The Rhyme: Wall Street Packages the AI Boom & Its 1998 Echo

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

◉ THE PRESENT

On Monday, the Financial Times reported that a consortium led by Nvidia, Apollo Global, Blackstone, BlackRock, and Brookfield is assembling a $500 billion funding package for AI infrastructure — chips, power plants, data centers, the physical spine of the machine-learning economy. This comes two months after Apollo and Blackstone anchored a $35 billion loan secured directly against Google's custom chips, and five months after Blackstone backstopped CoreWeave's $8.5 billion credit facility collateralized by Nvidia GPUs. Nvidia fell 3% on the news. Markets held near record highs, with the S&P 500 at 7,766, but oil climbed near $80 WTI on Hormuz doubts, and tomorrow's CPI print looms over everything.

The last time Wall Street built financial instruments this large around a single infrastructure thesis, the year was 1998, and the infrastructure was fiber optic cable.

◉ THE ECHO — AUGUST 14, 1998

The day Wall Street bet $399 million on glass in the ocean

Gary Winnick learned his trade from Michael Milken. He spent the first half of the 1980s at Drexel Burnham Lambert, the junk bond shop that blew up in scandal, and he walked away with the one lesson that mattered: you don't need profits to raise money. You need a story. In March 1997, Winnick put $15 million of his own cash into a company called Global Telesystems — soon renamed Global Crossing — and started telling anyone who would listen that he was going to lay fiber optic cable across the bottom of the Atlantic Ocean. At the time, every undersea cable in the world was owned by a consortium of old-line phone companies. Winnick's pitch was simple. The internet was doubling every hundred days. Somebody had to build the pipes. Why not him?

Wall Street listened. By June 1998, before a single paying customer had sent a single bit of data across Global Crossing's cables, Salomon Brothers, Merrill Lynch, and CIBC had helped Winnick raise $800 million in debt. Then came the real move. On Friday, August 14, 1998, Global Crossing went public on the NASDAQ at $19 a share, raising another $399 million. The stock closed its first day at $25.50. The market capitalization of a company that had never turned a profit and was still physically laying cable on the ocean floor hit $5.15 billion before the weekend. Winnick's $15 million investment was suddenly worth more than a billion dollars, making him a billionaire faster than anyone in American history at that point — faster than Gates, faster than Buffett.

Three days later, on Monday, August 17, something else happened. Russia defaulted on its sovereign debt. The ruble collapsed. Long-Term Capital Management, the hedge fund run by Nobel laureates and bond-market wizards, began hemorrhaging money — $553 million on a single day in late August alone. The Federal Reserve would cut rates three times that fall, in September, October, and November, to keep the financial system from seizing. Those rate cuts did the job. They also poured gasoline on everything. The NASDAQ, which had already blown past 2,000 in July, would rip from about 1,419 in October 1998 to 5,048 by March 10, 2000. And the telecom debt machine that Global Crossing's IPO had helped legitimize would become the most destructive financing engine since the railroad bond bubble of the 1870s.

Between 1996 and 2001, telecom companies raised $1.6 trillion on Wall Street. They floated $600 billion in bonds. They laid 80 million miles of fiber optic cable. When it was over, 96% of that fiber was dark — unlit, unused, a physical monument to money that believed its own projections. Global Crossing's stock hit $64 within seven months of the IPO. By the time the company filed for bankruptcy on January 28, 2002, carrying $12.4 billion in debt, the shares traded at thirty cents. WorldCom followed six months later, collapsing under $41 billion in debt and $3.8 billion in fabricated accounting entries. The NASDAQ lost 78% of its value. Five trillion dollars vanished.

The infrastructure survived. The fiber that Global Crossing and its competitors buried across the ocean floors eventually carried the modern internet. The investors who financed it did not survive with it.

◉ THE RHYME — WHAT'S IDENTICAL

When Wall Street starts packaging an infrastructure boom into financial products, it means the smart money has shifted from building to selling. The collateral is real. The multiples are not.

