When NVIDIA reported third-quarter revenue of $57 billion, up 62 percent, Wall Street breathed a collective sigh of relief. The AI bubble, it seemed, had been granted another reprieve. But beneath the headline-grabbing numbers lies a financial architecture that should trouble anyone who remembers the tech industry's previous monopolies—and the catastrophic collapses that eventually broke them.
The parallels to the 1990s "Wintel" duopoly are striking. Just as Microsoft and Intel created a remarkably dominant partnership dubbed 'Wintel' — Windows from Microsoft, on computers with CPUs from Intel, today's AI ecosystem is increasingly defined by the symbiotic relationship between NVIDIA's GPUs and the large language models they power—particularly those from OpenAI and Microsoft.
But the pattern becomes more troubling when we examine the financial entanglements. The circular capital flows that define today's AI boom bear an uncanny resemblance to the vendor financing schemes that brought down telecom giants like Cisco, Nortel, and Lucent in the early 2000s—a collapse that wiped out hundreds of billions in market value and countless jobs.
The Wintel Playbook: How Two Companies Controlled Computing
To understand where we might be headed, we need to understand where we've been. The Wintel monopoly of the 1990s wasn't just about market share—it was about creating an ecosystem so tightly integrated that escaping it became virtually impossible.
IBM may have deliberately chosen weak partners to ensure control over core functions when it selected Intel and Microsoft for its PC. What IBM didn't anticipate was that these "weak partners" would turn the tables, creating the "Wintel alliance" through Windows-Pentium intellectual property cross-licensing arrangements that locked in software developers and hardware manufacturers alike.
The network effects were devastating to competition. Applications written for one OS cannot run on another OS unless the developer undertakes the time-consuming and expensive process of adapting, or porting, the applications to another OS. Microsoft leveraged this to maintain its monopoly, prompting the United States District Court for the District of Columbia to rule that Microsoft's actions constituted unlawful monopolization under Section 2 of the Sherman Antitrust Act of 1890.
The parallels to today's AI ecosystem are unmistakable. NVIDIA's CUDA programming framework has created similar lock-in effects for AI developers, while the training costs for large language models—often running into hundreds of millions of dollars—create barriers to entry that would have made 1990s Microsoft executives blush.
The Return of Circular Financing: NVIDIA's Expanding Web
But here's where the historical echo becomes alarming. NVIDIA isn't just selling chips to the AI industry—it's increasingly financing it, in ways that recall the most problematic aspects of the dot-com bubble.
According to industry data, NVIDIA invested approximately $1 billion in AI startups in 2024, participating in 48 funding rounds. The company's investment portfolio spans the AI ecosystem: OpenAI (reportedly a $100 million investment as part of a $6.6 billion round valuing the company at $157 billion), xAI (participating in Elon Musk's $6 billion round), and dozens of other AI startups.
What makes this pattern concerning is where the money flows. Industry reports suggest major financing commitments to customers—creating circular arrangements where NVIDIA's capital enables companies to purchase NVIDIA's products. As one analyst noted, by leasing GPUs rather than requiring outright purchases, NVIDIA spares customers from taking accounting charges for high depreciation rates on the chips—but NVIDIA then bears those depreciation costs and the risk of holding devalued inventory if demand falters.
This circular flow of capital should set off alarm bells for anyone who remembers the telecom boom. During the dot-com era, Lucent committed $8.1 billion in vendor financing, Nortel extended $3.1 billion with $1.4 billion outstanding, and Cisco promised $2.4 billion in customer loans. The strategy seemed brilliant at the time: lend money to cash-strapped telecom companies so they could buy your equipment.
Until it wasn't. When the bubble burst, 47 Competitive Local Exchange Carriers (CLECs) went bankrupt between 2000-2003, and the equipment makers were left holding worthless debt.
An analysis comparing the two eras reveals troubling similarities. Lucent in 1999-2000 had vendor financing commitments representing roughly 20% of revenue, while estimates suggest NVIDIA's direct investments and financing commitments may represent a significantly larger share of annual revenue. The scale relative to business size appears substantially greater than what we saw before the telecom collapse.
GPUs as Collateral: A New Asset Class—Or a New Risk?
The circular financing pattern extends into an even more concerning innovation: AI startups are now using GPUs as collateral for massive loans—creating a debt market that didn't exist two years ago.
UK-based startup Fluidstack and others are securing loans using NVIDIA AI GPUs as collateral, with CoreWeave pioneering this model by raising $9.9 billion in funding through GPU-backed financing. Industry estimates suggest the total loan volume using GPUs as collateral has grown to exceed $20 billion.
The fundamental risk? The short product lifecycle of NVIDIA GPUs means the chips depreciate quickly, and lenders are increasingly concerned about how long the value of NVIDIA GPUs can be maintained. Indeed, lenders are now demanding higher interest rates and stronger collateral protections as they recognize these risks.
This should sound familiar. The assumption underlying this entire debt market is that GPUs will hold their value over 4-6 years—but NVIDIA is shortening its development cycles, potentially making existing GPUs obsolete faster than the loans backing them can be repaid. It's a dynamic that recalls how rapidly telecommunications equipment lost value when the bubble burst, leaving lenders holding collateral worth pennies on the dollar.
The Microsoft-OpenAI Tangle: When Partners Become Competitors
The financial entanglements extend well beyond NVIDIA. Recent leaked documents analyzed by tech blogger Ed Zitron—claims that have not been independently verified but align with other industry reporting—reveal the staggering interdependence between Microsoft and OpenAI.
According to these documents, in 2024, Microsoft received $493.8 million in revenue share payments from OpenAI, with that number jumping to $865.8 million in the first three quarters of 2025. But the relationship is even more complex: Microsoft also shares revenue with OpenAI, kicking back about 20% of the revenues from Bing and Azure OpenAI Service.
