On the day SpaceX prices the largest stock-market debut in history, the market underneath it has a case of nerves. The cause is not rockets. It is AI.
Several warning lights are flashing at once. Together they amount to the first serious test of a trade that has carried global markets for two years.
The clearest signal is in software. Wall Street has spent 2026 living through what traders at Jefferies dubbed the ‘SaaSpocalypse’, a rolling selloff that, by some tallies, has erased as much as $2tn from the S&P software index since its late-2025 peak.
The fear is specific. If AI agents can do the work of a team of sales reps, a company needs far fewer software seats, and the per-seat licensing model that built modern software starts to wobble.
The private-equity giant Apollo has turned that fear into policy. It now screens software deals for AI-displacement risk, holds zero private-equity software exposure, and keeps software below 2 per cent of its assets.
Its reasoning is concentration. Software swelled from about a tenth of global buyout volume to roughly 40 per cent at the peak, a level Apollo calls ‘a fairly significant red flag’.
The nervousness is spreading
It is no longer just software. Hong Kong and mainland Chinese shares fell on Wednesday, with tech stocks among the hardest hit, as AI-bubble fears tracked a retreat on Wall Street.
In Washington, Senator Elizabeth Warren introduced a bill, the AI Bubble Transparency Act, that would force banks to disclose their debt and equity exposure to chipmakers, data centres and hyperscalers. She casts the concentration as a systemic risk.
And KKR’s top macro strategist, Henry McVey, told clients the boom is real but will make the economy ‘more extreme than anything we have seen since the start of the second industrial revolution’ in the 1870s.
Some sectors are ‘starved’, he wrote, while a handful, tech, high-end services and government, run ‘flush’. Defence and power, the firm reckons, are the likeliest long-term winners.
Underneath all of it sits the capex. Hyperscaler infrastructure spending is approaching $660bn this year, the largest corporate investment programme in history outside wartime, and increasingly funded by debt.
Amazon’s borrowing has passed $225bn, and Oracle just overshot its own capex guidance, with tens of billions more to come.
The bear case is simple. Spending on this scale only pays off if AI moves from ‘copilot’ features to autonomous agents that justify the next order of magnitude of compute. If adoption plateaus, the return on $660bn a year falls below the cost of capital.
The bull case is just as real. This is not 2000.
As TNW has noted before, valuations and concentration sit above dot-com peaks on some measures, with the CAPE ratio near 38. But unlike the dot-com darlings, today’s leaders are enormously profitable, and the capex cycle has barely begun to produce results.
The honest answer to ‘is this a bubble?’ is that no one can know until the spending either delivers or it doesn’t. What changed this week is that the market started asking the question out loud, after two years of not wanting to.
SpaceX is not an AI company. But its debut, and the OpenAI and Anthropic listings lining up behind it, will be the closest thing to a real-time referendum on whether investors still believe. Even market-watchers who think the listing will not ‘break’ the bull market, as CNBC put it, are uneasy about what comes after it.
A wobble is not a crash. But for the first time in a while, the people writing the cheques are visibly weighing the question the boom has waved away: what, exactly, does all this return?
