Cisco's Dotcom Vendor Financing: The History

Jim Rickards has a claim about AI companies that keeps coming up in his recent presentations. He says they’re doing “circular financing” — lending each other money to buy each other’s products. The money goes in a circle. It shows up as revenue on somebody’s books. But nobody actually created any real value.

He compares it to Cisco during the dotcom bubble.

That comparison deserves a closer look. What did Cisco actually do? How big was it? And does the AI comparison hold up?

Let’s walk through it.

The Setup: Late 1990s

By 1999, Cisco was the undisputed king of networking equipment. Routers. Switches. The plumbing of the internet. John Chambers was the celebrity CEO, and Cisco’s stock was a holy grail — it had returned something like 90,000% over the previous decade.

But Cisco had a problem. Its fastest-growing customer segment was telecom startups — competitive local exchange carriers (CLECs), internet service providers, and other young companies trying to build out networks. They wanted Cisco’s gear. But they didn’t have cash.

Traditional banks wouldn’t touch them. These startups had negative earnings, no track record, and business plans that began with “we’re going to build a nationwide fiber network and beat the phone companies.”

So Cisco decided to become the bank.

Cisco Systems Capital

In the late 1990s, Cisco created a subsidiary called Cisco Systems Capital Corporation. Its job was simple: lend money to startups so they could buy Cisco equipment.

The mechanics were clean — too clean. Cisco would loan a telecom startup 100% of the purchase price for Cisco gear. The startup would buy the equipment through a distributor or Cisco partner. The distributor would pay Cisco. Cisco booked the sale as revenue. The startup got the equipment. Cisco got a loan receivable on its books.

The transaction was technically real. But the “customer” was spending Cisco’s money to buy Cisco’s product.

Think of it like this: A car dealership lends you $40,000 to buy a car from them. You drive off the lot. The dealership counts the full $40,000 as revenue and tells shareholders business is booming. But you haven’t actually paid anything. The dealership is out $40,000 in cash and has a loan from you on its books. If you never pay the loan back, that “sale” was always an illusion.

By 2000, Cisco executives confirmed that roughly 10% of the company’s $20 billion in annual revenue came through vendor financing and leasing deals.

The Terms Were Insane

You’d think a company lending its own money would be careful. Cisco was not.

The shareholder lawsuits filed in 2001 paint a damning picture. Cisco would lend up to a third over the retail price — meaning the loan amount frequently exceeded the equipment cost by more than 100%. Payment terms ran as long as nine years, with no payments due for the first two years.

Credit requirements? Waivable. Millions in credit could be approved without audited financial statements.

Salespeople could promise increased financing in exchange for a big order at quarter-end. Cisco would sometimes require customers to buy extra equipment they didn’t need or want — just to juice the numbers.

One customer, HarvardNet, had less than $5 million in sales and a negative net worth. When it compared Cisco’s gear to rival Paradyne’s, it concluded Paradyne was superior. So Cisco went straight to the top brass and offered up to $120 million in credit, with deferred payments and permission to spend on non-Cisco equipment. HarvardNet took the deal.

Another customer, Rhythms NetConnections, was founded by a major Cisco shareholder. It lost $36 million on $1 million in sales in 1998. S&P rated its debt deep into junk territory. Cisco shipped Rhythms $20 million in gear on credit just before its 1999 IPO.

American Metrocomm was already interviewing bankruptcy lawyers. Cisco lent them $62 million anyway.

The pattern was consistent: lend money to anyone who would buy Cisco gear, regardless of whether they could pay it back.

The Bill Comes Due

The telecom bubble burst in 2000-2001. CLECs started dropping like flies.

Winstar Communications filed for bankruptcy in April 2001. It had $5 billion in outstanding debts and owed Lucent (Cisco’s rival, running the same playbook) about $700 million.

By the end of 2001, Cisco had racked up $1.3 billion in financed-lease exposure across 735 customers. More than $233 million was tied to high-risk borrowers.

The reckoning was brutal. In April 2001, Cisco announced it was writing off $2.2 billion in excess inventory — one of the largest inventory write-downs in technology history. The company set aside nearly $900 million for bad loans.

Cisco posted a $2.69 billion loss that quarter. Its first loss in 11 years.

John Chambers called it a “100-year flood.”

The stock that had traded above $80 fell to $13. 8,500 people lost their jobs.

Cisco’s vendor financing arm shriveled from $1.9 billion in commitments in 2001 to $950 million in 2002 to just $90 million in 2003.

The revenue that looked real during the boom turned out to be Cisco’s own money, cycled through shaky startups, dressed up as customer purchases, and booked as growth. When the money stopped flowing, the illusion collapsed.

The AI Comparison: Fair or Not?

Rickards argues AI companies are running a similar playbook today.

Microsoft invests $13 billion in OpenAI. OpenAI commits to using Azure, Microsoft’s cloud platform. Microsoft books the cloud revenue. The money goes from Microsoft to OpenAI and back to Microsoft.

Nvidia invests in CoreWeave. CoreWeave buys Nvidia chips. Nvidia books the revenue.

OpenAI and Oracle sign a $300 billion partnership for data centers. Oracle builds them. OpenAI commits to paying for compute capacity over five years.

The structural similarity to Cisco’s vendor financing is real. In both cases, a supplier with deep pockets provides capital to a customer, and that capital returns to the supplier as revenue. The reported revenue growth includes money that started on the supplier’s own balance sheet.

But there are important differences.

OpenAI actually generates real products that real people pay for. ChatGPT has over 100 million users. That’s not WorldCom — which was a pure fraud, booking fake revenue on nonexistent capacity.

The question is one of degree. How much of Microsoft’s Azure AI revenue is genuine customer demand, and how much is OpenAI spending Microsoft’s own investment dollars? How much of Nvidia’s explosive data center growth comes from startups that Nvidia itself financed?

We don’t know. That’s the problem.

The Lesson

Vendor financing isn’t always fraudulent. But it’s always opaque.

When you can’t tell how much of a company’s revenue comes from real customers versus customers spending the company’s own money, you can’t value the business. That was the lesson from Cisco. It’s the question Rickards is asking about AI.

Cisco survived its 2001 crisis. It’s still around today, a $200 billion company. But the investors who bought Cisco at $80 in 2000 watched it fall to $13 and never saw that peak again. The stock trades around $50 today — 24 years later.

The money didn’t disappear. It just went in a circle. And when the circle stopped, someone was left holding the bag.

Related reads: If this dotcom-era financing pattern interests you, check out our breakdown of what the dotcom moment actually was, the Enron playbook comparison for AI debt, and the full Jim Rickards AI black paper promo breakdown.

— Sarge