OFF-CYCLE RESEARCH / BUBBLE MAP · v1.0.0
The bubble can be real. The revolution can still win.
AGI may define the most consequential investment decade of our lives. That does not make every AGI stock a good investment at every price. We separate technological truth, economic capture and market expectations—and pre-commit the conditions under which a new long-term investor may begin, add or wait.
OOP-AGI-BUBBLE-2026-09-12 · EVIDENCE CUTOFF 2026-09-12
Our evidence does not support “the whole market is a dot-com bubble,” nor “AGI fundamentals make price irrelevant.” The better description is a genuine technological revolution with elevated valuations, extreme concentration and near-dot-com investment intensity—but also enormous realized revenue, cash flow and mostly non-systemic aggregate leverage. The risk is selective overpricing and future capital overhang, not yet a proven economy-wide credit bubble.
01 / BOTH CAN BE TRUE
Three truths that can coexist
The technology may transform the economy even if today’s leaders later lose capital. Railways, electricity and the internet produced both durable infrastructure and ruined shareholders.
A great company can be a bad investment when its price already requires an implausibly perfect future. Investment return is the difference between what happens and what the price already assumed.
A bubble is not one number. Price acceleration, issuance, turnover, financing, concentration and narrative reflexivity must be read together; even then, crash timing is not reliably forecastable.
02 / SEVEN SIGNALS
The AGI bubble clock
Price already asks for a lot
The S&P 500 forward P/E was 20.85× in April 2026 versus a 16.00× historical median. The Fed describes equity valuations as elevated and the equity premium as low.
source ↗The index is making a concentrated bet
The ten largest S&P 500 constituents were 37.8% of the index and information technology 37.9% on 31 August 2026. Diversifying by ticker is not the same as diversifying by economic factor.
source ↗The build-out already resembles a boom
By Q1 2026, the IP and equipment investment share of GDP had risen 0.8 percentage points above 2024 and sat only slightly below its 2000 peak. The pace is comparable to the late 1990s, from a higher starting level.
source ↗Unlike 1999, the cash register is ringing
NVIDIA reported $89.0B of quarterly Data Center revenue, up 117% year over year. Microsoft reported $331B annual revenue and positive free cash flow despite $41B of quarterly capital expenditure. These are vendor-reported economics, not proof that every buyer will earn an adequate return.
source ↗Not yet a 2008-style balance-sheet bubble
The Fed reports moderate household and business debt and resilient banks, while hedge-fund leverage remains near historical highs. AI capex has largely been cash-funded, but the Bank of England warns that external and debt finance may rise.
source ↗Credit is not pricing panic
The US high-yield option-adjusted spread was 2.70% on 10 September 2026. Tight spreads do not call a top, but they leave less compensation if AI expectations or growth disappoint.
source ↗There is still cash outside the trade
US money-market fund assets were $7.97T on 9 September 2026. This is not automatically bullish—much of it is operational cash—but it argues against a story in which every marginal dollar is already fully invested in AGI equities.
source ↗03 / HISTORICAL BASE RATES
What history does—and does not—tell us
Internet: correct technology, wrong price and ownership map
Over 90% of installed fiber was initially dark. The overbuild destroyed capital but later lowered costs for the digital economy. The infrastructure can win while early owners lose.
source ↗Adoption and uncertainty can create rational-looking bubbles
Technological revolutions can produce bubbles observable in hindsight but difficult to time beforehand, especially when adoption is rapid and uncertainty is high.
source ↗A run-up raises crash risk; it does not provide a clock
Historical industry run-ups substantially increase the probability of a severe drawdown. Turnover, issuance, volatility and the price path improve the diagnosis, but do not make the crash date knowable.
source ↗Leverage determines whether a valuation bust becomes systemic
Credit-to-GDP, property prices and debt-service burdens are better early warnings for banking crises than equities alone. Today’s AGI risk can hurt portfolios without yet resembling the housing-credit mechanism.
source ↗Creators can be overpaid while adopters are underpriced
Market value can concentrate too much in visible innovators relative to realized social value, while competitive adopters capture benefits elsewhere. AGI alpha may migrate from model builders to scarce complements and productive users.
source ↗04 / ENTRY PROTOCOL
When should a new long investor enter?
Do not wait for a ceremonial AGI announcement—and do not buy an exciting narrative all at once. For a new long investor, the first entry becomes research-eligible only where a company can survive an AGI delay, the price no longer requires perfection, monetization is observable and portfolio concentration remains tolerable. The strongest add zone is price compression plus an intact causal thesis plus improving evidence. One or two of those three is a reason to wait.
Four gates before capital
Survival gate
The business remains solvent and strategically relevant under a 3–5 year AGI delay, lower AI utilization and materially lower gross margins.
Expectations gate
A reverse DCF shows that the market-implied growth, margins and reinvestment are plausible under more than one scenario—not only a perfect AGI path.
