Wall Street’s best-known skeptics think AI looks frothy. Most of them still aren’t willing to bet against it. According to The Information, the market’s big, bad bears are holding off on large short positions against AI stocks. That’s not because they’ve come around on valuations. It’s because timing a bubble is a great way to lose money while being right.
This matters because short sellers are often the first to call a top. When they sit it out, the market loses one of its main sources of pushback, and the AI trade keeps running on momentum and capex announcements.
🐻 Why the Bears Are Holding Back
The bearish case on AI is familiar by now. Hyperscalers are spending hundreds of billions on data centers, and revenue from AI products hasn’t caught up. Circular deals, where chipmakers invest in customers who then buy their chips, make the demand picture harder to read. Depreciation schedules on GPUs look generous.
But knowing a bubble exists and profiting from it are two very different skills. Several forces keep skeptics on the sidelines:
- Momentum is brutal to fight. Shorting a stock that keeps climbing means losses pile up and margin calls come fast. “The market can stay irrational longer than you can stay solvent” is an old line for a reason.
- Earnings keep showing up. Chip and infrastructure suppliers are posting real revenue and real profits. You can argue the demand won’t last. It’s harder to argue it isn’t there today.
- No clear catalyst. A good short needs a trigger, like a missed quarter, a capex cut, or a credit event. So far, the big spenders keep raising their budgets.
- Crowded pain. Skeptics who went public with bearish bets on AI names over the past couple of years have mostly watched those stocks keep grinding higher.
📊 The Other Side of the Argument
The bulls would say the bears are hesitating for a simple reason: they’re wrong. In their view, AI spending is a platform shift that’s closer to early cloud computing than to the dot-com collapse. They’d point out that companies are adopting agents and coding tools for real, and that inference demand keeps growing as models get cheaper to run.
The honest middle ground is that both things can be true. AI can be a lasting shift and some of today’s prices can still be too high. The dot-com era produced Amazon and Pets.com. The internet was real. Plenty of the stocks weren’t.
🔮 What Could Change in the Next 1-3 Years
If the bears do move in, I’d expect it to happen in stages and not as one big crash call:
- Capex guidance cuts. The first hyperscaler that meaningfully trims its AI infrastructure budget would give short sellers the catalyst they’ve been waiting for.
- Financing stress. More AI data centers are being funded with debt and private credit. If borrowing costs rise or a big project stalls, the weak spots show up fast.
- Margin pressure at the app layer. Model prices keep falling. Companies reselling model access with thin margins are easier targets than the chipmakers.
- Neocloud shakeout. Smaller GPU cloud providers with concentrated customers and heavy leverage are the likeliest early casualties.
What stands out here is that the bears aren’t saying “never.” They’re saying “not yet.” That’s a very different message, and investors should hear it that way.
🧭 What Practitioners and Businesses Should Do
You don’t need a trading desk to take something useful from this.
- Don’t build on one vendor’s pricing. If a correction hits, some AI providers will raise prices, cut free tiers, or disappear. Design your stack so you can swap models.
- Tie AI spend to measurable outcomes. Budgets that can’t show a return will be the first cut when CFOs get nervous.
- Watch your suppliers’ balance sheets. A startup that runs your critical workflows and depends on cheap funding is a risk worth tracking.
- Lock in favorable terms now. Compute and API pricing is competitive today. Multi-year deals signed during a boom can protect you in a downturn.
The smart money isn’t calling the top of the AI cycle. It’s watching for the first crack. Businesses should do the same: keep building, stay flexible, and don’t assume today’s cheap, abundant AI will last forever. The full report is at The Information.