AI Heatmeter

Tracking AI infrastructure spending, economics, funding and valuations.

Conclusions

Watch. Revenue is catching up with infrastructure investment. Funding is broadly resilient, with pockets of neocloud dependence on external capital. GPU payback fits within the four-year hardware horizon. Valuation premiums remain elevated.

Infrastructure Buildout Levels

Watch

At current growth, annual AI revenue catches annual investment in 2.7 years. Revenue is growing faster than investment.

Demand Absorption

Lower pressure

GPU quotes show no broad pricing stress.

Infrastructure Economics

Lower pressure

The H100 scenario recovers investment within the four-year hardware horizon.

Funding Quality

Watch

Hyperscalers and total leverage are resilient; the smaller neocloud segment keeps funding on Watch.

Market Euphoria

Elevated pressure

Public valuation premiums carry demanding expectations; private premiums imply particularly demanding growth.

Infrastructure Buildout Levels

Watch

KPI 01

Investment / AI Revenue

4.86×

Investment (last quarter annualized): $943.02B
AI revenue (annualized run-rate): $194.19B

Investment / annualized AI revenue4.86×Watch

KPI 02

Capex vs Revenue Growth

-169.8 pp

Investment growth (quarter YoY): +114.5%
AI revenue growth (annualized run-rate YoY): +284.3%

Investment growth minus revenue growth-169.80 ppLower pressure

Revenue catch-up horizon

2.71 yr

Annual revenue reaches annual investment at current growth.

Growth-adjusted revenue catch-up2.71 yrWatch

Funding Quality

Watch

KPI 06

Capex Self-Funding Ratio

111.0%

Hyperscaler adjusted OCF (LTM): $691.29B
Hyperscaler investment (LTM): $622.90B
Neocloud adjusted OCF (LTM): $12.39B
Neocloud investment (LTM): $43.61B

Hyperscalers · self-funding+110.98%Lower pressure

Neoclouds · self-funding+28.40%Elevated pressure

KPI 07

Debt & Commitments

0.89×

Net obligations (latest balances): $523.82B
Operating cash flow (LTM): $585.92B
Revenue (LTM): $1.63T

Net recognized obligations / reported OCF0.89×Lower pressure

Net obligations / revenue: 0.32×

Blended funding signal

Investment weights: 93.5% hyperscalers / 6.5% neoclouds.

Size-weighted funding blend35.00Watch

Demand Absorption

Lower pressure

KPI 03

GPU Rental Price Index

+27.8%

H100 (13w ago → latest; $/GPU-hour): $3.196 → $3.805
H200 (13w ago → latest; $/GPU-hour): $4.000 → $5.145
B200 (13w ago → latest; $/GPU-hour): $6.250 → $7.990

GPU rental quote change · 13 weeks+27.84%Lower pressure

Infrastructure Economics

Lower pressure

KPI 05

GPU Infrastructure Payback

2.29 yr

H100 investment (upfront / GPU): $40,000
Cash contribution (annualized / GPU): $17,499
Paid occupancy / cash margin (assumed): 75% / 70%

H100 scenario payback2.29 yrLower pressure

Market Euphoria · Public Markets

Watch

KPI 08

Public valuation premiums

Nasdaq-100

+34.1%

AI basket

+72.8%

Nasdaq-100 P/E (LTM): 29.22×
Nasdaq-100 revenue multiple (LTM): 6.61×
AI basket P/E (LTM): 26.88×
AI basket revenue multiple (LTM): 10.20×
Historical Nasdaq benchmark (P/E / P/S): 26.75× / 4.16×

Nasdaq-100 premium+34.06%Watch

AI basket premium+72.82%Watch

Market Euphoria · Private Markets

Elevated pressure

KPI 09

Private valuation premium

+486.9%

Median private transaction multiple (annualized revenue): 24.42×
Nasdaq-100 historical revenue multiple: 4.16×

Private median premium vs Nasdaq historical P/S+486.94%Elevated pressure

RESEARCH FRAMEWORK · VERSION 0.2

Methodology & definitions

The economic rationale, definition, exact formula, sources, inclusion rules, warning interpretation and known limitations for each of the eight active KPIs (original numbering retained; KPI 4 retired). These definitions also power the Calculator view.

Philosophy: A bubble can exist without being close to bursting. We monitor investment strain, GPU rental pricing, infrastructure returns, financial vulnerability and valuation heat separately. We do not assign a probability or composite score without historical calibration.

