They Buried a Trillion Dollars Underground. Then the Internet Was Born.
Part Two of the Great Buildouts
The Great Buildouts Series | Fiber Optics (1996–2010)
We previously covered the Great Railroad Wars (linked below) — the original infrastructure mania, the template for every bubble, crash, and second-owner windfall that has followed. This is Part II: the fiber optic buildout of the late 1990s, the closest modern parallel to what is happening right now in AI infrastructure.
In the late 1990s, the world decided the internet was going to change everything.
That part was true.
Then Wall Street (i.e., capital markets) and the telecom industry decided that meant we needed to build the entire future immediately — and finance it with the most aggressive debt structures the telecom industry had ever seen.
That part may or may not have been true (it wasn’t true in a “good” sense for the participants).
But it didn’t matter, because the capital was already raised, the bonds were already sold, and the deployment pressure was already unbearable. If you’ve read Part I of this series — the Railroad Wars — you already know the mechanism. The combustion engine is the same every time: a real technology, a policy catalyst, easy money, and a competitive dynamic that transforms formerly logical and methodical thinkers into crowd-following junkies.
Between 1996 and 2001, telecom companies buried tens of millions of miles of fiber optic cable across North America and Europe. The numbers are absolutely wild in hindsight: roughly 80 million miles of fiber, and close to $1 trillion of capital spent when you add up the network buildouts, the equipment, the acquisition spree, and the debt-financed balance-sheet leverage that made it all possible.
Most of the companies that built it went bankrupt.
And then the fiber sat there — dark, unused, waiting for a world that wouldn’t arrive for another five years. Literally. Between 1999 and 2004, it is estimated that less than 5% of the cable laid was “lit” - i.e., connected and carrying actual data.
The part most people get wrong is what happened next.
And it’s remarkable, my friends, how often history repeats. It’s not like that cable didn’t get put to use (eventually). It did. The companies that splurged and went wild on the buildout didn’t end up doing well. But somebody did.
As always, in a timeless pattern, the physical infrastructure did not fail. The capital structure did.
The Telecommunications Act and the Arms Race
Every infrastructure mania requires a policy catalyst. For the railroads, it was the land grants — 170 million acres of public domain handed to private companies between 1850 and 1871. For fiber, it was a single piece of legislation: the Telecommunications Act of 1996.
The Act was the most significant overhaul of American telecommunications law in sixty-two years. Its stated purpose was to promote competition by deregulating large parts of the industry — allowing local phone companies, long-distance carriers, cable operators, and new entrants to compete freely across previously protected markets.
But the practical effect of the law was to promote an arms race.
Before 1996, the American telecom landscape was a regulated oligopoly. AT&T had been broken up in 1984, producing the Baby Bells — regional monopolies that controlled local phone service — and a long-distance market dominated by AT&T, MCI, and Sprint. The system was stable, profitable, and slow.
The Telecom Act blew the doors open. Suddenly, anyone with capital and a business plan could enter any market. The Baby Bells could offer long distance. Long-distance carriers could offer local service. Cable companies could offer telephony. And a new generation of companies — the competitive local exchange carriers, or CLECs — could build their own networks from scratch and compete with everyone.
The logic was identical to the railroad logic of the 1860s: if the market is going to be open, the winner will be whoever builds the most network the fastest. First mover advantage. Land grab. Get the fiber in the ground before your competitors do, because once the routes are built, the economics of duplicating them become prohibitive.
This was not irrational. It was, in fact, strategically sound — for any individual actor. The problem, as with the railroads, was that when every rational actor reaches the same conclusion simultaneously, the collective outcome is massive overbuilding. Three competing railroads across the same Kansas prairie. Five fiber networks running through the same metro conduit. Same mechanism. Same result.
II. The Builders
The companies that built the fiber network of the late 1990s were big-league players, in general. They were serious operators with serious engineers and, in many cases, serious revenue. That’s what makes the story instructive — this was not a dot-com fantasy. This was physical infrastructure, built by companies that understood the technology and believed, correctly, that bandwidth demand would grow for decades.
WorldCom was the colossus. Founded in 1983 as a small long-distance reseller in Mississippi, it had grown through a relentless acquisition strategy orchestrated by CEO Bernie Ebbers — a former basketball coach and motel operator who understood deal mechanics the way Jay Gould understood stock manipulation.
By the late 1990s, WorldCom had acquired over 60 companies, including MCI Communications in a $37 billion deal that created the second-largest long-distance carrier in America. Ebbers was not building a fiber network from scratch. He was assembling one through M&A, layering acquisition upon acquisition, financing each deal with stock and debt that required the next deal to justify the last one.
The parallel to the railroad consolidators is almost exact. Ebbers was a financial engineer first, an operator second. The network grew. But the debt grew faster.
Global Crossing was the audacious one. Founded in 1997 by Gary Winnick — a former Drexel Burnham Lambert associate who had learned deal-making at the feet of Michael Milken — the company set out to build the first privately financed global fiber optic network. The flagship project was a transatlantic cable called Atlantic Crossing, connecting the United States and the United Kingdom. The company raised $750 million in its IPO and proceeded to build submarine cables across the Atlantic and Pacific, connecting four continents.