Evidence gate
Revenue quality, utilization, customer economics or bottleneck rents are visible in primary filings and improve after accounting for capex, dilution and financing.
Portfolio gate
The position does not duplicate the same compute, duration, liquidity or regulatory factor already dominating the portfolio. No leverage is required to make the thesis attractive.
A staged decision map
Narrative outruns evidence
Research only. Do not use price momentum as evidence.
All four gates pass
An illustrative first tranche may become eligible; size remains a separate portfolio decision.
Price falls, thesis survives
Re-run the four gates. A 20–30% thematic or issuer drawdown is a review trigger, not an automatic buy signal.
Evidence improves after the reset
A later tranche may become eligible at a scheduled governance window; never average down through a broken thesis.
05 / NO CRYSTAL BALL
The forward monitoring contract
AGI claims, price gaps, dispersion, short interest and whether price moves precede or follow evidence.
High-yield spreads, hyperscaler bonds and leases, private-credit structures and hedge-fund leverage.
Data-center completions, utilization, power availability, accelerator lead times and revenue per deployed unit.
Incremental revenue, gross margin, free cash flow after capex, dilution and customer concentration.
Growth duration, terminal margins and reinvestment implied by each AGI-linked security price.
Top-ten weight, effective number of bets, correlation, beta and exposure to compute, duration and power.
IPOs, secondaries, stock compensation, project finance and capacity announcements versus delivered capacity.
Value split among model labs, compute, energy, infrastructure, software adopters, labor and consumers.
06 / FALSIFICATION
What would change our mind
- 01
The “selective, not systemic” diagnosis fails if debt-funded AI construction expands rapidly while credit standards weaken and losses migrate into banks or core funding markets.
- 02
The monetization defence weakens if incremental AI revenue persistently trails depreciation, leases, energy and dilution while customer return on AI spend remains unproven.
- 03
The long-horizon opportunity weakens if capability progress plateaus, inference economics stop improving or regulation structurally prevents deployment across economically important domains.
- 04
The entry protocol fails if it cannot be executed from point-in-time data without hindsight or if its gates are relaxed after prices rise.
07 / SOURCE CONTROL
Evidence ledger
Valuations, leverage, banking resilience and AI-related risks.
www.federalreserve.gov ↗PRIMARY · REGULATORFederal Reserve — Financial Stability Report accessibility tablesForward P/E series and historical median.
www.federalreserve.gov ↗PRIMARY · CENTRAL BANK RESEARCHFederal Reserve — Do Major Technology Advancements Lead to Overinvestment?Current investment intensity and capital-overhang framework.
www.federalreserve.gov ↗PRIMARY · CENTRAL BANK RESEARCHBank of England — All chips in?AI valuation, concentration and financing channels.
www.bankofengland.co.uk ↗PRIMARY · INDEX PROVIDERS&P Dow Jones Indices — S&P 500 factsIndex concentration and sector weights as of 31 August 2026.
www.spglobal.com ↗PRIMARY DISTRIBUTOR · REFERENCE ONLYFRED — ICE BofA US High Yield Index Option-Adjusted SpreadCurrent credit risk pricing; no historical series is redistributed here.
fred.stlouisfed.org ↗PRIMARY · INDUSTRY STATISTICSInvestment Company Institute — Money Market Fund AssetsWeekly US money-market fund assets.
www.ici.org ↗PRIMARY · ISSUERNVIDIA — FY2027 Q2 financial resultsRevenue, Data Center growth and gross margin.
nvidianews.nvidia.com ↗PRIMARY · ISSUERMicrosoft — FY2026 Q4 earningsRevenue, capex, cash flow and forward investment guidance.
www.microsoft.com ↗ACADEMIC · NBERGreenwood, Shleifer & You — Bubbles for FamaHistorical industry run-ups, crash probability and bubble characteristics.
www.nber.org ↗ACADEMIC · NBERPástor & Veronesi — Technological Revolutions and Stock PricesWhy rapid adoption and uncertainty can generate bubbles that are hard to time.
www.nber.org ↗ACADEMIC · NBERHaddad, Ho & Loualiche — Bubbles and the Value of InnovationThe gap between innovator valuations and the eventual distribution of value.
www.nber.org ↗PRIMARY · BIS RESEARCHBIS — Early warning indicators of banking crisesCredit, property and debt-service indicators of systemic crises.
www.bis.org ↗PRIMARY · FED RESEARCHFederal Reserve — Learning by InvestingTelecom overinvestment, dark fiber and later infrastructure value.
www.federalreserve.gov ↗PRIMARY · LAB CLAIMOpenAI — Research acceleration: a view from inside OpenAIAI-assisted research acceleration and its measurement limits.
openai.com ↗PRIMARY · LAB RESEARCHAnthropic Economic Index — January 2026 reportReliability and bottleneck constraints on realized productivity.
www.anthropic.com ↗