Markets are no longer giving AI companies the benefit of the doubt just because they are spending heavily. What happened this week, and in the weeks before it, should be a wake-up call for anyone still treating this as one uniform trade.
Global AI investment is projected to exceed $2.5 trillion in 2026, yet the disconnect between roughly $400 billion in infrastructure spending and only around $100 billion in enterprise AI revenue has become impossible to ignore.
A recent Bank of America fund manager survey found 45% of respondents now flag an AI bubble as the market’s greatest tail risk, up from just 11% a few months earlier, with more than half saying they believe AI stocks are already trading in bubble territory.
This week made the reason for urgency unmistakable. Asian stocks slipped Thursday as a recent tech-led rally on Wall Street paused, with the down 0.2% and South Korea’s falling 1%.
The pulled back from a record high, and an index of semiconductor stocks lost more than 1%, even as Nvidia () itself advanced. SpaceX () tumbled 14% despite posting strong earnings, ahead of the release of roughly $101 billion of shares becoming available for trading Thursday.
One session captured the whole story. Strong results still triggered a sharp share-price fall, and a broadly steady chip sector still couldn’t prevent a wider pullback. Strong earnings did nothing to stop a 14% single-day fall.
This only happens when investors have already decided that headline growth isn’t enough on its own anymore.
The same split has shown up repeatedly in recent weeks. Nvidia shares fell 5% in a single session after reports it was pursuing a payment guarantee of up to $250 billion for OpenAI’s data centre lease, alongside discussions for up to $350 billion in additional financing, pushing the company’s market cap below Apple’s () for the first time in over a year. Its five-year credit default swap premium surged by the largest single-day amount on record on the news. SK Hynix () posted record quarterly revenue, up 257% year-over-year with a 76% operating margin, and still fell 9% on the earnings call.
None of this means the AI trade is finished. It means investors need to stop treating it as settled and start applying real scrutiny, because waiting for more certainty before adjusting positioning is itself a risk now.
Three things matter most for the rest of the year.
First, differentiate within the sector rather than treating it as one bet. Micron (), Applied Materials (), and Cisco () have each posted genuine earnings strength this year on the back of real component shortages and cloud-provider demand, with Cisco raising its 2026 revenue guidance to $62.8 to $63.0 billion on solid AI data-center orders. Some companies are seeing real, measurable demand for the physical components that power this build-out.
Others are increasingly reliant on complex financing arrangements to sustain their growth narrative. Lumping them together in one portfolio decision is no longer defensible.
Second, watch balance sheets, not just growth stories. SpaceX has erased roughly $1.2 trillion in market value since its record-setting June IPO, sitting 47% below its June 16 closing high, pressured by lock-up expirations, Starship test setbacks, and now a fresh $101 billion share unlock landing squarely on an already battered stock. Meanwhile, Alphabet (), Amazon (), Meta (), and Microsoft’s () collective 2026 capital expenditure is set to jump 77% to a record $725 billion, well above the $500 billion analysts originally expected, against a combined contractual backlog across the group of roughly $2.1 trillion.
The stocks under the most pressure this year aren’t simply the ones spending the most on AI infrastructure. They’re the ones carrying the largest financing entanglements and debt-guarantee exposure.
Third, expect volatility around each earnings date rather than a steady trend. The price-to-earnings ratio has climbed above 40, a level last seen before the dot-com crash, and this week’s 14% swing in SpaceX shares on strong earnings, not weak ones, shows exactly how easily financing and unlock events can overwhelm fundamentals in the short term.
This year has taught investors that AI-adjacent stocks can move 5% to 10%, and sometimes considerably more, in a single session on financing news alone, in either direction. Investors need to size positions accordingly. Sharp single-day moves around individual earnings dates are very different from a gradual repricing of the sector, and the two need to be told apart urgently.
The investors who do well for the rest of 2026 are likely to be the ones who stopped asking whether AI as a sector is a good bet months ago.
The more useful question is which parts of the AI trade are built on real demand and which are built on financing structures that still need to prove themselves. Getting that distinction right, and acting on it sooner rather than later, is the work in front of every investor holding AI exposure today.