Los Mag 7 perdieron 2,3 billones de dólares. Las acciones de chips se duplicaron.

Key Takeaways

• The split screen: The Magnificent Seven lost roughly $2.3 trillion in market value in June 2026, while the PHLX Semiconductor Index (SOX) rose 11% in the same month and finished the first half up 101%, the strongest first half in the index’s recorded history.
• Same dollars, two prices: Hyperscaler capital expenditure is chip-supplier revenue. Wall Street is currently pricing that spending as a mistake on the buyers’ income statements and as permanent growth on the suppliers’ income statements. Both cannot be right.
• The earnings are real: Micron’s fiscal third quarter produced $41.46 billion of revenue, up more than fourfold from $9.30 billion a year earlier, and $24.67 of earnings per share against $1.68. This is not 1999 vaporware. The question is durability, not existence.
• The 1995-or-2000 question: The only other first halves where the SOX gained more than 50% were 1995, three years before the cycle peaked, and 2000, three months after the dot-com top. Which one 2026 resembles is the most expensive open question in markets.

One Month, Two Verdicts

In June 2026, investors delivered two verdicts on the artificial intelligence (AI) buildout, and the verdicts contradict each other.

The first verdict hit the buyers. The Magnificent Seven, the seven megacap technology companies that dominate the S&P 500, shed roughly $2.3 trillion of combined market value during the month as investors grew skittish about whether their enormous AI-related capital spending can ever generate a return. Microsoft fell 17.2% in June by Yahoo Finance closing data, a decline Euronews called its worst month since December 2000. Amazon dropped 11.9%, Meta 10.9%, Alphabet 6.0%, Apple 7.3%, and Nvidia 5.2%, per Yahoo Finance closing data.

The second verdict rewarded the sellers. The PHLX Semiconductor Index (SOX), the benchmark basket of chipmakers and chip-equipment firms, rose 11.0% in that same June, from 12,829.38 on May 29 to 14,246.96 on June 30. For the first half of 2026, the index gained 101.1%, doubling from its December 31 close of 7,083.13. In daily closing data going back to 1994, the SOX has never had a stronger first half. Not in 1995, when it gained 74.2% in the first six months. Not in 2000, at the peak of the dot-com bubble, when it gained 61.9%.

Read those two paragraphs again. They describe the same money. The capital expenditure (capex) that investors are punishing Microsoft, Meta, and Amazon for spending is the exact revenue that investors are rewarding Micron, TSMC, and ASML for collecting. Wall Street has put two different prices on the same dollar, depending on which side of the invoice it looks at.

The Same Dollar, Priced Twice

The mechanism is not complicated. Five hyperscalers now spend more than $700 billion a year on AI infrastructure, according to Euronews’s tally of analyst estimates. That money buys graphics processors, high-bandwidth memory, advanced packaging, networking silicon, and the lithography machines that make all of it. Every dollar of it lands on a chip supplier’s revenue line.

What changed in June is how much of the buyers’ cash that spending consumes. Capital spending at the big platforms is on track to rise from roughly 70% of their operating cash flow in 2025 to roughly 100% in 2026, per Bank of America estimates cited by Euronews. Operating cash flow (OCF) is the cash a business generates from actually running; free cash flow (FCF) is what remains after capex. The arithmetic is unforgiving:

When capex approaches all of operating cash flow, free cash flow approaches zero. The Magnificent Seven earned their valuation premium over two decades as the greatest free-cash-flow machines in market history. A megacap that converts every operating dollar into servers is, from a cash perspective, a utility with better marketing. June was the month shareholders noticed. The Roundhill Magnificent Seven exchange-traded fund (ETF), a fund tracking exactly these seven stocks, fell about 13% from its late-May high and bled more than $700 million in monthly outflows, its worst month since launching in 2023. Michael Burry, the investor made famous by shorting the 2008 housing bubble, announced new bearish bets against AI-linked assets and called this “the beginning of the end.”

And yet the same skittish market spent June bidding up the companies on the receiving end of those capex checks. Micron rose 18.9% in June, TSMC 14.1%, and ASML’s US-listed shares 23.4%, per Yahoo Finance data. For the half, Micron gained 304%, ASML 86%, and TSMC 57%.

There is a name for this trade. In a gold rush, you sell shovels. The market has decided the shovel sellers are safe: they get paid immediately, in cash, regardless of whether the prospectors ever strike gold. The problem with the shovel-seller trade is embedded in its own logic: it works only as long as the prospectors keep buying shovels.

The Earnings Are Real. That Is Not the Question.

