Dot-Com Flashback: Capacity Swaps

This is the second article in the “Dot-Com Flashback” series. The first article, Financing the Customer, I covered how dot-com era companies used vendor financing to manufacture demand that didn’t exist, lending money to customers who then used it to buy the vendor’s own products. Revenue went up. Demand looked real. Until it wasn’t.

Today we go deeper into the Dot-Com’s playbook with the second trick: Capacity Swaps.

If vendor financing was the act of lending your customers the money to buy your product, capacity swaps were the act of two companies "selling" each other usage rights of network capacity they didn't need, booking the sale as revenue, and burying the purchase on the balance sheet as a capital expenditure, amortized over 20 years. Revenue recognized immediately. Cost spread over two decades. That was the trick. That was one of the ways companies to inflated demand and boost revenue to keep up with the narrative. And it worked beautifully, until it didn't.

---

Before I start: thank you to everyone who subscribes, follows, and supports this work. Very soon I will be moving to a paid subscription model, delivering the same content, analysis, and due diligence you have come to expect, but more frequently, and with a subscribers-only chat where we can discuss everything together, and much more. Until then, if you enjoy what I write, please subscribe and/or pledge using the button below. It is genuinely appreciated.

---

What Is a Capacity Swap?

The “product” being traded was something called an IRU, an Indefeasible Right of Use. In simple terms, it is a long-term lease on fiber optic capacity, usually spanning 20 or more years, that gives the buyer exclusive rights to use a specific amount of bandwidth on a specific route. Think of it like buying a reserved lane on a highway. You don’t own the highway, but you own the right to use that lane, and no one can take it from you.

In a normal, legitimate business, a telecom company would buy an IRU from another carrier because it needed capacity on a route it didn’t serve. Company A has network coverage in the U.S. but not in Europe. Company B has coverage in Europe. Company A buys capacity from Company B. Makes sense. There is a real business need. There is a real customer at the end of the chain who benefits.

But that is not what was happening at Global Crossing and Qwest.

Swap, Book, Repeat

What was happening was this: Company A “sells” $100 million of capacity to Company B. At the same time, Company B “sells” $100 million of capacity back to Company A. Both sides book $100 million in revenue. Both sides book $100 million in capital expenditure amortized over the life of the contract. The net economic value of the transaction is zero. But it doesn’t matter as the revenue was recorded and it looked “real” on the income statement.

Global Crossing reported $720 million in cash revenues from the sale side of its capacity swaps in just the first two quarters of 2001. Qwest reported over $1 billion in revenue from network capacity sales in 2001, and more than two thirds of those sales turned out to be swaps where Qwest simultaneously purchased similar amounts from its counterparties.

Yep. Two thirds of Qwest’s reported network capacity revenue in 2001 came from swaps, as Qwest looked to manufacture demand and boost revenue through round-trip transactions.

The Numbers Had to Be Met

Congressional hearings in September 2002 revealed through internal documents and emails that the sales teams were driving these swap transactions. The network people, the ones who would actually know whether additional capacity was needed, repeatedly questioned the rationale for these purchases. But the sales teams were under enormous pressure to meet quarterly revenue targets that executives had publicly promised to Wall Street.

One email from a Global Crossing executive captured it perfectly. When told that $75 million in capacity purchases had been approved for the quarter, the response was: “What the hell are we going to buy?” They had no idea why they needed the capacity because there was no business need. But the purchase side of the swap was irrelevant. The revenue side was all that mattered.

The dollar amounts of these swaps were “suspiciously flexible,” as the congressional investigation put it. The capacity being traded could be adjusted, stretched, rerouted, whatever was needed to make the numbers work. It was not the value of the transaction that drove the deal. It was the urgency to close it before the end of the quarter. The gap between projected revenue and actual revenue determined the size of the swap, not the other way around.

The Side Agreements

To book revenue from an IRU sale upfront rather than ratably over the 20-year contract life, certain accounting criteria had to be met. The most important one: the capacity had to be route-specific, meaning the buyer could not freely move it around the network. If the buyer had portability, it wasn’t a capital lease. It was an operating lease. And operating leases don’t let you book the full revenue on day one.

This created a problem. Companies didn’t want to be locked into specific routes they didn’t need. So Qwest’s sales team came up with a solution: enter into written contracts that met the accounting criteria on paper, then make oral side agreements promising the buyer they could swap routes later.

The accounting team and the auditors saw the written contract. They never saw the side agreements. Had they known, the revenue recognition treatment would have been completely different, and that is why Qwest was forced to restate $950 million in revenue.

