AI Bubble Risk: Circular Cash & Low Demand Impact Assessment

The recent AI news flow confirms that the concerns about circular investments and lack of AI demand are real and actively debated. Here’s what’s been reported in just the last 6 weeks:

• Nvidia reportedly backstopping $250B in OpenAI data center financing — Jim Chanos and Michael Burry both called this out explicitly as vendor financing, echoing the Lucent/dot-com playbook
• Big Tech off-balance-sheet AI debt reportedly tops $1.65 trillion (Nikkei study, Ed Zitron quoted)
• Apollo’s Torsten Slok stated the AI capex boom is growing nearly 2x as fast as the mid-2000s housing bubble as a % of GDP
• Amazon, Alphabet lead “AI debt tsunami” blamed for 20-year high in long-term interest rates.
• Hyperscaler AI capex projected to hit $916B in 2026 and potentially $1.17T in 2027

This is real systemic risk. But the counterargument is also data-driven — earnings are actually coming in strong.

The Top AI Player Performance

Here’s the current EV/EBITDA trajectory (the most honest valuation metric for these companies):

*GOOGL’s expansion is distorted by one-time investment gains from SpaceX and Anthropic (~$112B net income in Q2 is anomalous).

Are the PE Ratios Falling?

The compression is primarily earnings-driven, not panic selling. Net income has exploded:

• NVDA: $16.6B/quarter (Q2 2025) → $58.3B (Q1 2027) — 3.5x growth in earnings in 18 months
• MSFT: $24.7B → $35.8B — steady compounding
• AMZN: $15.3B → $62.7B — dramatic ramp

This means multiples fell because earnings grew faster than stock prices — not because the market is pricing in bubble risk. That’s actually the bullish interpretation. The bears would say: “Yes, but those earnings are funded by circular capex, and when the music stops, earnings collapse.”

The Market Reaction

QQQ is up ~52% over 2 years vs SPY +39.5%. However, it’s already off its recent high of ~$740 (Jun 2026), and the early 2026 selloff took it from $626 → $562 in about 5 weeks.

Analyst Sentiment Right Now

This is genuinely split at the top:

Bears (notable names actively shorting/warning):

• Michael Burry — expanded shorts on NVDA and MU, calls it a bubble
• Jim Chanos — “closer to a ’99-type moment than ’97”
• Steve Eisman (“Big Short”) — “AI has no moats, not a recipe for longevity”
• Aswath Damodaran (“Dean of Valuation”) — “Microsoft, Amazon, Meta, Google are collectively overinvesting; that’s betting, not investing”
• Apollo’s Torsten Slok — $500B software debt wall ahead, Fed hike risk

Bulls (sell-side consensus still overwhelmingly buy):

• Virtually every major bank (JPMorgan, Goldman, Morgan Stanley, Wells Fargo) still rates NVDA, MSFT, GOOGL as Buy
• MSFT surged 9% after Q4 earnings — Azure +43%, 30M Copilot paid seats, $214B full-year cloud
• GOOGL Cloud +82%, $514B backlog
• NVDA net income $58.3B/quarter

If the AI Bubble Bursts: Impact on QQQ

This is necessarily speculative — nobody knows the timing. But here’s the data-driven framework:

Nasdaq 100 / QQQ: The top 5-6 AI-related names (NVDA, MSFT, GOOGL, META, AMZN, AAPL) represent 60%+ of QQQ’s weight. A structural re-rating would be severe:

• If multiples reverted to long-run tech averages (~15x EV/EBITDA), NVDA would lose 50%+ from current levels even with good earnings
• In a hard bubble burst (earnings also collapse), drawdown of 50-70% for QQQ is analogous to the 2000-2002 Nasdaq, which fell 83%

Broader economy (S&P 500 / SPY):

• AI capex has been a major driver of GDP. Hyperscalers spent $600B+ on capex in the last 12 months. If that reverses, it hits GDP, employment (data center construction, chip fabs), and credit markets.
• Long rates are already at 20-year highs, partly due to hyperscaler debt issuance. A bust could sharply spike credit spreads and tighten financial conditions.
• However, vs. 2000, the AI companies actually have real earnings — NVDA, GOOGL, AMZN are all highly profitable. This is different from Pets.com.

Is Risk Being Priced In?

Partially, but mostly no. Here’s the breakdown:

1. Multiple compression is real — MSFT EV/EBITDA dropped from 23.5x to 14.1x. But it’s because earnings grew, not because the market priced in disaster.
2. CDS spreads on hyperscalers have doubled since 2025 — credit markets are starting to signal concern. This is a leading indicator.
3. QQQ is still at all-time highs — the market consensus is not pricing in a bust.
4. Polymarket traders give only ~16% odds of a “bubble burst” — the crowd is still bullish.
5. The sell-side has virtually zero sell ratings on the Magnificent 7

Bottom line: the market is pricing in a soft landing / continued AI monetization, not a bust.

If you believe the bubble thesis, the risk/reward for QQQ at current levels is not favorable. If you believe in the earnings trajectory, valuations are actually more reasonable now than they were 18 months ago.

What To Watch as Leading Indicators of a Bust

• Microsoft CFO’s comment: “we can slow GPU spending if demand cools” — this is the canary. Watch Azure growth rate deceleration.
• NVDA financing deals — if more circular vendor financing deals emerge, credit risk jumps
• Alphabet going cash flow negative — already happened in Q2 2026. If this persists, the market might panic
• META’s Q3 2025 net income was only $2.7B (vs $26.8B in Q1 2026) — huge earnings volatility is the norm, not the exception
• AI model pricing collapse — Box CEO and multiple analysts warn open-weight models could crash pricing. If inference becomes a commodity, capex returns evaporate.

Should You Buy QQQ

I can’t make that call for you — this is not trading advice. But here’s the factual tension:

The risk here is asymmetric and non-linear — if the bull case holds, you get modest further upside from already elevated prices. If the bear case materializes, the downside is severe.