In March 2000, the world was convinced the internet would change everything. And to be fair, it did.
At the center of that transformation stood Cisco Systems. The company sold routers and switches, the invisible plumbing that allowed the internet to function. As bandwidth demand exploded, Cisco’s revenues surged. Investors rewarded it generously. At its peak, Cisco briefly became the most valuable company on the planet, commanding a market capitalisation north of $500 billion.
Then the bubble burst.
Not because the internet failed. Not because Cisco was a weak company. But because expectations ran ahead of reality. Telecom operators had overbuilt. Corporate spending slowed. Excess capacity flooded the system. Cisco was forced to write down billions in inventory. The stock fell nearly 80 percent and took over a decade to revisit its previous highs.
Fast forward to today and you can see why Michael Burry is uneasy.
In recent commentary, Burry drew parallels between Cisco at the peak of the dot com boom and Nvidia Corporation in the middle of the AI surge. His focus is not on hype headlines or social media exuberance. It is on something far more mechanical and measurable: balance sheet commitments.
The $95 billion red flag: Nvidia’s growing commitments
Over the past year, Nvidia’s purchase obligations have ballooned. Long term supply commitments reportedly jumped from roughly $16 billion to around $95 billion within a year. When inventory and other contractual obligations are included, total exposure approaches $110 to $120 billion. That number sits uncomfortably close to Nvidia’s annual operating cash flow.
In plain English, Nvidia has locked itself into massive future production capacity.
Why would it do that? Because demand right now looks insatiable. Hyperscalers are racing to build AI infrastructure. Data center revenue has more than tripled year on year in recent quarters. Gross margins have crossed 70 percent. Free cash flow has exploded. Customers are placing orders months in advance.
From a management perspective, pre-securing supply in a constrained environment seems rational. If you believe demand will remain structurally high, locking in wafers and packaging capacity protects your competitive position. It ensures you do not lose sales to shortages.
But this is precisely where Burry sees echoes of Cisco.
In the late 1990s, Cisco ramped production aggressively because internet traffic growth appeared unstoppable. Telecom operators kept ordering more gear. Forecasts assumed exponential expansion. So Cisco locked in supplier contracts and expanded manufacturing capacity to match projected demand.
When spending cycles turned, those fixed commitments became a burden.
Asymmetric risk: When extraordinary growth normalises
Burry’s argument is not that AI will disappear. It is that growth rates eventually normalise. When they do, companies that have pre committed tens of billions of dollars face asymmetric risk. If demand continues at full throttle, the commitments look visionary. If demand cools even modestly, margins can compress quickly because supply obligations do not vanish overnight.
Consider the structure of Nvidia’s customer base. A significant portion of its AI revenue comes from a handful of large cloud providers. If those players pause capital expenditure for even a few quarters, the impact could be meaningful. Add to that the rise of custom silicon initiatives and alternative accelerators, and the assumption of perpetual pricing power becomes less certain.
History does not repeat perfectly, but it often rhymes.
Cisco remained profitable after the crash. The internet continued growing. Yet investors who bought at peak valuations endured years of underperformance because the stock had priced in perfection. When growth shifted from extraordinary to merely strong, multiples contracted sharply.
Nvidia today commands a multi trillion dollar valuation at its highs. Embedded in that price are assumptions about sustained dominance, high margins and relentless AI spending. The company may well deliver on many of these fronts. Its CUDA ecosystem creates switching costs. Its technological lead remains formidable. Its profitability metrics are far stronger than many dot com era companies ever achieved.
But valuation is a function of expectations, not just performance.
The distinction between tech success and price performance
Burry’s warning essentially asks a simple question: what happens if AI demand grows strongly, but not exponentially? What happens if spending becomes cyclical rather than parabolic? What happens if supply catches up faster than expected?
When forward commitments approach the scale of annual cash generation, flexibility declines. And markets tend to punish inflexibility when cycles turn.
The dot com era taught investors that being right about technology is not the same as being right about price. The internet reshaped the global economy, yet many stocks associated with it delivered poor returns from peak levels because growth expectations had overshot sustainable reality.
That is the lens through which Burry views Nvidia.
He is not dismissing AI. He is scrutinising the mechanics of expansion during euphoric phases. When companies behave as if current demand conditions will persist indefinitely, they implicitly reduce their margin for error.
Whether Nvidia ultimately mirrors Cisco’s trajectory remains to be seen. But the comparison forces investors to separate technological inevitability from financial inevitability.
And that distinction, as history has shown, can make all the difference.