The SEC later charged Qwest with securities fraud, alleging that between 1999 and 2002, the company fraudulently recognized over $3.8 billion in revenue. The SEC noted that Qwest employees had a name for these one-time IRU transactions that were used to fill the gap between actual revenue and projected revenue. They called them “heroin.” Others called them “cocaine on steroids.” The practice, they said, was an “addiction.”

The Metrics They Invented

The swaps alone were not enough. To make the numbers shine, Global Crossing needed the right presentation. So it invented its own metrics.

Under GAAP, when Global Crossing sold a 20-year IRU for $20 million, only $1 million could be recognized as revenue in the first year. The remaining $19 million sat on the balance sheet as deferred revenue, a liability. The company had the cash in its bank account. It was non-refundable. But it could not call it revenue. Not under GAAP. Not yet.

This was a problem. The cash was real, but GAAP revenue was small. And Wall Street cared about revenue growth.

So Global Crossing created two supplemental metrics: “cash revenue” and “adjusted EBITDA.” These included the full upfront cash payment from IRU sales, not the ratable GAAP amount. The company published these metrics alongside its GAAP results and told analysts they were “well understood by the marketplace.” In reality, they were designed to make the company look dramatically healthier than it was.

How dramatically? In filings submitted in the spring of 2001, Global Crossing reported an additional $531 million in the pro forma statement that was not included in the GAAP-compliant earnings. That represented a nearly 50% increase in reported earnings, entirely attributable to the way the company chose to present its swap-related cash receipts. As Chairwoman Sue Kelly put it during the March 2002 hearing: pro forma statements, because they are not required to follow GAAP, provided “an easy opportunity to cook the books.”

The SEC later quantified the damage. In Q2 2001, reciprocal transactions accounted for 32% of Global Crossing’s reported Cash Revenue. Without those transactions, the company’s Recurring Adjusted EBITDA would have gone from the reported $472 million to negative $43 million. Positive $472 million to negative $43 million. That is a $515 million swing in a single quarter, created entirely by round-trip swap transactions that had no net economic value. Without these reciprocal transactions, Global Crossing would not have met analysts’ estimates for either Cash Revenue or Recurring Adjusted EBITDA in that quarter.

And when Qwest’s president testified before Congress that same year, he described the company’s IRU swap exposure as “approximately 2 percent in 2000 and 3.5 percent of total reported revenues.” That sounded immaterial. But, surprise, he misled Congress. The SEC later found that Qwest had fraudulently recognized over $3.8 billion in revenue between 1999 and 2002. The number the president presented to Congress was a fraction of the actual exposure.

The Oversupply Problem

Behind all of this was the same fundamental problem that always sits behind these schemes: oversupply.

The telecom industry in the late 1990s and early 2000s built capacity at a pace that far exceeded demand. Dozens of companies raced to lay fiber across the country and across oceans, each convinced that internet traffic would grow exponentially forever. Equipment vendors like Lucent and Nortel sweetened the stakes by lending money to these carriers so they could buy more gear (vendor financing, the topic of my first article in this series). The result was a massive glut of fiber optic capacity. By 2004, analysts estimated that only about one tenth of installed fiber was actually being used.

Rather than acknowledge the oversupply, the entire sector formed a cartel of aggressive accounting. Qwest and Global Crossing weren't acting alone; they needed equally desperate counterparties to make the math work. Enron jumped in, using its Enron Broadband Services division to execute reciprocal swaps with Qwest, propping up its own doomed growth narrative. Qwest also bought massive amounts of redundant East Asia fiber from companies like Flag Telecom and TyCom Networks. Internal studies later showed Qwest had absolutely no business use for the majority of it. It was geographically duplicative. But buying it allowed Flag and TyCom to buy back from Qwest, creating the appearance of growing revenue in a shrinking market.

Global Crossing continued to engage in these transactions even while conducting an internal review to figure out how to get rid of excess capacity from previous swaps. That review found the company lacked the working capital to incorporate roughly $1 billion of its purchased capacity into its network. It ended up trying to sell that capacity for pennies on the dollar.

Global Crossing filed for bankruptcy on January 28, 2002. The fourth largest bankruptcy in U.S. history at the time. $54 billion in investor losses. Nearly 10,000 jobs gone.

One more thing worth mentioning. In March 2002, at the first congressional hearing on these matters, WorldCom’s general counsel testified under oath that “WorldCom does not participate in such transactions.” He explicitly distanced the company from the swap practices of Global Crossing and Qwest. The company that publicly declared itself clean, that sat at the witness table and told Congress it was above these games, turned out to be running the largest accounting fraud in American history at the time. $11 billion. WorldCom collapsed just months after that testimony. If there is a lesson in that, it is this: the company that volunteers the loudest denial is sometimes the one you should be watching most carefully.

