The One-Liner: "Bubble" Is Not a Single-Choice Question
Whenever someone shouts "AI is a bubble," you should first ask: "Which layer are you talking about?"
An industry bubble? An asset price bubble? Or a profits bubble? Discussing these three together is like lumping onions, garlic, and chives together as "alliums" — technically correct, but use the wrong one in a dish and the whole meal is ruined.
---
Layer 1: The Industry Bubble (Already Refuted)
Definition: Capex is deployed but nobody uses it; infrastructure sits idle.
This is the easiest layer to refute.
- Anthropic's inference gross margin is 70%. Not 7%, not 17% — 70%. Every 100 dollars of revenue leaves 70 dollars after direct costs. That's higher than Salesforce's software margins, and miles above Starbucks.
- ARR grew 10x in a year. Customers aren't "trying it out" — they sign long-term contracts, prepay, and queue for token quotas. Micron just signed a $22 billion long-term supply agreement. Customers locking in multi-year capacity is unheard of in traditional memory.
- The hardware chain rose 2-5x in four months. This isn't idle speculative capital; it's physical scarcity. An EUV lithography machine costs $200 million, and ASML's order book is three years deep. This isn't "hype" — it's "can't be built fast enough."
- The memory sector's PE was only 3-4x at the end of last year. That priced in near-bankruptcy. This year it re-rated to 10x PE — not a crazy spike, but a repair from "extremely undervalued" to "reasonable, still cheap."
- Analogy: oil tanker stocks during the 1970s oil crisis topped out at 3-5x PE. Markets knew the windfall wasn't sustainable. Memory faces the same logic — the market is already applying a discount.
- Leading platforms trade at 7-8x PE, versus the Nasdaq average of 25x and Nvidia above 50x. Memory valuations are downright "modest."
- Jim Chanos (the man who shorted Enron and Tesla) is also warning about an AI bubble, but his angle is smart: not shorting chip stocks, but shorting cloud providers' "depreciation time bomb." Chipmakers recognize revenue and profit immediately; cloud providers capitalize the costs — beautiful balance sheets now, ugly income statements once depreciation kicks in.
- 2024-2027: Extreme supply-demand mismatch; memory makers print money
- 2027-2029: New capacity releases; prices start to soften
- 2029+: Supply and demand balance; ROIC normalizes (20-30%)
- Howard Marks, *Is It a Bubble?* (Oaktree Memo, Dec 2025)
- Jim Chanos AI Bubble Warning (Jun 2026)
- Micron FY2026 Q3: revenue $41.46B (+345.7% YoY), gross margin 84.9%
- Micron $22B long-term supply agreement disclosure
- Korea stock market's 5th circuit breaker of 2026 (June 26; SK Hynix -9%)
- CICC: global AI is not in a full bubble, but crowding is high
Conclusion: This layer of the onion is hollow. Demand is real, payment is real, capacity bottlenecks are real.
---
Layer 2: The Asset Price Bubble (Valuations Aren't Absurd)
Definition: Stock prices far exceed intrinsic value; PE ratios defy gravity.
This layer needs careful peeling.
Conclusion: There's some froth in this layer, but not much. The market is already pricing memory as a cyclical.
---
Layer 3: The Profits Bubble (Real, But Has a Physical Ceiling)
Definition: Supply-demand mismatch creates extraordinary ROIC, but it cannot last.
This layer is solid.
Memory's ROIC is currently 200-400%. For reference: Apple is 30%, Moutai 25%. This isn't a "good business" — it's a money printer.
But the printer has physical limits:
1. A new fab takes 3-5 years from approval to mass production. Today's high prices reflect 2021-2022 investment decisions. Current expansion plans won't release capacity until 2028-2029. 2. ASML monopolizes EUV equipment, with capacity growth capped at ~30% annually. Not that they don't want to expand — the lithography machines can't be built faster. 3. Customer qualification locks. HBM requires 6-12 months of customer validation. Once Micron, SK Hynix, or Samsung passes qualification, switching costs are enormous.
So the "profit bubble" script:
Three years is the physical ceiling — not a prophecy, but a hard constraint of fab construction cycles.
---
Howard Marks' Framework: Is It Different This Time?
Oaktree's Howard Marks wrote a memo last year titled *Is It a Bubble?* His core point:
> "Our era is an extraordinary moment in world history. A transformative technology is rising, whose proponents claim it will forever change the world." > > "Early participants reap huge gains while onlookers feel intense envy and regret, piling in driven by the fear of missing out."
This is the classic bubble recipe. But Marks also reminds us: not every bull market is a bubble, and not every bubble bursts.
What's different about AI this time: 1. Cash flows already exist (Anthropic's 70% gross margin, Micron's 84.9%), unlike the dot-com "eyeballs economy" 2. Healthier capital structures (Big Tech total debt of $385 billion, leverage ~20% below historical heavy-investment cycles) 3. Long-term contracts lock in demand (Micron's $22 billion, Nvidia's nearly $100 billion in purchase commitments)
But the risks are real too: 1. The depreciation bomb (cloud providers' capitalized costs will eventually enter the depreciation cycle) 2. End-demand suppression (memory prices up 165%; smartphones up 7%, laptops up 15%, slowing diffusion) 3. Geopolitics (Korea's market hit its 5th circuit breaker of the year; Middle East conflict affects energy costs and rate paths)
---
Summary Comparison: AI Is Not Dot-Com 2.0
| Dimension | 2000 Dot-Com Bubble | 2026 AI Cycle | |------|----------------|--------------| | Revenue | Almost none | Real and fast-growing | | Gross margin | Negative | 70-85% | | Valuation | Infinite (no profits) | Memory at 10x PE, already discounted | | Capital structure | High leverage, cash burn | Strong big-tech balance sheets | | Demand certainty | "If only 1M people visit" | Multi-year locked contracts | | Capacity constraints | None (infinite bandwidth) | Physical limits (fabs 3-5 years) |
AI looks more like oil tanker stocks in the 1970s oil crisis — not "fake," but "cyclical." The windfall comes from supply bottlenecks, and it fades when supply releases. The key question: can you convert windfall profits into moats (technology, customer relationships, scale) during the windfall period?
---
Takeaways for Ordinary People
1. Don't think in "bubble / not bubble" binaries. Three layers, three different conclusions. 2. Memory's "profit bubble" has a 2-3 year window left. Determined by physical cycles, not sentiment. 3. Watch the "depreciation time bomb." Cloud providers (Microsoft, Google, Amazon) haven't yet absorbed AI infrastructure depreciation; the impact will show in 2027-2028. 4. Profit distribution along the chain is shifting — from the "shovel sellers" (chipmakers) to those who "dig gold with the shovels" (the application layer). But the shift takes time.
---