Portland, Oregon • Friday, February 6, 2026 • Special Report
BUBBLE & LOGIC
A data-driven comparative analysis of speculative manias

$600 Billion a Year in AI Spending. $650 Billion in Revenue That Doesn't Exist Yet.

The AI infrastructure boom rhymes with the dot-com era in structure but diverges in financial fundamentals. Here's what the data actually shows.

$600B+
2026 AI Capex
Combined hyperscaler spend
$650B
Revenue Gap
J.P. Morgan estimate
~45× P/E
NVIDIA TTM
vs. Cisco's 200× in 2000
25 yrs
Cisco Recovery
Peak to new high
−78%
NASDAQ 2000–02
$5T destroyed
34%
Mag 7 Weight
S&P 500 concentration
NVIDIA vs. Cisco: The CEO's company at peak euphoria

Cisco peaked at $80.06 on March 27, 2000, briefly becoming the world's most valuable company at ~$555 billion, trading at roughly 200–220× trailing earnings. CEO John Chambers proclaimed the internet was "the most significant economic shift since the Industrial Revolution." The stock then fell 89% and did not recover its 2000 high until December 2025.

NVIDIA sits at ~$4.5 trillion in market cap — 8× Cisco's nominal peak. But the valuation picture is fundamentally different. NVIDIA's trailing P/E of ~45× is one-fifth of Cisco's at its peak. The earnings are real, audited, and accelerating. The question is whether the growth rate can sustain the multiple.

Exhibit A
Valuation at Peak Euphoria (P/E Ratio)
Exhibit B
Revenue Scale Comparison ($B)
Full Comparison — Cisco (March 2000) vs. NVIDIA (February 2026)
MetricCisco 2000NVIDIA 2026Ratio
Market Cap$555B$4,500B8.1×
Trailing P/E~200×~45×0.23×
Forward P/EN/M~25×
Annual Revenue$19B$130.5B6.9×
Revenue Growth55%62–69%~1.2×
Gross Margin~65%73–75%~1.1×
Customer ConcentrationTelecom carriers~4 hyperscalersBoth high
Peak Drawdown−89%TBD
Recovery Time25 yearsTBD
"Is there going to be a train wreck? Yes — for those that aren't able to translate the technology into a sustainable competitive advantage."
John Chambers, Former Cisco CEO — October 2025
The Infrastructure Arms Race: $600 billion and counting

The dot-com telecom buildout consumed an estimated $500B–$2T between 1996–2001, financed almost entirely by debt. By 2001, 95% of installed fiber was dark. Today's spending dwarfs that era — but the spenders generate enormous cash flows rather than running on borrowed money. The risk isn't insolvency; it's whether returns ever justify the outlay.

Exhibit C
Hyperscaler AI Capital Expenditure — 2024 Actual vs. 2026 Projected ($B)

Critical difference: dot-com telecoms funded buildouts with debt and went bankrupt. Today's hyperscalers fund capex with cash flow — Alphabet generated $132B in net income in 2025. But debt IS growing: hyperscalers raised $108B in bonds during 2025, and J.P. Morgan projects $1.5T in investment-grade bonds needed over five years.

Exhibit D — The $600 Billion Question
AI Infrastructure Spending vs. AI Application Revenue ($B annual)
"Our biggest fear would be a repeat of the telecom and fiber buildout experience, where the revenue curve failed to materialize at a pace that justified continued investment."
J.P. Morgan — AI Spending Analysis, 2025
The Great Narrowing: concentration has shattered all records

The Magnificent Seven represent ~34% of the S&P 500, having peaked near 37% in October 2025. The top 10 account for ~40–41% — roughly 50% higher than the ~27% set during the dot-com peak. But the valuation story is more nuanced than raw weight suggests.

Exhibit E
S&P 500 Top-10 Concentration (%)
Exhibit F
Weight vs. Earnings Share — Then & Now

Michael Burry has taken a $1.1B short position against NVIDIA and Palantir, warning that passive indexing concentration could make any correction "more drawn-out" than the dot-com bust. Only Microsoft from the 2000 top 10 remains in today's top 10 — dominance is impermanent.

The Money Carousel: circular financing inflates demand signals

A pattern of circular AI financing has emerged where hyperscalers invest in AI startups that immediately spend the capital on those same hyperscalers' cloud infrastructure. The revenue looks organic until you trace where it came from.

Exhibit G
The Circular Flow — Invested vs. Returned to Cloud ($B)

Microsoft invested $13B in OpenAI, which spent $12.43B on Azure — a 96% recapture rate. Amazon invested $8B in Anthropic, which spent $2.66B on AWS — exceeding Anthropic's total revenue. Google invested ~$3B in Anthropic while signing a massive GCP deal. AI captured ~50% of all global VC funding in 2025 ($202B), with the vast majority flowing to cloud compute.

The Dot-Com Wreckage: a field guide to destruction & recovery

The NASDAQ fell 78% from its peak of 5,048 (March 10, 2000) to 1,114 (October 9, 2002), destroying approximately $5 trillion. The index did not recover until April 2015 — fifteen years later. At least 4,854 internet companies were acquired or shut down.

Exhibit H
Dot-Com Peak-to-Trough Drawdowns (%)
In Memoriam, 2000–2009
WorldCom
$9B fraud · $41B debt · Largest U.S. bankruptcy
−99.9%
Nortel Networks
Was 35% of Toronto Stock Exchange
−99.6%
Amazon
Survived with 10 months cash · Recovered 2009
−95.1%
Yahoo!
Eventually sold to Verizon for parts
−93.0%
Cisco Systems
$2.25B inventory write-off · Recovered Dec 2025
−89.0%
Intel
Never recovered dot-com high
−84.0%
NASDAQ Composite
$5T destroyed · 15-year recovery
−78.0%
GPU Hoarding, Price Collapse, & the Hype Cycle Turn

During 1999–2000, Cisco's supply chain suffered catastrophic double-ordering — a team needing 10,000 units triggered 30,000 in production. Today, Meta aimed for 350,000 H100s (~$10B worth), NVIDIA reports a $500B order backlog, and companies hoard GPUs as strategic assets. But the first cracks are visible.

Exhibit I
H100 Cloud Rental Price Collapse ($/GPU-hour)

Microsoft canceled leases for up to 2 GW of data center capacity. AWS paused international colocation discussions. Over 300 new GPU cloud providers entered the market in 2025, driving aggressive price competition. Gartner's 2025 Hype Cycle moved generative AI into the Trough of Disillusionment — projecting 2–5 years before the Plateau of Productivity. Less than 30% of CEOs reported satisfaction with GenAI ROI.

Bear Case

The Revenue Never Comes

The $650B gap persists. Circular financing unwinds. Capex slashed. AI names correct 50–80%. Cisco's 25-year recovery becomes the template.

Base Case

Rationalization, Not Ruin

30–50% correction in AI-levered names. Capex growth decelerates but doesn't collapse. Transformation plays out over 10–15 years. Patient capital wins.

Bull Case

The Gap Closes

Enterprise adoption hits escape velocity. Revenue accelerates faster than capex. NVIDIA's earnings justify valuation. The Trough is shallow and brief.

The people who made generational wealth from the internet didn't buy Pets.com in 1999 — they bought Amazon at $5.51 in 2001 after it had fallen 95%. The AI equivalent of that trade is probably still ahead of us.
Risk management isn't pessimism. It's patience with a plan.

Sources: NVIDIA IR, SEC filings, Sequoia Capital, J.P. Morgan, Gartner, IMF, Goldman Sachs, CNBC, Bank of England

For informational purposes only — Not financial advice — February 2026