convinced that whoever built the smartest model would command the next era of technological leadership. But the world's greatest economic opportunity is not selling intelligence. It is applying intelligence to the physical economy.
The third mistake is only now beginning to reveal itself. The investment case for proprietary Frontier Language Models rested upon a simple proposition: intelligence would remain scarce. Scarcity would create durable software moats capable of justifying unprecedented valuations and capital expenditure.
That proposition is collapsing under the weight of open-weight AI.
The extraordinary pace at which open-weight AI models — many originating in China — have improved has fundamentally altered the economics of frontier AI. Whether proprietary models ultimately retain commercial leadership is almost beside the point. The more important development is that intelligence itself is becoming a commodity.
The value migration
When scarcity disappears, value migrates.
It migrates away from software and towards the physical systems on which software depends. Electricity grids, semiconductor fabs, fiber networks, water utilities and factories cannot be open-sourced, downloaded or forked. They remain stubbornly physical assets. Telecommunications infrastructure sits at the center of this transition: every AI application, whether proprietary or open source, ultimately depends on the network.
Ironically, the more intelligence becomes commoditized, the more valuable the infrastructure required to deploy it becomes. That is the paradox almost nobody in the United States appears to recognize.
While America's AI leaders poured resources into ever more capable proprietary large language models (LLMs), much of the rest of the world pursued a very different objective: applying AI to increase productivity across manufacturing, transport, logistics, energy and critical infrastructure. Their ambition was not simply to create more intelligent software, but to build more intelligent economies.
These are fundamentally different investment propositions. Silicon Valley measures success by how intelligent AI becomes (and how much money it makes from it). Industries measure success by how much AI improves productivity.
One seeks to improve AI itself. The other seeks to improve the industries that underpin the global economy. History has generally rewarded the application of intelligence more than the pursuit of intelligence itself.
The consolation prize
That's the bad news.
The good news is that bubbles have an odd habit of leaving behind extraordinary infrastructure long after they have destroyed extraordinary amounts of capital.
The railroad mania of the 19th century bankrupted investors but left behind railroads that powered the Industrial Revolution. The dot-com bubble erased trillions of dollars in market value, yet the surplus fiber installed to carry internet traffic that barely existed quietly became the backbone of the digital economy.
The investors remember the pain. Everyone else enjoys the infrastructure.
If the AI boom achieves nothing else, it will at least force the U.S. to confront the consequences of decades of underinvestment in its industrial foundations. Faced with explosive growth in electricity demand and hyperscale data centers, the country has finally begun expanding its electrical grid, transmission networks, semiconductor manufacturing capacity and fibre infrastructure.
Ironically, this investment has been driven not by a coherent strategy for industrial renewal, but by the immediate demands of proprietary AI. America thought it was financing a software revolution. It may instead be financing the infrastructure renewal it should have begun years ago.
That is America's AI consolation prize. It is not a winner's medal. It will not erase China's lead in industrial digitalization. But if it leaves the U.S. with stronger electrical grids, greater manufacturing capacity, more fiber infrastructure and a more resilient industrial economy than it would otherwise have possessed, then at least something of lasting value will have emerged from one of the greatest episodes of technological hubris in modern history.
Ask not what the AI bubble will do to your portfolio. Ask what it will do for your country.
How bad Is bad?
These are the facts. The unknown is not whether America has inflated an AI bubble, but how severe its eventual collapse will prove to be. The answer depends largely on how the hyperscalers have financed the hundreds of billions of dollars they have poured into an extraordinarily ambitious — spectacularly ill-judged — AI data center infrastructure buildout.
That, in turn, depends on the legacy of the financial crisis in 2008. Unlike the dot-com bubble, the financial crisis left no infrastructure dividend; its supposed legacy was regulatory. New rules were designed to prevent another catastrophe built on opaque leverage, off-balance-sheet financing and complex financial engineering.
If those reforms worked, the fallout may prove painful but manageable — a repeat of 2000 rather than 2008. If they did not, America may be about to discover that 2008 was merely a rehearsal.
Note: Later this week I'll publish Part 2 of this analysis, identifying which companies and nations are most exposed to the AI bubble, which are relatively insulated, and which may ultimately benefit from its collapse. In other words, a spot of light reading for anyone with a 401(k).
Stephen M. Saunders MBE is a communications analyst and USPTO-registered inventor examining how digital infrastructure — 5G, cloud, and AI — is reshaping industry, power and society, as well as underpinning the emerging, ubiquitous global digital economy. As anchor of FNTV and a longtime industry insider, he focuses less on growth narratives and more on execution, risk and how hyperscale technology is distorting markets, governance and society at scale.
Opinion pieces from industry experts, analysts or our editorial staff do not represent the opinions of Fierce Network.