// learn ยท Market history
The AI Bubble: hype vs. what holds up
"Is AI a bubble?" is the wrong question. The useful one is narrower: which parts of the AI trade are paid for by real revenue today, which are priced on a future that may or may not arrive, and how do you tell them apart before the market does it for you?
Bubbles and real revolutions look identical early on
The internet was both a genuine revolution and a bubble that wiped out most of the companies riding it. Both things were true at once. Railways, electricity, the dot-com era, every real technology shift has drawn in far more capital than the winners could ever justify, then repriced hard. AI can be transformational and overvalued in the same breath. Treating "it's real" and "it's a bubble" as opposites is the first mistake.
Follow the cash, not the narrative
The cleanest filter is boring: who is actually being paid, and by whom?
- Picks and shovels. Chipmakers and infrastructure providers are booking real, auditable revenue now. The risk isn't "is it real", it's whether current growth rates are already priced in for a decade.
- Platforms. Large software and cloud firms are monetizing AI features into existing paying customers. Grounded, but the incremental revenue is often smaller than the valuation implies.
- Pure story stocks. Companies whose AI revenue is a roadmap, not an income statement. This is where bubble dynamics concentrate, and where the biggest moves, up and down, happen.
The red flags on the table right now
None of these prove a bubble on their own. Together, they're the reason serious people are asking the question (figures as of late 2025, via BrokerChooser and public filings):
- Stretched valuations. NVIDIA has traded around a P/E of ~54, roughly double the market average, and Palantir's multiple has run into the hundreds, hard to square with its growth and margins.
- Index concentration. A single chipmaker has reached ~8% of the S&P 500 and ~14% of the Nasdaq-100. When one name carries that much weight, the whole index rides its story.
- Derivatives boom. The market went from ~100โ200 new ETFs a year pre-2020 to 900+ in 2025, over half using derivatives, more than a third leveraged. Both record highs.
- Margin debt. Investor borrowing against portfolios hit ~$1.2 trillion in late 2025, up ~45% year-on-year, much of it presumed to be in AI names.
- Single-headline swings. Oracle added ~$250B in market value overnight on one AI-infrastructure announcement, then gave it back, a sign the market's conviction is thinner than the price implies.
Signs you're paying for hype
- Valuation justified only by a total-addressable-market slide, not by margins or cash flow.
- "This time is different" used to dismiss profitability entirely.
- Revenue that depends on a handful of customers who are themselves unprofitable.
- Price moving on announcements and partnerships rather than shipped, paid-for products.
If you think it's a bubble: how people short it, and why it's brutal
Deciding the AI trade is overvalued is the easy part. Profiting from that view is one of the hardest trades in markets, because being right and being right on time are two different things. As the old line goes, the market can stay irrational longer than you can stay solvent. These are the common instruments, and the catch built into each.
- Short selling shares. Borrow the stock (the crowded AI names: NVDA, PLTR, MSFT, META, GOOG), sell it, buy it back lower. Downside is theoretically unlimited, you pay borrow fees, and if too many shorts pile in you can get caught in a squeeze. Needs a margin account with short access. Highest risk of the lot.
- Short or inverse ETFs. Betting on the sector instead of one name spreads single-stock risk (one earnings call can move a stock 20โ30%). You can short broad AI ETFs (QQQ, AIQ, ARKK) or simply buy an inverse ETF, no margin, no borrow: PSQ (inverse Nasdaq-100), SARK (inverse ARKK), QQQD (a Magnificent-7 short). The catch: fees are higher (PSQ ~0.95% vs QQQ ~0.20%) and daily rebalancing makes them drift over time, built for short holds, not a months-long bubble call.
- NASDAQ-100 futures (NQ). The Nasdaq-100 is ~50% AI/big-tech by weight, so shorting NQ is a direct sector bet, and going short is as simple as going long. But margin is steep (~$30k for full NQ, ~$3k for Micro NQ) and the leverage is unforgiving, a ~6% index move against a full contract can wipe the margin.
- Put options. Loss capped at the premium, the cleanest defined-risk way to position bearish, with built-in leverage. But time decay is the enemy: you must be right on direction and timing, spreads can be wide (favour high open-interest contracts), or the option expires worthless even if you're eventually right.
- CFDs. Leveraged short exposure with a small account, but you trade against the broker (wider spreads), pay overnight financing, and it's banned or restricted in many places (notably U.S. retail). Roughly 3 in 4 CFD traders lose money over time.
Fuller broker-by-broker breakdown: BrokerChooser, how to short the AI bubble.
How Peaky Radar treats the AI trade
We don't call tops or bottoms. The radar flags momentum, volume and new listings across the AI complex, including the speculative small caps, and labels them for what they are. A signal that a name is moving is not a claim that it should be moving. You get the scan and the context; the decision is yours.
Educational market information, not financial advice. Nothing here is a recommendation to buy or sell any asset. Some links on this site are affiliate links, always disclosed. Markets carry risk of loss, do your own research and consider a licensed advisor.