The $5 Trillion Question
NVIDIA's market capitalisation crossed $1 trillion in May 2023, $3 trillion in June 2024, $4 trillion in July 2025, and $5 trillion in October 2025 MarketWatch / Bloomberg. No company has ever appreciated by that much absolute value that quickly. It now trades at roughly 40–50 times trailing earnings — expensive, but not the 100-plus multiples that defined Cisco at its 2000 peak.
The concentration behind it is historically unprecedented. The largest ten US stocks account for roughly 35–40% of total S&P 500 market capitalisation Bloomberg / Goldman Sachs, 2025 — above the late-1990s peak, above 1929, above any point in the modern record. And the spending it implies is enormous: the four largest hyperscalers plan roughly $725 billion of AI infrastructure investment in 2026 alone, nearly double 2025's already-record level Company capex guidance, aggregated by Bloomberg.
A bubble can only be confirmed in retrospect. Railways, telephony, fibre optics, and the internet were all genuinely transformative and all catastrophic investments for their early financiers. Transformative technology and investor losses are not opposites — historically they are companions.
A Gallery of Booms
Every major infrastructure buildout since 1600 has followed a recognisable shape: a real technological breakthrough, a financing frenzy, overcapacity, a crash — and then the infrastructure quietly becoming indispensable while the original investors mostly lost their money.
| Episode | Peak | Peak-to-trough | Recovery time | Source / confidence |
|---|---|---|---|---|
| Dutch tulip mania | 1637 | ~99% (some contracts) | No organised market survived | Historical estimates Unresolved |
| South Sea bubble | 1720 | ~−80% | Decades; 1720 high never meaningfully revisited | Historical estimates Partly substantiated |
| UK railway mania | 1845 | −50–60% average; many lines bankrupt | Network survived; most shareholders did not recover | Historical estimates Partly substantiated |
| Wall Street crash | 1929 | −89% (Dow Jones) | ~25 years (1954) | Dow Jones index record Substantiated |
| Japanese asset bubble | 1989 | −82% (Nikkei 225) | ~34 years (recovered 2024) | Nikkei index record Substantiated |
| Dot-com crash | 2000 | −78% (NASDAQ, 5,048 → 1,114) | ~15 years (April 2015) | NASDAQ index record Substantiated |
| Telecom crash | 2000–02 | −90%+ sector; ~$2T value destroyed | Many names never recovered (WorldCom, Global Crossing bankrupt) | SEC filings / press Substantiated |
| US housing bubble | 2006 | −33% (Case-Shiller national) | ~6 years | S&P Case-Shiller Substantiated |
| Crypto cycle | 2021 | −77% (Bitcoin) | ~3 years (new high 2024) | Market record Substantiated |
| AI boom | 2025–? | — | — | In progress Unresolved |
The telecom crash is the most instructive comparison. Between 1996 and 2001, roughly $2 trillion was spent laying fibre optic cable Telecom crash post-mortems, Forbes / NYT. WorldCom, Global Crossing, and 360networks went bankrupt. Yet that "overbuilt" fibre became the physical substrate of the entire internet economy — the winners (Google, Netflix, Amazon) rode infrastructure built by people who lost everything.
NVIDIA vs Cisco: Two Trajectories
The chart below indexes both companies to 100 at the start of their signature run-ups — Cisco in January 1999 (~$71B market cap), NVIDIA in May 2023 ($1T). Cisco multiplied nearly 8× in 14 months, then lost 86% of its value. NVIDIA's ascent has been slower in percentage terms — 5× over 29 months — but it started from a base fourteen times larger, meaning the absolute dollars at risk are of a different order entirely.
The Ascent — and the Precedent
Market capitalisation, indexed to 100 at run-up start · Cisco Jan 1999–Oct 2001 vs NVIDIA May 2023–Oct 2025 · log scale · approximate values from public market data
Note NVIDIA's mid-2025 drawdown: after the DeepSeek R1 release in January 2025, NVIDIA briefly shed roughly $800 billion in market value in weeks Market record, Jan–Apr 2025 — the largest single-company drawdown in history, recovered within months. That episode is the bull case in miniature: demand shocks hit, and so far buyers have stepped back in.
What Is Different This Time
- Earnings are real. Unlike Cisco in 2000 or the dot-com cohort, the AI trade is led by companies with enormous verified profits. NVIDIA's fiscal 2025: $130.5B revenue, $72.9B net income NVIDIA 10-K, FY2025. The Magnificent Seven collectively earn over $250B a year.
- The buyers are solvent. The 1999–2000 boom was fuelled by retail margin debt and telecom vendor financing. Today's capex is funded by the largest cash piles in corporate history — though, as Thread 03 documents, that changed meaningfully in 2025 when hyperscalers issued $121B in new debt.
- Usage is real. ChatGPT reached 100 million users in two months. Enterprise AI spending is documented in audited filings, not projections. The adoption gap is real (see Thread 06) but nothing in prior manias had this level of verified uptake.
What Rhymes
- Concentration. US equity concentration is at its highest level in the modern record — above 1999 Bloomberg / Goldman Sachs.
- The revenue gap. Sequoia's 2023 "$600B question" — how much revenue must AI generate to justify chip spending — has been updated by Goldman Sachs (Oct 2024) and others: even at $725B of 2026 spend, documented AI revenue covers only a fraction of the required return Sequoia, Goldman Sachs research Partly substantiated.
- Circularity. Over $1 trillion of deals in which investments become revenue for the investor — documented in Thread 02 Substantiated.
- Insider behaviour. Venture and corporate insiders are monetising at record pace through secondary sales and debt-funded dividends while public markets absorb new supply FT / The Information, 2025–26 Partly substantiated.
Jim Chanos, who shorted the dot-com boom: the current market is "a bubble" — not because AI isn't real, but because the capital being deployed will earn returns far below its cost, exactly as it did in fibre, railways, and telephony. The technology wins; the financiers lose.
Confidence Assessment
| Claim | Status | Confidence |
|---|---|---|
| US market concentration is at a modern-era high | Index arithmetic, verifiable | Substantiated |
| AI leaders have real, audited earnings | SEC filings | Substantiated |
| Planned spend exceeds revenue demonstrably needed to justify it | Analyst consensus; assumptions contested | Partly substantiated |
| The current boom will end in a crash | Unfalsifiable in advance; base rates from history are high | Unresolved |
| "This time is different" | It always is, until it isn't | Opinion |