Winnick understood narrative. He understood that the story of a global fiber network was worth more, in the capital markets, than the near-term economics of operating one. Global Crossing’s market capitalization reached $38 billion. Its actual revenue never remotely justified that valuation. But in the late 1990s, the gap between narrative value and economic value was the whole game — just as it had been for the railroad promoters of the 1870s.
360networks was the Canadian entry — a Vancouver-based company that planned to build 112,000 kilometers of fiber optic network across North America. It raised $900 million in its 2000 IPO, one of the largest in Canadian history. The company was burning through capital at a pace that required the bandwidth market to grow at rates that, even by optimistic forecasts, were unrealistic.
Level 3 Communications was the one built to survive — and I’ll come back to them, because they are the most important character in this story.
The others — Flag Telecom, Genuity, ITC DeltaCom, Williams Communications, McLeodUSA, Winstar — were variations on the same theme: serious technical operations financed with capital structures that required perfect market timing.
III. The Mania Mechanics
The fiber buildout followed the same combustion sequence we documented in the Railroad Wars. Every infrastructure mania runs on the same engine, and recognizing the parts is how you avoid being the one holding the bag when it stalls.
A genuine, transformative technology. Fiber was real. A single strand of glass fiber could carry more data than the entire copper telephone network of the United States. The physics were not speculative. The technology was proven. Dense wavelength division multiplexing — DWDM — allowed multiple signals to travel simultaneously on a single fiber, multiplying capacity by orders of magnitude. The underlying technology was as real as steel rails and steam locomotives had been in 1870.
A policy catalyst that created competitive urgency. The Telecom Act of 1996 did for fiber what the Pacific Railroad Acts of 1862 and 1864 did for the transcontinental railroad: it turned a technical possibility into a competitive arms race. Once the market was deregulated, standing still was not an option. You built, or you were built around.
Capital availability that disconnected deployment from economics. The bond market was wide open. Telecom companies issued hundreds of billions of dollars in high-yield debt between 1997 and 2000. The equity markets were even more permissive — IPOs for companies with no earnings, minimal revenue, and ambitious network plans raised billions.
And the critical dynamic: once the capital was raised, deployment pressure became irresistible. If you are a CEO sitting on $2 billion in bond proceeds and IPO capital, you cannot call your board and say, “Actually, I think we should wait and see if demand materializes.” The capital itself creates the obligation to deploy. The money was raised to build. It had to be spent building. Whether the economics justified building right now was a question nobody was incentivized to ask.
This is the iron law we identified in Part I, and it applies with equal force here:
Once the money is raised, the buildout happens — whether or not the economics justify it.
IV. The Crash
The fiber market began to crack in late 2000, and by mid-2001, it was in free fall.
The proximate cause was straightforward: supply had massively outrun demand. Industry estimates varied, but the consensus was brutal — somewhere between 85% and 97% of the fiber that had been laid was “dark.” Not lit. Not carrying traffic. Not generating revenue.
The companies that had built the network were carrying debt loads sized for a demand curve that had not arrived. When revenue failed to materialize at the pace the capital structures required, the math became impossible. You cannot service $10 billion in debt on $2 billion in revenue, no matter how good your engineers are.
360networks filed for bankruptcy in June 2001. The $900 million IPO, completed just fourteen months earlier, was effectively a total loss.
Global Crossing filed in January 2002, in a roughly $25 billion collapse. Gary Winnick had already sold over $700 million worth of his personal shares before the filing. The parallel to the Crédit Mobilier promoters — who extracted their returns during construction and left the bondholders with the wreckage — is impossible to miss.
WorldCom collapsed in July 2002 in what was then the largest bankruptcy in American history — $107 billion in assets. And then, an internal audit uncovered $3.8 billion in fraudulent accounting entries (later discovered to be closer to $11 billion!) — expenses reclassified as capital expenditures to inflate earnings. Ebbers had been running a financial machine that required perpetual growth to justify the debt, and when the growth stopped, the fraud was the only thing keeping the numbers alive. He was convicted in 2005 and sentenced to 25 years.
The Nasdaq fell 78% from its March 2000 peak. An estimated $5 trillion in market value was destroyed across the technology and telecom sectors. More than 500,000 telecom workers lost their jobs.
And the fiber — 80 million miles of it, buried in conduits under cities and strung across ocean floors — sat there. Dark. Waiting.
V. The Dark (Ages) Period
Between 2002 and roughly 2006, the conventional wisdom about the fiber buildout was simple: it was a catastrophic misallocation of capital. A cautionary tale about speculative excess. The greatest destruction of telecom shareholder value in history.
All of that was true — for the first owners.
But the fiber did not disappear. Glass doesn’t rot. Conduit doesn’t decay. Submarine cables don’t dissolve. The physical infrastructure that had been built at a cost approaching $1 trillion was still there, intact, maintained at minimal cost, depreciating on balance sheets that had been restructured in bankruptcy court.
