The AI Bubble Is Real. These Are The Stocks I Would Avoid First.

Wall Street still argues the AI boom like a cartoon. Either every chip order proves endless demand, or the whole thing is fake. Both takes are lazy.

My take is simpler. Artificial intelligence is a real industrial buildout. Models need power, land, memory, and advanced chips. That does not mean every stock priced as pure AI destiny deserves your capital. The bubble I care about is the one where reported growth, multi-year leases, and friend-group financing get treated like end customers quietly paying full freight forever.

What this bubble actually is

This is not 1999 with empty websites and no sales. The big cloud platforms and chip vendors are booking huge revenue. Nvidia (NVDA) still sits near the center of the hardware story. The big spenders are still pouring hundreds of billions a year into AI factories.

The soft spot is different. Free cash flow at several of those spenders has been crushed by capital budgets. A big slice of the buildout now runs on bonds, special vehicles, and long leases that have not even started. Street tallies put unfinished hyperscaler lease promises near $1.5 trillion, with a huge share still off the balance sheet, plus close to another trillion in future purchase promises for compute, chips, gear, and power.

On top of that sits circular financing we have already covered on this desk. Supplier money leaves as equity or credit and comes back as “demand.” Model usage prices have been sliding even while the cost of building capacity stays high. Real technology can still support a financing bubble when the market prices endless payday on top of stretched plumbing.

Three stocks I would avoid first

I am not calling for the death of artificial intelligence. I am naming the seats where the story breaks first if growth slows, credit tightens, or one big tenant stumbles.

Oracle (ORCL) became the market’s favorite AI landlord. Shares have already been cut hard from highs near $320 into the low $130s, and that drop does not clear the risk. Oracle disclosed about $288 billion of extra data-center lease promises that have not started yet, mostly multi-year deals into the end of the decade. Last week investors learned Oracle sent a force majeure notice on its giant New Mexico campus, Project Jupiter, to protect itself if the site slips past 2028. Oracle says the project remains on schedule. The stock still sold off. That is leverage and delivery risk getting loud in the paperwork.

CoreWeave (CRWV) is the purest public bet on rented Nvidia capacity. The backlog is real and huge, often cited well above $100 billion. So is the cash burn. Capex guides in the mid-thirties of billions against roughly low-teens revenue leave a hole filled with debt. Net debt has run near the value of the shares. Guided interest expense can eat a large share of adjusted operating profit in one quarter. If lenders get picky, or one giant tenant slows take-or-pay deals, the equity is the shock absorber. Backlog is a promise. Debt is a schedule.

Super Micro Computer (SMCI) is the thin-margin AI server builder Wall Street still treats like a chip peer. Gross margins have lived near the low teens or worse. The business is closer to packing advanced parts into racks at speed than owning the scarce silicon roadmap. A record order book can convert, or it can lump and reprice if the big buyers pause. Add the legal cloud from the DOJ case around alleged diversion of AI servers toward China, and you need almost perfect execution for this name to look clean.

What I am not saying

I am not putting Nvidia on this avoid list by default. It still owns scarce designs and real cash customers, even if some demand is financed inside a friend group. I am also not calling Microsoft (MSFT) or Amazon (AMZN) worthless. Cash-flow quality, customer mix, and who funds whom still matter more than a ticker sold as an “AI story.” The bubble risk concentrates where leverage, customer concentration, thin margins, and multi-year promises stack on the same balance sheet.

What would change my mind

Oracle gets cleaner if free cash flow stabilizes and campuses go boring and on time. CoreWeave gets cleaner if interest coverage improves without endless refinancing. Super Micro gets cleaner if margins thicken and controls hold. If AI software sales cover the wear-and-tear bill across the stack and credit stays easy, the avoid list can shrink. Until then, I would rather miss a bounce in the fragile names than pretend every AI ticker is the same trade.

Bottom Line

The AI bubble is not “chips are fake.” It is real factories on stretched cash flow, unfinished leases, circular financing, and growth hopes with little room for a miss. The first stocks I would avoid are Oracle as a leveraged AI landlord with a delivery crack, CoreWeave as a debt-funded rental cloud, and Super Micro as a thin-margin box builder with legal baggage. Own the scarce pieces of the stack if you must own AI. Do not confuse a financing club with a free lunch.