FIVE-CATEGORY MAP

Infrastructure Buildout Levels

1. Investment / AI Revenue · 2. Capex vs Revenue Growth

Demand Absorption

3. GPU Rental Price Index

Infrastructure Economics

5. GPU Infrastructure Payback

Funding Quality

6. Capex Self-Funding Ratio · 7. Debt & Commitments

Market Euphoria

8. Public AI Valuation Premium · 9. Private AI Valuation Premium

Proposed monitoring limits

Analyst-proposed screening limits, not empirically calibrated. Green means lower pressure on this dimension, not a buy signal. Confidence is separate. Public trailing and private annualized revenue multiples are not directly comparable.

Growth-adjusted buildout and size-weighted funding

ln(investment/revenue) / ln((1+revenue_growth)/(1+investment_growth)); assumes current growth rates persist; compares annual revenue and annual investment rates, not cumulative cash recovery.

Half investment-weighted self-funding pressure plus half aggregate leverage pressure. A red funding subgroup with at least 5% of tracked investment sets a minimum pressure of 35 (Watch near the green boundary).

Overall Boom / Bubble thermometer

The five category signals have equal weight: green = 0, amber = 50 and red = 100. The overall position is their arithmetic mean. Green is below 33⅓, amber is 33⅓ to below 66⅔, and red starts at 66⅔. The opening word follows the overall color: Boom for green, Watch for amber and Bubble for red. All five category signals must be available. Buildout uses the revenue catch-up horizon. Funding uses investment-weighted self-funding plus aggregate leverage, with a material-risk Watch floor. Other categories use their underlying signals. Funding contributes its blended pressure position; other colors contribute 0, 50 or 100.

KPI / displayed bar basis	Green	Amber	Red	Rationale
1 · Investment / annualized AI revenue	< 3×	3–< 6×	≥ 6×	Three times revenue already represents substantial capital intensity; six times flags a larger monetization burden.
2 · Investment growth minus revenue growth	≤ 0 pp	> 0–20 pp	> 20 pp	Investment growing faster than revenue raises absorption pressure; a gap above 20 points warrants closer review.
3 · GPU rental quote change · 13 weeks	> −10%	> −25%–−10%	≤ −25%	A moderate decline may reflect efficiency and hardware aging; a 25% quarterly decline flags pricing pressure, not proven weak utilization.
5 · H100 scenario payback	≤ 4 years	> 4–< 5 years	≥ 5 years	User-selected recovery horizon: payback within four years is green, above four and below five is Watch, and five years or longer is red.
6 · Capex self-funding · observed subset	≥ 100%	50–< 100%	< 50%	100% covers the measured investment; below 50% indicates substantial reliance on external financing.
7 · Net recognized obligations / reported OCF	≤ 2×	> 2–4×	> 4×	Two and four years of reported OCF are proposed leverage screens. The covered aggregate omits additional commitments and advance adjustments.
8 · Valuation premium vs Nasdaq history	< 25%	25–< 75%	≥ 75%	Mean of the earnings and revenue multiple premiums versus their respective Nasdaq historical averages.
9 · Private median premium vs Nasdaq historical P/S	< 25%	25–< 75%	≥ 75%	Private median annualized-revenue multiple versus Nasdaq historical P/S. Different revenue definitions remain documented in the calculator.

Bars show their numeric scale and clamp out-of-range markers at the edge. Missing values have no marker. KPI 6 shows the two funding groups separately; KPI 7 uses only the matched covered subset. KPI 8's bar uses revenue multiples even though its headline shows P/E. These supporting screens do not fill unavailable formal KPIs.

KPI 01 / Infrastructure Buildout Levels

Investment / AI Revenue

Is investment running ahead of monetization?

Why we track it

Capital investment precedes revenue, but an expanding infrastructure-to-revenue imbalance is a defining feature of unsustainable booms. This measures the scale of the bet relative to customer payments.

Formula

Σ (LTM cash capex + new financed equipment) ÷ LTM deduplicated end-customer AI revenue

Construction and inputs

Fixed 11-company basket including hyperscalers, neoclouds and xAI. Count actual investment, including newly financed assets, not payments on existing finance leases. Revenue is bottom-up vendor reporting, net of estimated intermediary overlap. Use matched LTM periods.

Bubble risk interpretation

Higher ratio and sustained upward drift increases industry strain.

Important caveats

Do not divide partial reported capex by an annualized AI revenue estimate and label it a published KPI. Until matched LTM revenue and lease additions are available, only a partial comparison can be shown.

Current data limitations

Cash capex 11/11; financed assets 10/11, including approved proxies.