Anyone calling this a rerun of 1999 vapor has not read Micron’s income statement. The memory maker’s fiscal third quarter, which ended May 28, 2026, produced $41.46 billion of revenue, against $9.30 billion in the same quarter a year earlier. Net income came in at $28.24 billion, or $24.67 per diluted share, versus $1.68 a year prior, a nearly fifteenfold jump. Operating cash flow of $25.39 billion more than doubled from $11.90 billion the previous quarter. “Micron’s record fiscal Q3 financial results and even stronger outlook for Q4 reflect the strategic value of memory in the AI era,” said Sanjay Mehrotra, Micron’s chairman, president, and chief executive.

The driver is a genuine physical shortage. Dynamic random-access memory (DRAM), the working memory every AI server needs in enormous quantities, saw prices rise as much as 98% in the first quarter alone, per Euronews. The site has covered both halves of that squeeze: the AI boom cannibalizing smartphone memory supply and the way two memory chipmakers came to represent half of South Korea’s entire stock market before a single-day 10% crash exposed the concentration.

So the bull case for the suppliers is not fantasy. It is audited, taxed, dividend-paying cash. In dot-com terms, this looks less like Pets.com and more like 1995, when the SOX’s 74.2% first half preceded three more years of genuine, earnings-backed expansion.

But notice what Micron’s blowout quarter is made of. Its revenue is the hyperscalers’ capex. The fifteenfold earnings jump is funded by the same spending that Bank of America’s math says is about to consume 100% of the buyers’ operating cash flow. Micron’s income statement and Microsoft’s cash-flow statement are the two ends of one wire transfer.

The Data

The first-half scoreboard, from Yahoo Finance daily closing data:

The June column is almost embarrassingly clean: every pure supplier rose, and every buyer fell. Nvidia is the one company that is both, a Magnificent Seven member whose revenue is other members’ capex, and in June the market traded it like a buyer, not a supplier. Even the index’s own internals tell the story. The S&P 500’s 493 non-Magnificent-Seven stocks were up 13.7% for the year through late June while the Magnificent Seven were down 6.6%, per Euronews.

1995 or 2000?

The SOX has posted a first-half gain above 50% exactly three times in its recorded history: 1995, 2000, and 2026. The first two are a matched pair of opposite lessons.

The 1995 surge was early-cycle. The personal computer and internet buildouts were real, demand kept growing for years, and buying the dip in semiconductors was the correct call for the rest of the decade.

The 2000 surge was the top. The buildout was also real (the fiber got laid, the routers got installed), but the customers doing the building were burning cash they did not have. When the telecom carriers hit their funding wall in 2001, equipment revenue did not slow down gently. It fell off a cliff, taking down the era’s shovel sellers. Lucent Technologies, the equipment giant spun out of AT&T, had juiced its boom-era sales by lending customers the money to buy its own gear; Fortune later described how the loan money showed up on Lucent’s income statement as new revenue while the shaky debt sat on its balance sheet as an ostensibly solid asset. When those practices came to light and the telecom customers hit their funding wall, Lucent’s revenue and stock collapsed, and the company never recovered as an independent firm.

The vendor-financing echo is already here. Anthropic is acquiring roughly $36 billion of Google’s custom tensor processing units (TPUs), the AI chips Broadcom co-develops with Google, leased through a special-purpose vehicle and financed by a record private credit deal syndicated by Apollo and Blackstone. Broadcom is providing a credit backstop on the deal’s senior debt, which PE Insights’ breakdown puts at roughly $6 billion of A1 notes and $25 billion of A2 notes, about $31 billion combined. The site’s breakdown of that $36 billion financing structure called it what it is: the company that builds the chip guaranteeing that the company buying the chip can pay for the chip. That is not proof of a top. Lucent’s financing looked prudent right up until 2001. It is, however, the exact instrument that turns a customer slowdown into a supplier balance-sheet event.

And there is already one documented case of a buyer pulling back in physical terms, and it predates the June selloff by more than a year. In an April 2025 analysis, SemiAnalysis reported that Microsoft had frozen 1.5 gigawatts of near-term self-build data center projects that had been scheduled to come online in 2025 and 2026, and had walked away from more than 2 gigawatts of non-binding lease letters of intent. The episode is covered in the site’s April capex analysis. No hyperscaler has yet cut its dollar capex guidance. But gigawatts are the leading indicator; dollars are the lagging one.

The Cracks in the Shovel Trade

If the supplier rally were priced for any capex disappointment, the last week of June would have looked different.

Micron reported its record quarter after the close on June 24. The stock touched $1,255 intraday the next session and then gave it all back, closing July 1 at $1,032.28, down 10.6% on that day alone. The SOX itself peaked on June 22 at 14,634.72 and closed July 1 at 13,353.28, roughly 9% below the top. The Wall Street Journal noted that even Micron’s blowout results failed to arrest a five-session decline in which the S&P 500 and Nasdaq fell every single day.