This is the dark period — the interregnum between the mania and the ignition. It’s the years between the Panic of 1893 and the industrial boom that turned J.P. Morgan’s reorganized railroad bonds into compounding machines. It’s the gap between the peak of the narrative and the arrival of the actual demand.
During the dark period, the fiber changed hands. Bankruptcy courts sold assets. Distressed debt funds bought bonds at pennies on the dollar. New operators acquired network assets at fractions of their replacement cost.
The most important transaction of this era was the transformation of Level 3 Communications. Founded in 1998 by James Crowe — who had previously built MFS Communications, one of the first successful competitive telecom companies — Level 3 had survived the crash, barely, by being marginally less leveraged than its competitors. Through the dark period, Level 3 systematically acquired distressed fiber assets, buying network capacity at prices that would have been inconceivable during the mania.
The logic was pure Second Owner Advantage. Level 3 was acquiring at economic price — the price the cash flows could actually support — rather than at narrative price. The fiber was the same fiber. The routes were the same routes. Only the basis had changed.
The first owner paid the narrative price. The second owner paid the economic price. Those two numbers are almost never the same.
VI. Ignition
Between 2007 and 2012, the world the fiber had been built for finally arrived.
The iPhone launched in June 2007. Within three years, smartphones had fundamentally changed how human beings consume data. Mobile data traffic, which had been negligible in 2005, began doubling every year. Every smartphone was a node on a network that needed backhaul — and backhaul runs on fiber.
YouTube, acquired by Google in 2006 for $1.65 billion, scaled into mainstream consumption. Video — which consumes orders of magnitude more bandwidth than text or email — became the dominant form of internet traffic. Netflix launched its streaming service in 2007, and by 2010 was responsible for a significant percentage of North American internet traffic during peak hours.
Amazon Web Services, launched in 2006, made compute elastic. The cloud computing revolution that followed required data centers connected by — fiber. Massive amounts of fiber. The same fiber that had been sitting dark under cities and oceans for five years.
The demand curve that the builders of 1998 had projected for 2001 arrived in 2008. They were right about the destination. They were wrong about the timeline by roughly seven years.
And seven years, in infrastructure finance, is an eternity. It is the difference between servicing your debt and defaulting on it. Between collecting the payoff and going bankrupt before it arrives.
The fiber lit up. Dark fiber became the most valuable infrastructure asset class in telecommunications. The companies that had acquired it during the dark period — at pennies on the dollar — found themselves sitting on assets whose replacement cost was multiples of their acquisition basis.
The builders were gone. The payoff arrived. Someone else collected it.
This is the pattern. It was the pattern in 1893, when Morgan’s syndicates bought reorganized railroad bonds and held them through the industrial boom. It is the pattern every time.
VII. What This Teaches Us About the AI Buildout
I am not saying the AI infrastructure buildout is the fiber bubble. I am saying the mechanism is identical — and if you understand the mechanism, you can position for what comes after.
The fiber buildout and the AI buildout share every structural feature that produces the same arc:
Government narrative creates inevitability. The Telecom Act of 1996 told the market that bandwidth was the future and competition was the policy. Today, the CHIPS Act, the Infrastructure Investment and Jobs Act, and bipartisan consensus around AI dominance have created the same “this is necessary” narrative that insulated the fiber buildout from normal capital discipline.
Rational actors produce irrational collective outcomes. Every fiber company that built in 1999 was making a defensible individual decision — if you don’t build the network, your competitor will, and you’ll be locked out of the market. Today, every hyperscaler building data centers is making the same defensible individual decision. The collective result is the same: parallel construction in competitive corridors, aggregate capacity that massively exceeds near-term demand, and capital structures that require optimistic utilization assumptions.
The capital creates the deployment pressure. WorldCom raised billions and had to spend it. Global Crossing raised $750 million in its IPO and immediately began laying cable. Today, Meta syndicated $38 billion in data center financing in 24 hours. Microsoft committed $80 billion in a single fiscal year. OpenAI has announced partnerships implying over a trillion dollars in infrastructure spending. When the capital is raised, it gets deployed. It always does.
The demand curve will arrive late. Fiber demand arrived roughly seven years after the buildout peaked. The applications that justified the infrastructure — streaming video, cloud computing, mobile data — required ecosystem development that could not be rushed. AI infrastructure faces the same dynamic: the applications that will justify the data centers being built today may require enterprise adoption cycles, regulatory frameworks, and workflow integrations that take longer than the capital structures assume.
The question is not whether AI infrastructure will matter. It will. The fiber proved that. The question is who pays for it — and who gets to use it when the demand finally arrives.
In every great buildout, the builders are praised. Then they are ruined. Then the patient inherit the assets.
VIII. The Second Owner Framework: Fiber Edition
In Part I of this series, we laid out the five principles the Railroad Wars teach the modern investor. The fiber cycle confirms every one of them — and adds a critical refinement that applies directly to where we are in the current AI infrastructure arc.
