Nvidia just dropped the most anticipated earnings report of the year: $57 billion in quarterly revenue, +62 % YoY, beat every Wall Street estimate by a mile.
Jensen Huang took the mic and basically said: “AI bubble? What bubble? We don’t see it.”
The stock popped, headlines screamed “AI fears overblown,” and the usual suspects declared the bubble debate officially dead.
I’m calling nonsense.
This wasn’t proof the AI boom is healthy. It was the loudest warning yet that we’re standing on a $7+ trillion house of cards — and the bottom cards are already sliding out.
The Four-Layer Fragile Cascade That 93 % of Nvidia’s Revenue Depends On
Layer 1: The Data Centers Are Bleeding Cash (Even at Peak Pricing)
~93 % of Nvidia’s revenue now comes from data-center GPUs.
You’d think with GPU compute prices at all-time highs and hyperscalers in a mad rush to add capacity, these data centers would be printing money.
They’re not.
Latest estimates (Praetorian Capital, Epoch AI, and leaked capex models):
- New AI hyperscale facilities built in 2024–2027 will generate ≈ $16–22 B in annual revenue at full utilization
- Annual depreciation + power alone: ≈ $42–48 B
That’s negative $25 B gross margin before you pay a single engineer or kilowatt of electricity.
Translation: Even while demand is “sky-high,” the core infrastructure buying Nvidia’s chips is deeply unprofitable — and that’s using today’s inflated GPU rental rates. Once supply catches up (and it will in 2026–2027), prices crash and the losses get uglier.
Layer 2: The AI Labs Are Burning Money Faster Than Ever
The customers renting those data centers (OpenAI, Anthropic, xAI, Inflection clones, etc.) are in even worse shape.
OpenAI’s own internal forecasts leaked in late 2024:
- 2025 expected loss: $14–27 B
- 2029 projected annual loss (on their own bullish revenue numbers): >$200 B
That’s not a path to profitability — that’s a death spiral.
Why? Diminishing returns on the compute frontier.
Each new model generation costs 5–10× more to train but delivers only marginal capability gains. Meanwhile, enterprise adoption has slammed into the “hallucination wall”:
- MIT / Rand / Carnegie Mellon studies (2024–2025): 82–96 % of real-world AI pilots fail to show positive ROI once you account for human oversight and error correction.
- Result → revenue growth collapsing from 250 % YoY → 55–70 % YoY → soon single digits while compute spend keeps 4–6×ing.
Layer 3: The Money Spigot Is Already Closing
Where did all the cash come from in the first place?
Phase 1: Big-Tech cash hoards (mostly spent) Phase 2: Circular cross-investments (Nvidia → OpenAI → Oracle → Microsoft → back to Nvidia) Phase 3: Debt — and this is where it gets scary.
2023–2025: Big Tech + AI pure-plays issued ≈ $1.3 trillion in bonds to fund AI capex.
But the bond market just blinked.
CDS (credit default swap) spreads on AI-heavy debt — the “insurance” investors buy in case of default — exploded from $25 B notional in Q1-Q3 2025 to over $140 B in Q4 alone.
That’s the market screaming: “We no longer believe these cash flows are coming back.”
Higher CDS → higher borrowing costs → less new debt → capex cuts → fewer GPU orders → Nvidia revenue cliff.
We’re watching the loop reverse in real time.
Layer 4: Nvidia Itself — The Most Undiversified Trillion-Dollar Company in History
Current Nvidia market cap: ~$3.4 T ~93 % of current revenue and nearly 100 % of the growth story tied to the three layers above.
Forward P/E is still 38× on numbers that assume the debt keeps flowing and hyperscalers keep buying forever.
When (not if) the capex cycle turns:
- Consensus 2027 revenue estimates get cut 60–80 %
- Multiple compresses from 38× → low-20s overnight
- That’s a 75–90 % drawdown from today’s price using simple math.
We’ve seen this movie before — Cisco in 2000, TSMC in 2021 crypto winter, anything tied to a single mega-theme that “can’t possibly fail.”
Why Wall Street Cheered Anyway (The Shovel-Seller Paradox)
Investors aren’t stupid. They see the same math.
But right now Nvidia is the premium pick-and-shovel play in a modern gold rush. As long as the rush continues for another 4–8 quarters, early shovel-sellers still make life-changing money — even if every miner goes bankrupt later.
That’s why the post-earnings pop happened. Not because the bubble was disproven, but because the music is still playing and Nvidia has the best seat.
Bottom Line — This Wasn’t Disproof. It Was the Top Signal.
Strong earnings at the absolute peak of capex intensity, while every downstream layer is hemorrhaging cash and the debt market is flashing red, is not evidence of health.
It’s the classic signature of a bubble in its final manic phase.
We’re not in 1995 internet. We’re in 1999 — right before the adults walk in, turn off the credit spigot, and the whole thing comes crashing down.
Nvidia’s monster quarter didn’t kill the AI bubble narrative. It just showed us exactly how the collapse will happen — and how fast.
Stay sharp. Position accordingly.