The compute costs are staggering. According to the leaked documents, OpenAI spent $8.7 billion on inference costs with Azure alone in the first three quarters of 2025, more than double the $3.7 billion reportedly spent in 2024.
Most troubling of all: while OpenAI's training spend is mostly non-cash—paid by credits Microsoft awarded as part of its investment—the firm's inference spend is largely cash. The implication, if these figures are accurate: OpenAI could be spending more on inference costs than it is earning in revenue.
Microsoft's stake in OpenAI now stands at approximately $135 billion, representing roughly 27-32.5% of OpenAI, making this one of the largest bets in tech history. Yet the relationship has grown increasingly strained, with negotiations between the two companies over restructuring becoming contentious enough that OpenAI has reportedly considered asking antitrust regulators to examine potential competition issues.
Central Banks Sound the Alarm
Unlike in the dot-com era, this time central banks are raising red flags before—rather than after—a potential crash. The European Central Bank warned that stock markets, particularly in the US, have become increasingly reliant on a small group of technology companies seen as beneficiaries of the AI boom, stating "This concentration among a few large firms raises concerns over the possibility of an AI-related asset price bubble".
The International Monetary Fund and Bank of England have joined the growing chorus warning of an AI bubble, while research firm Capital Economics projected that the current AI-driven rally will culminate in a bubble expected to burst by 2026.
Some analysts offer even more dramatic assessments. Julien Garran of MacroStrategy Partnership estimates that "misallocation of capital in the US" led by AI is 17 times bigger than the dot-com bubble and four times bigger than the 2008 real estate bubble.
Whether these specific predictions prove accurate or not, the fact that major financial institutions are issuing warnings suggests the patterns are visible to those tasked with monitoring systemic risk.
Why This Time Might Be Different—For Better or Worse
For all the parallels, there are crucial differences that could make this era either more resilient than the dot-com bubble—or more catastrophic.
The case for resilience:
Unlike Lucent's customers, which were leveraged startups burning cash, NVIDIA's top four customers generated $451 billion in operating cash flow in 2024 (Microsoft $119 billion, Alphabet $125 billion, Amazon $116 billion, Meta $91.3 billion). These companies have resources to weather a downturn.
Moreover, while fiber networks in 2000 used less than 0.002% of capacity, Microsoft and AWS report AI capacity constraints in 2025, suggesting real demand rather than purely speculative overbuild.
The case for concern:
Yet the concentration risk appears more severe. Lucent's top two customers—AT&T at 10% and Verizon at 13%—accounted for 23% of revenue in fiscal 2000, while NVIDIA reportedly derives 39% of revenue from just two customers and 46% from four customers—nearly double Lucent's concentration.
More concerning: Microsoft is developing Maia accelerators, aiming to use "mainly Microsoft silicon in the data center," meaning if hyperscalers shift to in-house chips, NVIDIA's vendor financing becomes exposure to customers building competitive alternatives.
The Antitrust Question: Lessons Unlearned?
The Wintel monopoly eventually faced regulatory scrutiny, though not before cementing Microsoft and Intel's dominance for over a decade. The District Court initially ordered a breakup of Microsoft, requiring the company to be split into two separate units, one to produce the operating system and one to produce other software components, though this was later overturned on appeal.
Today, the US Department of Justice has launched two separate antitrust probes into NVIDIA, evaluating whether the company has abused its market dominance and forced companies to buy additional products to receive GPUs while penalizing those that buy rival chips. The DOJ is also reportedly examining the company's $700 million acquisition of Run:ai in April 2024 and its 2022 purchase of software firm Bright Computing.
But here's the historical lesson: history shows that market forces, rather than public policy, have been much more effective at curbing monopoly power in the technology industry. The Wintel monopoly was ultimately undermined not primarily by antitrust action, but by Apple introducing the iPod, iPhone, and iPad in quick succession—light, portable smartphones and tablets that made heavy, clunky desktops and laptops seem obsolete.
The question is whether a similar market disruption will emerge before the current financial architecture is tested—or whether we'll see a dot-com-style correction first.
What the Patterns Suggest
The 1990s Wintel monopoly demonstrated that technical lock-in combined with network effects can create seemingly unassailable market positions. The dot-com bubble showed that vendor financing and circular capital flows can create the illusion of sustainable demand long after real market forces would have corrected prices.
Today's AI boom appears to combine elements of both: the technical and ecosystem lock-in reminiscent of Wintel, paired with circular financing patterns that echo the telecom bubble. Add GPU-backed debt instruments, stir in customer-investor relationships where the boundaries between buyer and seller blur, and you have a financial architecture that looks disturbingly familiar to those who study market history.
The ECB warned that "a correction in AI stock valuations could trigger global spillovers, given the interconnected nature of global equity markets". Given that 88% of NVIDIA's revenue comes from data centers, and that the entire AI startup ecosystem depends on access to compute, the concentration of risk warrants attention.
The question isn't whether today's AI market shows patterns that recall both Wintel and the dot-com bubble—the parallels are evident. The question is whether the differences—stronger balance sheets among major players, genuine technological breakthroughs, and real demand for AI capabilities—are sufficient to prevent those patterns from playing out similarly.
Markets have a way of testing assumptions built on circular financing and inflated valuations. The emerging GPU-backed debt market is built on the assumption that GPUs will hold their value over 4-6 years. History suggests such assumptions get tested much sooner than investors expect.
And when markets test assumptions built on vendor loans, circular capital flows, and concentrated customer dependencies, the results are rarely gentle.
Unlock the Future of Business with AI
Dive into our immersive workshops and equip your team with the tools and knowledge to lead in the AI era.