Primary source families

SEC 10-Q/10-K XBRL company facts
Company lease and equipment notes
Bottom-up AI supplier revenue register

KPI 02 / Infrastructure Buildout Levels

Capex vs Revenue Growth

Is the imbalance widening or narrowing?

Why we track it

A forward-looking direction indicator. Even if investment/revenue is high, much faster revenue growth can repair the economics before a correction occurs.

Formula

100 × [(Current LTM capex / prior LTM capex − 1) − (Current LTM AI revenue / prior LTM AI revenue − 1)]

Construction and inputs

Calculate year-over-year growth from matched LTM observations for the same company universe and the same deduplication model. A separately labelled, low-confidence preview compares mixed-date annualized 2026 and 2025 revenue estimates with quarterly investment growth; its assumptions and prior overlap calculation are shown in the audit trail.

Bubble risk interpretation

Positive and rising percentage-point gap indicates deterioration; negative gap indicates catch-up.

Important caveats

Newly tracked companies must be backfilled historically or excluded from both periods. A denominator reclassification can create false growth.

Current data limitations

Growth comparison inherits KPI 1’s coverage and proxy limitations.

Primary source families

Same underlying datasets as KPI 1

KPI 03 / Demand Absorption

GPU Rental Price Index

Is the market price of AI compute capacity weakening?

Why we track it

Market rental prices are a relatively timely signal of demand/supply tightness and may turn before reported earnings.

Formula

Median across H100/H200/B200 of [(current matched GPU hourly rental quote ÷ quote 13 weeks earlier) − 1] × 100

Construction and inputs

Use weekly GetDeploying provider-median on-demand USD/GPU-hour quotes, matched by GPU slug and exact 13/52-week dates with a common anchor. Provider identities and configuration fields are unavailable in the aggregate, so this measures changing-panel quoted medians. The latest week is provisional; dated external level checks use separate definitions.

Bubble risk interpretation

Broad and persistent declines across new and mature GPUs are a potential capacity or competitive-pricing warning.

Important caveats

Listed prices are not realized contract prices. Changing seller panel and generational depreciation create noise. A single H100 change is not a full index.

Current data limitations

Matched prices: 3/3 models at 13 weeks; 3/3 at 52 weeks.

Primary source families

GetDeploying weekly rental dataset, CC BY 4.0
GPU Economy price series as cross-check

KPI 05 / Infrastructure Economics

GPU Infrastructure Payback

Will investments recover cash before equipment becomes economically obsolete?

Why we track it

Even strong demand can coexist with unattractive returns if installed GPUs cost too much, are underutilized or rental prices fall.

Formula

Fully installed GPU capex ÷ [GPU-hour price × 8,760 × paid occupancy × contribution margin]

Construction and inputs

Standardized H100, H200, B200 model; median only once costs for all three are validated. Cross-check against separately dated provider disclosures: CoreWeave historical committed contracts, Nebius new Q2 2026 deals, and IREN selected recent contracts. Do not average these distinct definitions.

Bubble risk interpretation

Payback approaching or exceeding economically useful hardware life indicates thinner returns and higher overinvestment risk.

Important caveats

Advertised rental quote is not realized price; H100 $40K cost is an equipment-level scenario that excludes some data-center capex. Contract paybacks are management forecasts.

Current data limitations

Payback is scenario-based; the H100 cost excludes some infrastructure.

Primary source families

GetDeploying GPU rental quotes
Hardware pricing disclosures / industry benchmarks
CoreWeave, Nebius and IREN management contract economics

KPI 06 / Funding Quality

Capex Self-Funding Ratio

Can infrastructure expansion be financed from cash generated internally?

Why we track it

Externally financed investment is more exposed to credit tightening, equity market shocks and failure to refinance. Hyperscalers and neoclouds should be shown separately.

Formula

(LTM operating cash flow − financing-related customer advances) ÷ (LTM cash capex + newly financed equipment)

Construction and inputs

Apply the same company coverage and cash/lease definitions each period; exclude material advances functioning as project financing. Show hyperscaler and neocloud aggregates separately.

Bubble risk interpretation

Sustained declining self-funding, especially among leveraged providers, increases financing vulnerability.

Important caveats

Advances may be embedded in operating cash flow and equipment may arrive via finance leases. Do not interpret missing lease additions as zero; incomplete basket is only a partial reading.

Current data limitations

Self-funding ratios remain partial with the approved investment proxies.

Primary source families

SEC cash flow statements via companyfacts API
Lease commitments disclosures
Customer-prepayment footnotes

KPI 07 / Funding Quality

Debt & Commitments

How large are binding financial obligations relative to ongoing operating strength?

