Here is the contradiction most people never resolve: in June, the Magnificent Seven torched more than $2.3 trillion in market value — the largest monthly loss ever recorded for the group — and yet the Nasdaq-100 closed the month essentially flat, at -0.85%. Separately, per Crypto Briefing, US tech funds reportedly pulled in $14 billion in a single week, putting 2026 on pace for a record $152 billion in inflows.
A market that is fleeing tech does not do both of those things. It arguably does them when it is re-sorting who deserves the money.
This week that thread got a concrete repricing event and a government paper trail to match: a violent intramural rotation inside the Mag 7, and a draft Treasury report quietly warning that the whole financial system is now leaning on AI to deliver.
Start with the tape, because the tape is more precise than the headlines.
The selloff was not evenly distributed. It was aimed. Microsoft fell 19% in June, its worst month in years, and is down roughly 22% year-to-date. Amazon reportedly shed 8.77%, Meta 6.11%, Alphabet 4.99% in June.
Now look at the supplier side. NVIDIA slipped 10.72% for the month — real pain — but it is still up 4.6% for the year and 29% over twelve months. The buyers got hammered. The seller got a haircut and kept its gains.
That is not a market questioning whether AI is real. That is a market pricing the difference between who writes the checks and who cashes them.
The reason for the split is sitting in the filings. As CNBC's MacKenzie Sigalos put it, hyperscaler capex is up 84% from a year ago and now amounts to 98% of cash flow from operations.
Read that number again. Almost every dollar these companies generate from running their actual businesses is being recycled into chips, memory, and power.
Microsoft booked $30.88 billion of capex in a single quarter, up 84% year-over-year. Alphabet now guides $180–190 billion for the full year and says 2027 will "significantly increase" on top of that. Meta's CFO admitted the company has "continued to underestimate our compute needs even as we have been ramping capacity significantly."
Bulls hear a demand signal. Bears hear a company that cannot budget for its own core input.
The market, in June, sided with the bears — but only against the buyers. Suppliers, memory makers, power operators, and networking vendors are still getting the benefit of the doubt. The hyperscalers are being asked to show their work.
And the work looks circular. Google has committed up to $43 billion to Anthropic; Amazon as much as $50 billion into OpenAI. The labs then spend that money back on cloud compute, and everyone points at the resulting revenue as validation. Money loops from cloud to lab to cloud, and it is not yet clear who ultimately gets paid back.
This is the actual signal of June — not the tired bubble-versus-supercycle binary, but an intramural repricing of payback risk. The market appears to have decided the buildout is real and started auditing who profits from it.
This is the actual signal of June — not the tired bubble-versus-supercycle binary, but an intramural repricing of payback risk. The market started auditing who profits from the buildout.
Which is exactly where the second story of the week comes in, and it is a heavier one.
According to NOTUS, career analysts inside the US Treasury have completed a draft report — awaiting sign-off — arguing that the AI market rhymes with the 2000 dot-com bubble, and that if it bursts, the damage ripples through the entire economy.
The report's transmission chain is specific: an AI downturn would hit the stock market, private credit markets, the firms financing data-center construction, cloud providers, chip manufacturers, and utilities. None of those are insulated from the others.
But the most important finding is not the crash scenario. It is the anatomy of the risk.
Unlike the dot-com bubble, which was driven largely by retail investors, this boom is institutional. Large banks, hedge funds, and private-credit firms are the primary backers of the bet.
That distinction is the whole ballgame. A retail bubble bursting harms household wealth. An institutional bubble bursting threatens the stability of the financial system itself — because the risk has not been diversified, it has been concentrated into a handful of interconnected, too-big-to-fail players.
This is why the June rotation matters beyond the scoreboard. There reportedly appears to be less retail froth to bleed off here than in the dotcom era; the excess is sitting on institutional balance sheets and inside financing structures that touch banks and power grids.
The Treasury paper even lists electricity bottlenecks and utility shortfalls as chokepoints that could stall the whole thing — where power capacity refuses to move at the speed of a funding round.
The report offers itself an escape hatch, and it is a revealing one. If productivity growth expectations are met and commercialization paths prove viable, the bubble might never burst.
Flip that sentence over and you see the trap. The stability of the entire financial system now rests on the premise that AI must deliver on its promises.
That is a bet, not a floor.
Officially, the administration is not looking at the drawer. A Treasury spokesperson called the report "unvetted" and not representative of the department's views, while reaffirming that AI "will be a key driver of America's new golden age." Secretary Scott Bessent, meanwhile, was on a New York stage praising the roughly $750 billion tech giants are pouring into AI this year and warning that the real risk is China, not a bubble.
So which is it — super cycle or bubble? The honest answer, and the useful one, comes from Lightspeed's Bejul Somaia, who refused to pick.
Are we in a super cycle? Yes. Are we in a very rich valuation environment? Yes. Will some investments prove extremely rewarding despite that? Yes. Will some lose money? Yes. "All of those things can be true."
That is not fence-sitting. It is the correct frame. The June wipeout was the market beginning to sort the "extremely rewarding" from the "lose money" — punishing the payers, sparing the suppliers — rather than rendering a verdict on the category.
Which leaves one variable that actually resolves the question, and it is not sentiment or positioning. It is monetization showing up before the circular cloud-lab money loop runs out of patience.
Here is the tell to watch, and it is uncomfortable. Economist Dean Baker notes that productivity growth is running around 1% — aggregate hours grew at a 1.3% rate against GDP near 2% — reportedly following 0.3% in the prior quarter. "There is zero evidence of any sort of productivity uptick in these data."
The late-1990s dot-com era, for all its excess, at least delivered roughly 2.8% productivity growth over four years. Baker's math is blunt: current market levels need something closer to 4% to make sense of the profits being priced in.
We are getting 1%, and it is "invisible" in the data. "It is surprising," Baker writes, "that the continuing weakness of productivity doesn't bother stock investors more."
That gap — 1% delivered against 4% required — is the space the whole trade is living in. Capex is real, the buildout is real, the institutional exposure is real. The payback is a forecast.
June did not answer whether AI is real. It answered a narrower, sharper question: the market has stopped paying up for the dream and started auditing the margins. The buyers are being asked to show their work, the suppliers are getting a grace period, and an outside commentary (36kr) — not the Treasury analysts — drew the parallel that if the productivity numbers don't turn, the drawer the report is sitting in becomes a 2005 mortgage memo.
Watch the productivity line, not the price line. That is where June's question gets settled.