The AI Industry Needs 18x More Revenue – Every Year, Forever
“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.”
OpenAI lost $5B+ in 2025. It projects $14B in losses for 2026. Most of that $37B in AI revenue? Generated at a loss.
AI infrastructure investment accounted for 92% of US GDP growth in H1 2025. What happens when it stops?
NVIDIA's revenue depends on companies whose spending depends on AI revenue that doesn't yet exist. The same dollars, circling.
LLMs Are a Breakthrough – But Breakthroughs Don't Guarantee $650B
The technology is real. The math isn't.
- Coding assistants saving developers hours per week
- Enterprise search across millions of documents
- Automated customer support and content generation
- Real productivity gains across Fortune 500 companies
$37B
Current generative AI industry revenue (2025)
- Adoption rates exceeding the smartphone
- New software markets worth hundreds of billions — that don't exist yet
- AGI-level capabilities — a step-change no one has shipped
- Revenue growth sustained indefinitely
$650B/yr
Revenue needed to justify current AI capex (JPMorgan)
“Current LLMs are not able to plan, reason, or think critically. They are fundamentally limited in ways that scaling will not fix.”
Meta's own head of AI research says the next step-change requires a different approach than what that $125B is buying.
“Almost all AI companies will go bankrupt, and much of the AI spending will be written off.”
The investor who shorted the 2008 housing market sees the same pattern in AI infrastructure spending.
The capex is priced for a future the technology hasn't reached yet. The gap doesn't close with incremental adoption – it requires either a step-change or adoption rates that exceed the smartphone.
A Composite Score Built on Real Signals – Updated Daily
No opinions. No predictions. Just data.
Widening spreads = credit stress = bearish
FRED, FINRA TRACE
↑12bps
Book-to-bill < 1.0 or TSMC deceleration = trouble ahead
SIA, TSMC IR, FRED
↑4pts
Rising puts + skew = smart money hedging = bearish
CBOE, OCC
↑2.1pts
Falling prices = oversupply = bearish
AWS, Azure, GCP, Vast.ai, Lambda Labs
↓3.2%
Approaching 100% = must borrow to fund capex = unsustainable
SEC quarterly filings
↑8pts
Decelerating growth = peak approaching = bearish
NVIDIA quarterly earnings via SEC EDGAR
↓2pts
Concentrated insider selling = insiders see risk = bearish
SEC Form 4 filings via EDGAR
↑3.5pts
Underperformance = equity investors losing confidence in DC demand
Market data (EQIX, DLR, VNQ); SEC quarterly filings
↓1.8pts
Shift from "accelerating AI" to "rationalizing AI" = demand problem
SEC EDGAR filings
↑5pts
If AI adopters underperform, the ROI thesis underlying hyperscaler spend is failing
Market data (SNOW, CRM, NOW, PLTR vs SPY)
Picks-and-shovels companies reflect physical buildout demand 6-12 months ahead
Market data (VRT, ETN vs XLI)
Dramatic outperformance vs. utilities = market pricing unsustainable AI power demand expectations
Market data (CEG, VST, NRG vs XLU)
Every number has a source. Every source is public.
Founding Member – $348/year $199 once, Pro access forever. 57 of 200 seats claimed.
Claim Your SeatMichael Burry Is Warning Again – And the Numbers Are Starting to Agree
- Debt-fueled construction boom
- Assumption: prices never fall
- Banks didn't understand their exposure
- Exotic instruments hid the risk
Burry shorted it. The rest of the market took two more years to notice.
- $600B in annual capex
- Assumption: demand will justify the buildout
- Hyperscalers extending depreciation schedules
- $14B in projected losses at OpenAI alone
The same pattern. Different asset class.
“All the capital being spent and lent by the richest companies on earth will not buy enough time – by the very definition of mania.”
January 21, 2026
The Hidden Accounting
GPUs have a 2-3 year product cycle. Hyperscalers are depreciating them over 5-6 years – extending their useful life to suppress expenses and inflate reported earnings. Burry estimates this earnings overstatement will understate costs by $176 billion across the Big Five from 2026-2028.
$176B
hidden depreciation costs
$176B hidden depreciation costs
“Massively ramping capex through purchase of Nvidia chips/servers on a 2-3 yr product cycle should not result in the extension of useful lives... Yet this is exactly what all the hyperscalers have done.”
November 10, 2025
5-6 years vs 2-3 years standard
5-6 years vs 2-3 years standard
5-6 years vs 2-3 years standard
5-6 years vs 2-3 years standard
5-6 years vs 2-3 years standard
Source: Burry's estimates from public 10-K filings
This Has Happened Before
In 2008 it was housing. In 2026 it's data centers.
“In 2008, banks discovered they owned far more US housing risk than their internal reports suggested. They might soon discover the same about data-center and digital infrastructure risk.”
Where We Are in the Cycle
Two views of the same question. Where the score has been – and where this pattern has gone before.
“If he had one failing in the dot-com cycle, it was being early... The housing bubble? It was being early.”
December 28, 2025
See It Coming
For the curious
- The gauge + verdict – always public
- Weekly newsletter
- Signal status dots
- 30-day composite score chart
- 1 composite score alert
For the believers
$348/year – pay once, forever
200 people will lock in lifetime access before launch. When they're gone, this price disappears.
- Everything in Pro, forever
- Daily score breakdown
- Signal-level drill-downs
- Score attribution – see what's moving
- Unlimited custom alerts (composite + per-signal)
- 90-day history with trend lines
- Historical bubble overlays
- Monthly deep-dive report
- Embeddable gauge widget
- API access (10 calls/hour)
Limited to 200 seats. Never offered again.
Dashboard and data powered by stressindex.ai
For the serious
- Daily score breakdown
- Signal-level drill-downs
- Score attribution – see what's moving
- Unlimited custom alerts (composite + per-signal)
- 90-day history with trend lines
- Historical bubble overlays
- Monthly deep-dive report
- Embeddable gauge widget
- API access (10 calls/hour)
Dashboard and data powered by stressindex.ai
Think the AI bubble will pop? Bet on it → Polymarket
Still doubt our data? See the full methodology
Subscriptions managed by stressindex.ai · Not affiliated with Michael Burry.