Nvidia is the most valuable company in the world because it sells the one thing the entire AI build-out cannot proceed without, and because a software layer, CUDA, makes switching away from it painful. In late 2025 it became the first company ever to cross $5 trillion; it briefly touched $5.5T this spring, and even after a sharp early-June pullback it sits near $5T, worth more than every public company on Earth and more than every national stock market except the United States. With record FY2026 revenue of $216B (about $194B of it data center) and roughly 80% of the AI-accelerator market, it is priced as if that position is permanent.

We have seen this exact movie before, and we know how the middle acts play out. From the 1980s through the 2000s, the company that owned computing's center of gravity was Intel. It owned the x86 instruction set, it stamped Intel Inside on the box, and it taxed nearly every PC on Earth. In August 2000 Intel was the most valuable public company in the world at $509 billion. The two stories rhyme so loudly that the lazy take almost writes itself, in either direction. This piece argues the lazy takes are both wrong.

Before the parallels, the mechanism. Monopolies in semiconductors do not come from making the fastest chip for one year. They come from owning a standard, a piece of the stack that everyone else writes their software and builds their products against, so that leaving you means abandoning their own work. x86 was that standard for the PC. CUDA is that standard for AI. The pattern from there is almost mechanical, and it ends the same way every time.

Four parallels do most of the work. One, the moat is a standard, not a chip: x86 and CUDA both win because the world's software is written against them, so the switching cost is other people's code. Two, the customer is also the rival: the PC makers that lived on Intel and the hyperscalers that live on Nvidia are the same firms with the most reason to build an alternative, and today's hyperscalers have far more cash and chip-design talent than any 1990s PC maker ever did. Three, the valuation prices permanence: Intel at $509B in 2000 and Nvidia near $5T in 2025-26 were each, at their peak, the most valuable company on Earth, priced as though the tax never ends. Four, a single manufacturing dependency sits underneath: Intel's was its own fabs; Nvidia's is TSMC and its CoWoS packaging, the exact bottleneck we covered in The CoWoS Ceiling.

Now the three differences, because they cut the other way and they matter. First, the business model. Intel was an IDM: it designed and manufactured. So when its fabs fell behind TSMC, the whole company fell with them, and you cannot re-staff a fab the way you re-staff a sales team. Nvidia is fabless and asset-light; a manufacturing stumble is TSMC's problem to solve, not a decade-long Intel-style retooling. Second, the kind of moat. x86 was bound to Intel's own process; CUDA is software that travels across foundries and nodes, which arguably makes it the stickier lock-in. Third, the market's maturity. The PC market Intel rode was sliding into a slow replacement cycle by 2005; AI compute is still early. The pie is growing faster than any single rival can take share, which is why Nvidia's revenue keeps climbing even as its share drifts from roughly 90% toward 80%.

People use "bubble" to mean two different things, and Intel's history separates them cleanly. There is the market bubble, the moment a price detaches from any reasonable future, and there is the business plateau, the slower discovery that the monopoly was never as permanent as the price assumed. Intel had both, three years apart, and conflating them is how people misread Nvidia today.

The market bubble popped first, and it popped on a specific day. On September 21, 2000, Intel warned that revenue would miss, blaming a weak euro. The stock fell about a fifth in a session and erased on the order of $120 billion in a day, contagion ripping through the whole chip complex. Ten years of "priced for permanence" met one press release. Crucially, Intel the business was still dominant for years afterward; what broke in 2000 was the price, not the company. The plateau, missing mobile, losing the process lead to TSMC, came later and slower.

For Nvidia, the analog is the AI-capex question. The valuation assumes hyperscaler AI spending compounds for years and that Nvidia keeps almost all of it. The early-June 2026 pullback that knocked it briefly under $5T is a reminder that the price can correct hard while the business stays dominant, exactly the Intel-2000 pattern. The honest distinction: a price correction is a Tuesday; the platform shift is the decade. Watch which one you are actually looking at.

The most useful and least-told part of the Intel story is not the missed iPhone. It is that Intel spent decades unsure of its own core competency, and the one time it answered the question correctly, it saved the company. In 1985, crushed by Japanese memory makers, Andy Grove famously asked Gordon Moore what a new CEO would do, then walked back in and did it himself: Intel exited DRAM, the business it was founded on, and bet everything on microprocessors. That clarity, "we are a logic company, not a memory company", built the monopoly.

Intel then spent the next thirty years slowly losing that clarity. Was it a chip designer or a manufacturer? An IDM whose edge was owning both, or a products company trapped by failing fabs? It tried to be everything, delayed the hard call, and let TSMC, a company that did exactly one thing, make chips for everyone else, take the manufacturing crown. The irony of 2026 is that Intel has finally answered the question again: it is leaning hard into foundry, and it just won a marquee AI-chip manufacturing order from Google and is in talks to make chips for Apple, the very company whose defection symbolized its decline. That is why the stock has roughly quadrupled off its 2024 lows.

The lesson transfers directly to Nvidia, and it is sharper than the usual "don't get complacent." Nvidia's core competency is not the GPU; it is CUDA and the developer ecosystem around it. The day it starts defending the chip instead of the ecosystem, optimizing for this quarter's margin instead of the next decade's lock-in, is the day it starts becoming Intel, regardless of the share price.

If you remember a single chart from this piece, make it the next one. The thing that makes Nvidia genuinely different from Intel-at-the-top is that its market is still growing faster than it is losing share. Nvidia's slice of AI accelerators has slipped from about 90% to about 80% as AMD and the hyperscalers' custom chips scale, and yet its data-center revenue has gone from $27B to $216B across the same window. Intel, by 2005, was defending a flat pie; Nvidia is bleeding share into a flood. Both can be true: the monopoly is eroding at the margin and the absolute business is still compounding. That is the tension every Nvidia model has to hold in its head at once.

The startup opening. It is in portability and the picks-and-shovels of a multi-accelerator world: compiler and abstraction layers that make models vendor- and foundry-agnostic, inference-optimization tools for custom silicon, and the qualification services buyers need to run a credible second source. Every monopoly funds the market for its own alternative; the CUDA era will be no different.

1. Nvidia FY2026 results: record $216B revenue, ~$194B data center [8 min]

2. Washington Post archive: 'Intel Shares Plunge on Earnings Forecast' (Sept 22, 2000) [5 min]

3. Barron's 'Burning Up' (Mar 20, 2000), the call that marked the dot-com top [6 min]

4. TSMC passes $2 trillion: Taipei Times market-cap coverage [4 min]

5. Clayton Christensen, 'The Innovator's Dilemma': why incumbents miss the next platform [12 min]