Sound Familiar?

Fast forward to today.

In the AI boom, we are watching a version of the same mechanics play out in real time, with updated terminology. Replace “fiber optic capacity” with “GPU compute.” Replace “IRUs” with “Cloud Credits.” Replace “capacity swaps” with “GPU for cloud credit swaps” and “compute barter transactions.” Replace “Qwest and Global Crossing” with the names of companies you already know. Nvidia and the Neoclouds, Microsoft and OpenAI, Amazon and Google with Anthropic, etc.

Consider the structure: Company A (a GPU maker) invests equity in Company B (a Neocloud). Company B uses the investment to buy GPUs from Company A. Company A books the sale as revenue. Company B books the GPUs as assets and starts depreciating them. Company A then signs a contract to lease compute back from Company B, meaning Company A is now paying Company B for access to the very GPUs it just sold them.

The “product” has changed. The mechanism hasn’t. The circular flow of money, the dependency, the problem, all the same.

During the dot-com telecom bust, the swaps existed because real demand was declining and executives needed to manufacture revenue to meet Wall Street expectations. Today, we are told that demand for AI compute is endless. Maybe it is. Maybe it isn’t. But the question is not whether demand exists in aggregate. The question is whether the specific transactions being used to book revenue reflect genuine, arms-length demand from independent customers, or whether they reflect a circular ecosystem where the supplier, the buyer, and the financier are all connected, and where the money flows in a loop that inflates everyone’s numbers.

When Nvidia invests in a neocloud, and that neocloud uses the capital to buy Nvidia GPUs, and Nvidia then leases compute back from that neocloud, and that neocloud’s entire business model depends on Nvidia’s next generation allocation and continued support, Is that an arms-length transaction? Or is that a capacity swap with equity characteristics? Or maybe it is a company-SPV relationship? Or maybe it is all of them?

When a hyperscaler signs a multi-billion dollar commitment with a neocloud for GPU capacity, and simultaneously backstops that neocloud’s debt facility, is that genuine demand? Or is it a mechanism to move capex off the hyperscaler’s balance sheet and onto someone else’s, while still controlling the capacity?

When an AI lab throws non-binding commitments worth hundreds of billions across the ecosystem, and those commitments flow into the RPO lines of hyperscalers and the revenue projections of neoclouds, and everyone points to everyone else’s numbers as proof that demand is real, how is that different from two telecom companies swapping capacity and both calling it revenue?

And when companies today invent their own metrics to tell a better story, “remaining performance obligations,” “annualized revenue run rate,” “contracted backlog,” metrics that combine binding and non-binding commitments, real revenue and projected revenue, current demand and aspirational demand, all into a single number that always goes up, how is that different from Global Crossing inventing “cash revenue” and “adjusted EBITDA” to include the full value of swap cash that GAAP would not let them book as revenue? some food for thought.

---

I recommend reading my following article:

Nvidia Presents the Cloud Credits Boom

and the follow-up note:

---

References:

1. U.S. Congress, House Committee on Energy and Commerce, Subcommittee on Oversight and Investigations. Capacity Swaps by Global Crossing and Qwest: Sham Transactions Designed to Boost Revenues? Hearing, 107th Cong., 2nd sess., September 24, 2002. Serial No. 107-129. Washington: Government Printing Office, 2002. Video available at https://www.c-span.org/program/house-committee/telecommunications-accounting-scams/130356. Transcript available at https://www.govinfo.gov/content/pkg/CHRG-107hhrg81961/html/CHRG-107hhrg81961.htm.

2. U.S. Congress, House Committee on Energy and Commerce, Subcommittee on Oversight and Investigations. Capacity Swaps by Global Crossing and Qwest: Sham Transactions Designed to Boost Revenues? Hearing, Day 2, 107th Cong., 2nd sess., October 1, 2002. Serial No. 107-129. Washington: Government Printing Office, 2002. Video available at https://www.c-span.org/program/house-committee/telecommunications-accounting-scams-day-2/129827.

3. U.S. Congress, House Committee on Financial Services, Subcommittee on Oversight and Investigations. The Effects of the Global Crossing Bankruptcy on Investors, Markets, and Employees. Hearing, 107th Cong., 2nd sess., March 21, 2002. Serial No. 107-63. Washington: Government Printing Office, 2002. Video available at https://www.c-span.org/program/house-committee/global-crossing-bankruptcy/164235. Transcript available at https://www.govinfo.gov/content/pkg/CHRG-107hhrg78601/html/CHRG-107hhrg78601.htm.