Capital spending was raised, yet the stock fell.

Produced by | Miaotou App

Author | Dong Bizheng

Editor | Ding Ping

Header image | AI-generated

Even great results could not save the stock.

Late on July 16, 2026, TSMC delivered an almost flawless report card.

TSMC’s second-quarter net profit rose 77% year on year, its gross margin reached 67.7%, and its third-quarter revenue and gross margin guidance both beat Wall Street expectations.

What Wall Street cared about most was capital spending. After all, capex growth is seen as a barometer for the AI industry cycle.

TSMC said it raised its full-year 2026 capex guidance from $52 billion-$56 billion to $60 billion-$64 billion, adding $8 billion in one move and setting a record high.

By the script of the past three-year AI bull market, July 16 should have been a night of sharp stock gains. The picks-and-shovels player had beaten expectations again, the whole industry cycle had moved up another notch, and bulls should have kept adding to their positions.

But this time, the script did not repeat.

TSMC’s U.S.-listed shares fell 2.32%. Memory stocks fared worse: SK Hynix dropped 13.69%, Micron fell 5.65%, and SanDisk declined 12.63%. NVIDIA, Intel, Tesla and others also fell, with Apple and Microsoft the only Magnificent Seven stocks to rise.

Could capital spending, the market’s faith anchor, be priced in reverse for the first time?

Expectations Anchor Starts to Wobble

To understand the unusual move in U.S. stocks on July 16, it helps to start with the logic underpinning the AI bull market of the past three years.

After ChatGPT emerged in late 2022, capital markets kept revising up their bullish view of AI. First came NVIDIA’s explosive results, then the fundraising mythology around model companies, and then the grand narrative that computing power is the new oil, water, electricity and coal.

At the end of that narrative, capital markets found their most durable faith anchor: the capital spending by tech giants on AI infrastructure.

Capex guidance from NVIDIA, Google, Meta, Microsoft, Amazon and other giants became the unit of measurement for the health of the entire AI supply chain. Higher guidance meant an upgrade to the industry cycle; lower guidance meant a bubble warning.

That logic had been validated again and again.

Every new capex increase tended to send shares of most AI-related companies, including Micron, Intel, NVIDIA and TSMC, higher.

Under this expectation that a rising tide lifts all boats, capex was more than a financial number. It was a narrative symbol: as long as the number kept rising, faith in the AI bull market remained intact; once it began to fall, the story would have to be retold.

So when TSMC sharply raised its 2026 capex from $52 billion-$56 billion to $60 billion-$64 billion on July 16, the market should have treated it as a major positive.

But part of this capex increase is inflated.

TSMC CEO C.C. Wei gave two reasons for the capex increase: first, demand is still rising and customers are applying pressure; second, inflation has emerged in the semiconductor industry, and TSMC cannot avoid higher raw material costs.

Miaotou believes the expected impact of the capex increase will be somewhat discounted because prices are rising across the semiconductor supply chain.

Earlier, on July 1, Meta’s plan to sell excess computing capacity to external customers for revenue prompted capital markets to start worrying about a computing power glut.

Meta later said it planned to invest $10 billion to build its first data center in Canada, its largest data center outside the United States with 1 GW of power capacity. In the first half alone, Meta had already signed for more than 5 GW of data center resources, excluding self-built projects.

After the news broke, many Wall Street firms argued that capex would still grow in 2027 and that concerns over a computing power glut were unnecessary.

Still, capital markets responded only mildly, and many related AI stocks remained weak.

Micron also raised its full-year capex again after its latest earnings report, lifting the figure from $25 billion to $27 billion, with 2027 capex set to be even higher. That did not give investors much additional confidence.

Miaotou therefore believes the investment logic that capex intensity is the valuation anchor for AI investing is starting to loosen.

The Next Anchor for AI Investing

Next, major U.S. tech giants will enter earnings season and disclose their capex levels. Investors can use the capital market response to judge whether the anchor for AI investing has truly changed.

If capex keeps being revised upward while the downtrend in U.S.-listed AI-related stocks does not stop, investors who believe in AI should be careful. This is the last psychological line of defense for the current U.S. AI bull market.

If capital spending is no longer investors’ holy grail, what will be the next anchor for AI investing?

Miaotou believes the next anchor for AI investing will be whether the pace of profit realization matches capital spending.

The earnings season at the end of April already offered a preview.

Google parent Alphabet raised its 2026 capex guidance from $175 billion-$185 billion, set in February 2026, to $180 billion-$190 billion, and its shares jumped 9.96% after hours. Meta also raised guidance, but its shares fell 8.55% after hours.

The difference was not how much they spent, but whether there was revenue attribution.

Google Cloud reported $462 billion in backlog, nearly doubling in a single quarter. Meta’s advertising business, by contrast, was carrying $100 billion in spending on its own, while management gave vague answers on a conference call when pressed about investment returns. Its reasons for raising capex were higher component costs, including GPUs, and rising data center costs.

It is clear that capital markets are no longer applauding the absolute scale of capex. They are starting to applaud how much revenue each dollar of spending can generate.

Goldman Sachs’ statistics from the first-quarter earnings season were even more sobering: among all companies that mentioned the relationship between AI and productivity, only 11% actually quantified productivity gains in specific scenarios, and only 2% quantified AI-driven productivity gains at the profit level. In the previous quarter, those two figures were 10% and 1%, respectively, showing almost no substantive progress.

The capex has been spent, but evidence of productivity gains has barely appeared.

Miaotou believes the fundamental reason this capex faith anchor is starting to fail is not funding, nor orders, but returns.

Next, the AI bull market in U.S. stocks will enter a moment of changing anchors.

The AI industry is still running hot. But capital markets are once again reexamining this AI bull market. The real risk has never been whether capex rises, but whether returns can materialize after capex has risen.

This content may not be reproduced without permission. For authorization, please contact hezuo@huxiu.com.

End

💥 Starting at 09:00 on July 18 (tomorrow), Huxiu’s WAIC livestream will host seven CEO-level roundtables.

👨‍💻 More than 20 AI entrepreneurs will gather on site for seven and a half hours of debate.

📊 From models, agents, embodied intelligence and terminals to organizational efficiency, every session will cut straight to the industry’s real issues.

🤖 Scan the code to reserve the livestream, break down the AI industry’s real ledger together, and seize new opportunities.