◉ THE DIVERGENCE — WHAT'S DIFFERENT THIS TIME

Where the Pattern Breaks

AI has paying customers. Fiber did not. Global Crossing never turned a quarterly profit. Its revenue model depended on other telecom companies buying "capacity swaps" that were often just circular accounting entries dressed up as sales. Nvidia, by contrast, sold $193.7 billion worth of data center chips in fiscal 2026. The AI companies borrowing against those chips — Anthropic, CoreWeave, the hyperscalers — have real, measurable workloads running on real hardware. The demand is not imaginary. The question is whether $500 billion worth of demand exists.
The collateral depreciates on a different clock. Fiber optic cable, once buried under the Atlantic, holds its physical value for decades. That made the eventual write-downs worse — you owned an asset nobody wanted. AI chips lose half their value in eighteen months as new generations arrive. If the loans start souring, the collateral won't be sitting there waiting to be sold at a discount. It will be technologically obsolete. That makes the downside faster and sharper than it was in telecom.
Private credit, not public bonds. In 1998, telecom debt was packaged into publicly traded bonds that anyone could buy — pension funds, mutual funds, retail investors. The AI financing is flowing through private credit vehicles at Apollo, Blackstone, and Brookfield. That concentrates the risk in fewer, larger hands. It also means the losses, when they come, will be harder to see in real time. There's no daily mark-to-market on a $35 billion private loan secured against Google chips.
The macro threat is inflation, not deflation. The 1998 crisis was deflationary — Russia defaulting, emerging markets collapsing, the Fed cutting rates. Today's backdrop is the opposite. Oil near $80, CPI at 3.5%, and a Fed that three governors want to push toward a rate hike. If inflation forces rates higher, the cost of servicing $500 billion in AI infrastructure debt goes up at the worst possible time.

◉ THE RECKONING — WHAT HAPPENS NEXT

Here is what happened after August 14, 1998. The Russia crisis and LTCM collapse knocked the NASDAQ down about 30% from its July peak to its October low. The Fed panicked. Alan Greenspan cut the federal funds rate from 5.50% to 4.75% in three moves between September 29 and November 17. The market found its floor. And then — this is the part people forget — the rally that followed was the most violent upside move in American stock market history to that point. The NASDAQ more than tripled from October 1998 to March 2000. Global Crossing's stock went from $19 to $64. Money that had been sidelined during the Russia scare poured back in, and the telecom financing machine went from warm to white-hot. In 1999 alone, the industry made $1.3 billion in underwriting fees. Four hundred and fifty-seven companies went public. First-day IPO returns averaged over 70%.

The infrastructure thesis was correct. Internet traffic was growing at a rate that did eventually require all the cable being laid. The timing was wrong by about a decade. By the time the traffic arrived, the companies that built the pipes had been bankrupt for years, their assets sold for pennies on the dollar. The investors who financed the buildout — the ones who bought the bonds at par and the equity at IPO prices — lost everything. The second-wave buyers, the ones who picked up dark fiber for five cents on the dollar after the bankruptcies, made fortunes.

The pattern is mechanical. Wall Street doesn't create $500 billion financing vehicles at the beginning of a cycle. It creates them near the end, when the easy equity returns have been captured and the only way to keep the machine running is leverage. The Global Crossing IPO wasn't the start of the fiber boom. The first transatlantic fiber optic cable was laid in 1988. The IPO came years into the buildout, right when the industry needed more money than equity alone could provide. The Nvidia consortium isn't the start of the AI infrastructure boom, either. Nvidia's data center revenue started accelerating in 2023. This is three years in. The private credit machine is arriving on schedule.

The edge: In 1998, the financialization of the telecom boom marked the start of the last, fastest leg higher — and the beginning of the end. The market didn't top for another 19 months. The investors who understood that timeline made money on the way up and got out before the bonds defaulted. The ones who confused the financing for validation rode it all the way down. The clock starts when Wall Street begins lending against the hardware itself. That clock started in March 2026.

◉ TOMORROW’S WATCH

July CPI drops at 8:30 a.m. Wednesday. If headline inflation comes in above 3.4%, the Fed's three hawkish dissenters will have the data they need to push for a September hike — and the cost of financing $500 billion in AI infrastructure will suddenly look a lot less theoretical. Watch the 10-year Treasury yield. In November 1999, when the 10-year crossed 6%, it was the first crack in the telecom financing model. The AI version of that moment may be closer than the bond market thinks.