Why we track it

Infrastructure bubbles often break through financing rather than technology failure. Debt-like obligations plus contracted buildout can require refinancing ahead of expected revenue.

Formula

Primary: (debt + recognized finance/operating leases − cash + PV of additional binding commitments) ÷ adjusted LTM OCF. Secondary: (debt + recognized leases) ÷ LTM revenue. Unadjusted supporting ratio: (financial debt − cash and equivalents) ÷ reported LTM OCF; available independently of missing leases.

Construction and inputs

Recognized leverage includes debt and both lease types; track uncommenced leases separately and discount additional contractual payments consistently. Include debt/revenue for loss-making firms. Show maturities as context.

Bubble risk interpretation

Higher obligations / cash flows, especially with near-term maturities and low self-funding, raise refinancing risk.

Important caveats

Do not add undiscounted lease payments to recognized present-value liabilities. OCF may be negative: label the ratio not meaningful, retain debt/revenue. Do not use net debt for debt/revenue.

Current data limitations

Recognized debt and leases are distinct from unvalued future commitments.

Primary source families

SEC debt and lease notes
Binding contracts and maturity schedules
Company revenue and cash-flow statements

KPI 08 / Market Euphoria

Public Valuation Premiums

How far are Nasdaq-100 and AI basket valuations above the historical Nasdaq benchmark?

Why we track it

AI enthusiasm becomes speculative when valuation expectations outpace future earnings power. Distinguish general technology-market heat from an AI-specific valuation premium.

Formula

Mean of [(P/E ÷ historical Nasdaq P/E − 1) × 100] and [(P/S ÷ historical Nasdaq P/S − 1) × 100], calculated separately for Nasdaq-100 and the AI basket.

Construction and inputs

Use the matched October 20, 2026 GuruFocus Nasdaq snapshot and publisher-reported long-term means. AI basket compares to the same Nasdaq benchmark; it is not a premium to the basket own history.

Bubble risk interpretation

Persistent large premium above historical norms, especially if earnings revisions weaken.

Important caveats

Publisher historical start/end dates are not disclosed. These are long-term averages, not verified ten-year means. AI basket earnings and sales definitions differ from the index. See the dated benchmark review.

Current data limitations

Public premiums use a Nasdaq long-term benchmark; they are not AI-basket own-history premiums.

Primary source families

Siblis Nasdaq-100 P/E and forward P/E
SEC historical financial statements
Paid or licensed point-in-time historical share prices / consensus forecasts

KPI 09 / Market Euphoria

Private Valuation Premium

How far is the median private transaction multiple above the historical Nasdaq revenue multiple?

Why we track it

A private-market premium provides context for speculative financing even when public multiples appear more restrained.

Formula

(Median private equity valuation / annualized revenue multiple ÷ historical Nasdaq P/S − 1) × 100.

Construction and inputs

Use the median of six transaction multiples; compare only with the historical equity P/S benchmark. Display the median multiple as numerical context. Underlying transaction identities remain in the research registry.

Bubble risk interpretation

Very high or expanding premium not justified by relative expected growth increases valuation risk.

Important caveats

Private annualized revenue and public LTM sales are different bases. Funding valuations may precede revenue disclosures. The result is a reference premium, not comparable forward EV/revenue.

Current data limitations

Equity valuation / annualized revenue is a discovery proxy, not forward EV/revenue.

Primary source families

Verified funding rounds and tender valuations
Company revenue announcements / financial statements
Consensus next-12-month public-company revenue data

Data governance & calculation rules

Unit of analysis

Capex and funding use the fixed 11-investor research registry. KPI 7 uses consolidated SpaceX in place of its AI segment. Public valuation uses 16 issuers; private valuation uses six. Keep these scopes distinct.

Time window

Prefer matching last-12-month numerator and denominator for flow ratios. Never treat annualized monthly revenue as recognized LTM revenue.

Leases & financing

Add newly financed equipment to capex when acquired; do not count future lease repayments as new capex. Existing operating leases are generally on US GAAP balance sheets.

Economic scenarios

Occupancy and effective price expectations are currently unobserved assumptions. Keep forecasts distinct from financial disclosures.

Evidence hierarchy

Audited/reported data → source-derived arithmetic → external estimate/forecast → declared assumption. Do not allow missing observations to silently become zero.

Historical calibration

Buildout uses the growth-adjusted revenue catch-up horizon. Funding blends investment-weighted self-funding and aggregate leverage with a material-risk Watch floor. Other categories use their available KPI readings. Unknown inputs remain unavailable.