The Rhyme: the AI Capex Boom Just Cracked & Its 2000 Echo

"History doesn't repeat… but it rhymes." — Mark Twain

Oracle declared force majeure on its $165 billion Project Jupiter data center last Thursday, and the aftershock is still working through Wall Street this morning. The stock closed Friday at $137 — down 60% from its peak of $345.72 — while the project's $18 billion in construction loans traded at 89 to 91 cents on the dollar, an effective borrowing cost north of 8.5%. Oracle spent $28.5 billion on capital expenditures last quarter alone, pushing free cash flow to negative $5.4 billion, and interest expense climbed 55% year over year.

ORCL −60% from peak  |  AI capex $600B+ projected 2026  |  10Y yield 5.22%  |  Oracle FCF −$23.7B (FY)  |  Project debt at 90¢/$1

The defining technology of a generation, financed by the most aggressive debt structures the industry has ever seen, running into the highest borrowing costs in nineteen years. The last time an infrastructure boom hit this kind of wall was October 2000, in a building in Murray Hill, New Jersey.

Rich McGinn had promised Wall Street 20 percent annual growth. As CEO of Lucent Technologies — the AT&T spinoff that made the switches, routers, and fiber optic gear powering the internet revolution — he delivered on that promise for fourteen consecutive quarters. Lucent's stock hit $84 in December 1999. The company had 157,000 employees, nearly $38 billion in revenue, and a market capitalization north of $230 billion. With 4.6 million individual shareholders, it was the most widely held stock in America. McGinn told investors the bandwidth buildout was just getting started, that the internet would swallow more capacity than anyone could imagine.

He was right about the internet. He was wrong about the math.

On January 6, 2000, the first crack appeared. Lucent missed its quarterly earnings estimate, snapping a fourteen-quarter streak of beats. The stock dropped 28% in a single session, erasing $64 billion in market value. Analysts called it a speed bump. The Nasdaq was still climbing toward its March peak of 5,048, and nobody on Wall Street wanted to hear that the biggest infrastructure supplier in the hottest sector on earth had a real problem.

Three more warnings followed that year. By autumn, Lucent was quietly extending vendor financing to its own customers — lending them money so they could keep buying Lucent's products and keep the revenue line intact. It was the telecom version of a builder financing every buyer who walked in because the banks had stopped writing loans. The revenue looked real on the income statement. The cash was not.

On Monday, October 23, Lucent's board convened at headquarters on Mountain Avenue in Murray Hill, New Jersey. They fired Rich McGinn. Henry Schacht, who had run the company during its 1996 IPO, came back at age 66 to try to stop the bleeding. The company issued its fourth earnings warning of the year and told investors that first-quarter 2001 revenue would fall 7%. The stock, already down from $84 to about $22, barely moved on the news. The Street had priced in the warning. What it had not priced in was what the warning meant for everything else.

Lucent was not just one company having a bad year. It was the first domino in an infrastructure buildout that had gone parabolic. At the peak in 2000, telecom companies were spending $120 billion a year — over $213 billion in today's dollars — burying fiber optic cable under streets, across ocean floors, and through the mountains of two continents. The cumulative bet topped $500 billion. Capital markets financed all of it: bonds, high-yield loans, equity raises, capacity swaps that counted as revenue but never moved a dollar of real cash. And at the bottom of the entire thesis sat a single claim, originally made by WorldCom, that internet traffic was doubling every hundred days. That claim was false. But by the time the industry figured it out, the cable was already in the ground, and between 85 and 97 percent of it was sitting in the dark. Not lit. Not carrying traffic. Not making money.

The parallel is the same: when the cost of money rises faster than the infrastructure can generate returns, the math breaks. And math always wins.

After Lucent's board fired McGinn in October 2000, most of Wall Street treated it as a company-specific problem. Bad management. Poor execution. An isolated story, not a sector story.

They were wrong.

Within six months, the first telecom bankruptcies arrived. Winstar Communications — a broadband provider that owed Lucent $700 million — filed Chapter 11 in April 2001. The Canadian fiber company 360networks, which had raised $900 million in its 2000 IPO, went under in June. By January 2002, Global Crossing, which had built a $15 billion fiber optic network spanning 27 countries and 200 cities, filed for bankruptcy. And then the biggest domino of all: WorldCom declared Chapter 11 on July 21, 2002, in what was at the time the largest bankruptcy filing in American history.

The Nasdaq fell from around 3,468 the day McGinn was fired to 1,114 by October 2002 — another 68% down from a level that already felt like the bottom. Greenspan started cutting on January 3, 2001 — an emergency intermeeting 50-basis-point cut that sent the Nasdaq up 14% in a single session. The rally lasted two days. He cut eleven times that year alone, slashing the fed funds rate from 6.50% to 1.75% by December. It was not enough. The rate cuts could not fix the core problem, which was that $500 billion in infrastructure had been built for demand that would not arrive for another decade.

Here is what the smart money did. They did not short technology wholesale. They got surgical. They sold the companies that had financed growth with debt and could not generate free cash flow — the Lucents, the Global Crossings, the 360networks. They held the companies that had real earnings and could survive the bust: Cisco and Intel took massive hits but lived, consolidated market share, and eventually recovered. The technology was real. The financing was not sustainable. Those two facts can coexist, and knowing the difference was the entire edge.

The first company to crack in an infrastructure boom is never the last. In 2000, Lucent cracked first, and the sector did not bottom for two years. Oracle is the most leveraged player in the AI buildout. Watch whether Micron's data center revenue on Wednesday confirms demand or starts to gap against the capex numbers. The distance between those two figures is the distance between a healthy buildout and a 2000-style bust.

Micron reports after Wednesday's close — the first major read on whether AI chip demand still matches the $600 billion capex pipeline. If data center revenue disappoints while spending stays elevated, it echoes Corning's early-2001 earnings warning that confirmed the fiber glut nobody wanted to see.