When a company beats expectations by that margin and its stock falls double digits within a week, the market is telling you the price already assumed perfection. A 101% first half does not leave room for “merely excellent.”

The steelman for the suppliers deserves a fair hearing, though. Three arguments carry real weight:

1. Backlogs are sticky. Memory is sold on contracts with prepayments, and foundry capacity is booked quarters or years ahead. Even a genuine hyperscaler retrenchment would take several quarters to reach supplier income statements.
2. The cash is banked. Unlike 2000-era dot-coms, the chipmakers of this cycle are collecting historic margins in real time. Micron generated more operating cash flow in one quarter ($25.39 billion) than it did in entire good years of previous cycles. That cash cushions any downturn.
3. The spenders can afford it. The Magnificent Seven are not 2000’s debt-fueled telecom carriers. They fund most of their capex from the largest operating cash flows in corporate history, although CNBC notes that some of the AI investment is now being fueled by debt. Capex at 100% of OCF is a choice they can still reverse, not a funding wall they have already hit.

All three points are true. All three were also true, in period-appropriate form, of the equipment makers in mid-2000. Sticky backlogs delay the reckoning; they do not repeal it. The third point is actually the bears’ point wearing a bull costume: spending that is a choice is precisely the kind that gets cut when shareholders revolt, and June’s $2.3 trillion was the revolt.

What Happens Next

The collision has a date. The hyperscalers report second-quarter earnings in late July, and every call will feature some version of the same question: are you slowing capex? Each company faces a prisoner’s dilemma. The one that cuts early risks conceding the AI race; the one that cuts late risks becoming the shareholder-value cautionary tale. Watch the language, not the numbers: phrases like “optimizing the pace of deployment” or “capital efficiency” would be the dollar-guidance version of Microsoft’s quiet gigawatt freeze.

For the suppliers, the tell will be memory contract pricing and any change in prepayment behavior. DRAM prices that rose that steeply in a single quarter can reverse just as violently once buyers stop panic-ordering, a dynamic the site examined in the DRAM concentration crash.

For the index math, the stakes are systemic. The S&P 500 spent 2025 being carried by the Magnificent Seven and is spending 2026 being carried by their suppliers. If the capex tap tightens, both engines sputter together: they were always the same engine viewed from opposite ends.

What This Means for You

If you own an S&P 500 index fund, you own both sides of this contradiction. Your Microsoft shares are being punished for the same dollars your Micron shares are being rewarded for collecting. The index cannot win this argument with itself; it can only discover, one earnings season at a time, which side was wrong.

If the answer is 1995, the suppliers grind higher for years and the Magnificent Seven eventually get credit for infrastructure that pays off. If the answer is 2000, supplier earnings are at a cyclical peak funded by capex their own shareholders are demanding be cut, and the half-year the SOX just booked will be remembered the way its 2000 run-up is remembered.

There is no cheat code for telling which it is in advance. But one heuristic has aged well through every capex cycle since the railroads: when the companies writing the checks start getting punished for writing them, the companies cashing the checks are living on borrowed time, no matter how real this quarter’s earnings are.

Frequently Asked Questions

Why did the Magnificent 7 stocks fall in June 2026?

Investors grew skeptical that the more than $700 billion the five largest hyperscalers are set to spend on AI infrastructure this year can generate returns that justify it, especially as hyperscaler capex approaches 100% of operating cash flow. The selloff erased about $2.3 trillion of combined market value during the month. Microsoft fell 17.2% by Yahoo Finance closing data, which Euronews called its worst month since December 2000.

Why are chip stocks rising while Big Tech falls?

Chip suppliers are paid in cash up front for the AI buildout regardless of whether it ultimately earns a return for the buyers. Micron’s fiscal third quarter revenue more than quadrupled year over year to $41.46 billion on surging memory prices. Investors rotated from the companies spending the money to the companies collecting it, pushing the SOX index up 101% in the first half.

Is the 2026 AI chip rally a bubble like 1999?

The comparison is imperfect in both directions. Unlike the dot-com era, the chipmakers of 2026 report record real earnings and cash flow. But the SOX’s only comparable first halves came in 1995, which preceded years of genuine growth, and 2000, which marked the top before a multi-year collapse. The outcome depends on whether hyperscalers sustain their capex, which their own shareholders began punishing in June.

The Bottom Line

June 2026 was the month the AI trade stopped being one trade. The market now believes the spending is a mistake and the revenue from that spending is a growth story, simultaneously, at extreme prices on both legs. The suppliers’ earnings are real, the buyers’ cash flows are real, and the contradiction between them is just as real. One of these two prices is wrong, and the second-quarter earnings calls at the end of July are where the market starts finding out which.