Hyperscale data center at night
INVESTMENT RESEARCH • JULY 27, 2026

The AI Bubble
in 12 Charts

What investors are actually paying for — and where the market has already priced in perfection.

Valuation • Capex • Chips • Data centers • Power • Dot-com precedent
EXECUTIVE VERDICT
Short answer

AI is a real industrial boom inside a selective valuation bubble.

The cash flows, revenues and infrastructure are real. The speculative excess is concentrated in a handful of stocks whose prices require exceptional growth through the end of the decade.

$358B

2025 capex by the four largest U.S. hyperscalers.

60× sales

Palantir’s current multiple — above Cisco at the dot-com peak.

2× power

Global data-center electricity demand by 2030.

The evidence argues against “all hype” — and against “price no longer matters.”02
01
CHART 01

AI-exposed leaders now carry $18 trillion of market value

Chart 01

That is roughly one-third of the S&P 500 — but much of it sits in diversified, cash-generative platforms.

01 / 12
Source — Financial Datasets market snapshots and metrics, July 27, 202603
02
CHART 02

The bubble is concentrated in the upper-right corner

Chart 02

Fast growth supports rich multiples. The dangerous zone is where 20×–60× sales leaves almost no room for execution mistakes.

02 / 12
Source — Financial Datasets market metrics, July 27, 202604
03
CHART 03

Even 2028 consensus does not fully de-risk the priciest names

Chart 03

At unchanged prices, Palantir would still trade above 20× projected 2028 revenue. NVIDIA and Broadcom remain near 10×.

03 / 12
Source — Financial Datasets analyst estimates and market caps, July 27, 202605
04
CHART 04

The capex cycle has already gone vertical

Chart 04

Microsoft, Alphabet, Amazon and Meta increased annual capital spending from $140B to $358B in two years.

04 / 12
Source — Company filings via Financial Datasets; fiscal years as reported06
05
CHART 05

Three companies have already announced almost another $353B

Chart 05

Amazon, Alphabet and Meta alone plan 2026 spending comparable to the entire Big Four's 2025 outlay — before Microsoft.

05 / 12
Source — Amazon, Alphabet and Meta official 2026 guidance; midpoint used for Meta07
06
CHART 06

Cash generation is funding the boom — but the buffer is shrinking

Chart 06

Amazon spent 94% of 2025 operating cash flow on capex. The cycle is moving from surplus cash toward balance-sheet commitment.

06 / 12
Source — Company 2025 cash-flow statements via Financial Datasets08
07
CHART 07

The physical pipeline is huge — and mostly sold before completion

Chart 07

North America had 6.5 GW under construction in H1 2025; 72% was already preleased and primary-market vacancy was just 1.5%.

07 / 12
Source — CBRE North America Data Center Trends H1 202509
08
CHART 08

Chip demand looks less like a story and more like an industrial supercycle

Chart 08

Global semiconductor sales reached $792B in 2025 and WSTS projects roughly $975B in 2026.

08 / 12
Source — Semiconductor Industry Association and WSTS, 202610
09
CHART 09

NVIDIA has grown into the valuation faster than dot-com icons did

Chart 09

Revenue more than tripled in two years while operating margins stayed near 60% — real earnings power, not just narrative.

09 / 12
Source — NVIDIA FY2024–FY2026 10-K statements via Financial Datasets11
10
CHART 10

Power is becoming the binding constraint

Chart 10

U.S. data-center electricity demand could rise from 176 TWh in 2023 to 325–580 TWh by 2028.

10 / 12
Source — Lawrence Berkeley National Laboratory / U.S. Department of Energy12
11
CHART 11

The AI buildout is large enough to reshape national grids

Chart 11

Global data-center electricity demand is projected to more than double by 2030, with AI the primary growth driver.

11 / 12
Source — International Energy Agency, Energy and AI, 202513
12
CHART 12

The dot-com lesson is not ‘technology was fake’ — it is ‘price mattered’

Chart 12

The Nasdaq lost 78% after 2000. Today’s leaders have far stronger profits, but isolated valuations are already beyond Cisco-at-the-peak territory.

12 / 12
Source — FRED NASDAQ Composite; Cisco 2000 peak valuation; current Financial Datasets metrics14
BUBBLE SCORECARD

Five signals say “boom.” Two say “bubble.”

REAL REVENUE

NVIDIA revenue reached $216B with roughly 60% operating margins.

REAL BUYERS

Hyperscalers are paying from operating cash flow, not venture funding.

REAL BOTTLENECKS

Power, land, transformers and preleased capacity confirm physical demand.

PRICE RISK

20×–60× sales requires years of flawless execution and durable margins.

CAPEX RISK

Spending is rising faster than near-term monetization and free cash flow.

CYCLE RISK

Semiconductors and infrastructure can overshoot even when demand is secular.

Base case — the AI economy keeps expanding, while returns disperse sharply between infrastructure winners, platform owners, and overvalued application stocks.

Investment implication — own cash flows and bottlenecks; be skeptical of distant narratives.15
WHAT WOULD BREAK THE THESIS
“The market has not lost its mind about AI. It may have lost its discipline about what any one AI company is worth.”
Watch cloud AI revenue, GPU utilization, capex-to-cash-flow, data-center vacancy, power interconnection delays and 2028 revenue revisions.
The AI Bubble in 12 Charts16
METHODOLOGY & SOURCES

Built from filings, market data and primary infrastructure research

Markets & companies

Financial Datasets real-time market snapshots, financial metrics, analyst estimates and SEC-derived annual statements for NVIDIA, Microsoft, Alphabet, Amazon, Meta, Broadcom, Oracle, Palantir and Cisco. Market data timestamp — July 27, 2026. Fiscal-year comparisons follow each issuer’s reporting calendar.

Infrastructure & history

Amazon Q4 2025 earnings call; Alphabet Q2 2026 earnings call; Meta Q1 2026 results; CBRE North America Data Center Trends H1 2025; SIA and WSTS market releases; Lawrence Berkeley National Laboratory 2024 U.S. Data Center Energy Usage Report; IEA Energy and AI 2025; Federal Reserve Bank of St. Louis NASDAQ Composite series.

Valuation ratios are snapshots, not price targets. 2028 revenue figures are consensus estimates and can change materially. The selected companies are AI-exposed examples, not a diversified portfolio or a pure-play index.

Research cut-off — July 27, 202617