Semis Are Back Because AI Capex Just Hit $760 Billion-But This Rally Has a Knife in It

$760 Billion in Hyperscaler Spending Keeps the Semi Rally Alive

This is the core signal: roughly $760 billion of hyperscaler capex is expected this year, up from $413 billion in 2025. That points to more than one or two strong quarters for semis. It suggests the AI buildout is still creating broad demand across GPUs, networking, memory, and related infrastructure. The spend is back, and the pipeline is large.

Why the scale matters

The forward outlook is what makes this more than a brief rebound. Goldman now sees $5.3 trillion of hyperscaler capex from 2025 to 2030, with a $7.6 trillion baseline projected from 2026 to 2031 across compute, data centers, and power. That framing makes the AI buildout look less like a short-lived trade and more like a multi-year infrastructure program.

Bulls vs. bears

Bulls have the stronger case, but the bear case is real. Amazon, Microsoft, and Alphabet still showed solid operating momentum alongside higher spending, and the market is learning that not all capex gets the same reward. Last week, Alphabet's higher 2026 guidance landed poorly, and Alphabet's stock slid 7% while Amazon, Meta and Microsoft also fell. Investors are no longer giving hyperscalers unlimited credit for spending alone.

That still leaves room for the supplier trade. If hyperscaler budgets hold up, semiconductor vendors are among the first places that spending shows up.

Semiconductor ETFs Are Rebounding Because the Supply Chain Is Still Getting Funded

After the panic, SOXX rose 18 straight sessions. Combined SOXX and SMH April flows also reached $5.5 billion, a record for the semiconductor ETF category. The message is not that every chip stock is automatically back. It is that capital is rotating back toward broad exposure to the AI supplier stack.

Why sector breadth matters

Hyperscaler spending does not stop at one winner. It moves through designers, foundries, memory, equipment, packaging, and test. SOXX captures that broader supplier base, while SMH offers a more concentrated mix of foundry and equipment exposure. Both are benefiting from the same spending cycle: hyperscaler AI capital spending projected near 25% annual growth through 2030 is distributing dollars across the chain, not just inflating one name.

What the selloff actually changed

The noise was unmistakable: more than $1 trillion wiped from market caps of top chip stocks, and the SOX down nearly 20% over the past month. But much of that looked like repricing of financing and timing risk rather than proof that demand disappeared. Fundamentals still held up better than sentiment: the sector was still delivering >100% earnings growth in Q2, while global semiconductor sales rose from roughly $110 billion in April to about $120 billion in May.

That is the more useful anchor. The capex debate may be uncomfortable, but the economic footprint is already showing up in sales and earnings.

The live catalyst: whether capex holds after the backlash

The next test is whether hyperscaler spending survives the earnings-season pushback. Even after the market reaction, the largest US cloud and AI infrastructure providers are still targeting between $660 billion and $690 billion in 2026 capex, with most of that directed at AI compute, data centers, and networking. That gap between infrastructure spending and current application revenue is the tension in the trade. Hardware demand can stay strong before AI software monetization fully catches up.

What could break the thesis

Bears have one important point: this spending is getting harder to fund. One recent analysis noted hyperscalers have guided to roughly $725 billion in 2026 capex and are using debt for more of the buildout, while capex growth is running ahead of revenue growth. That helps explain why the market punished Alphabet's hike and sent shares of Amazon, Meta and Microsoft tumbling even with healthy cloud demand.

What to watch now:

• Bullish: continued ETF inflows, another breadth-led rebound, and no fresh evidence of demand softness.
• Bearish: new capex downgrades, financing strain that slows spending, or a rebound that narrows to just a few mega-caps.

Demand is still feeding the chain, but the market now wants discipline, not just headline capex growth.

Position for Selective Semi Exposure, Not Broad AI Hype

The cleaner trade remains AI buildout suppliers rather than every semiconductor name catching a volume tailwind. Even after record fiscal 2026 revenue of $215.9 billion, the market is still rewarding selective positioning more than blind sector beta. That matters with hyperscaler spending still expected at a massive scale expected to hit $760 billion this year and Amazon and Microsoft are set to report results this week.

How to think about ETF exposure

If the capex wave is real, dollars still have to pass through equipment, foundries, memory, and enablement suppliers before end-market growth fully shows up in units. That argues for owning the bottlenecks rather than chasing every name with an AI label.

What could confirm or invalidate the setup

Confirmation would come from sustained hyperscaler spending, continued ETF inflows, and broad participation across the semiconductor complex rather than a narrow handful of stocks. Invalidation would come from lower guidance, financing strain, or signs that capex growth is slowing before utilization and monetization improve.