The July semiconductor selloff was built around one uncomfortable question: What happens when Big Tech finally slows its artificial-intelligence spending?
So far, the opposite is happening.
Bank of America now expects hyperscaler capital expenditures to top $1.2 trillion over the next 12 months, according to semiconductor analyst Vivek Arya.
That is a striking escalation from where the AI investment cycle stood only months ago.
“Hyperscale appetite to investing remains strong,” Arya said on Monday.
Earlier this summer, the five largest U.S. hyperscalers were projected to spend roughly $700 billion during calendar 2026.
The implication for semiconductor investors is simple. The AI infrastructure cycle isn’t merely lasting longer than expected. Its dollar intensity keeps increasing.
And BofA sees nine Buy-rated semiconductor stocks offering at least 30% upside to its 12-month price objectives.
Big Tech’s Capex Problem Is The Chip Industry’s Opportunity
There is a strange redistribution happening inside the AI economy.
For hyperscalers, the infrastructure race is becoming increasingly expensive. For the companies supplying that infrastructure, those same costs become revenue.
Microsoft’s capital spending rose 63% year over year earlier this year, while free cash flow fell 10%. Hyperscaler capex had already climbed toward 100% of operating cash flow.
That explains why investors have increasingly questioned the economics of the AI buildout.
Yet the suppliers sit on the other side of that equation.
Every additional dollar spent on AI factories requires GPUs, custom accelerators, memory, networking equipment and semiconductor manufacturing capacity.
BofA’s latest capex estimate suggests that spending pool is still expanding.
Other Wall Street estimates point in the same direction. Morgan Stanley recently projected hyperscaler capex at roughly $800 billion in 2026 and $1.2 trillion in 2027, up from around $450 billion estimates a year earlier.
The bottleneck is increasingly becoming supply rather than demand.
That distinction matters.
If hyperscaler spending keeps rising, the semiconductor correction increasingly looks less like the end of the AI cycle and more like a valuation reset inside an expanding investment boom.
9 Semiconductor Stocks With At Least 30% Upside
Arya groups the beneficiaries into five buckets: compute, memory, semicaps, power semis and optics.
| Stock | Aug. 2 Price | BofA Price Objective | Implied Upside |
|---|---|---|---|
| Marvell Technology Inc. (NASDAQ:MRVL) | $187.56 | $365 | 95% |
| Micron Technology Inc. (NASDAQ:MU) | $823.03 | $1,550 | 88% |
| Intel Corp. (NASDAQ:INTC) | $90.20 | $160 | 77% |
| NVIDIA Corp. (NASDAQ:NVDA) | $200.75 | $350 | 74% |
| Credo Technology Group Holding Ltd. (NASDAQ:CRDO) | $206.99 | $340 | 64% |
| KLA Corp. (NASDAQ:KLAC) | $182.82 | $260 | 42% |
| Applied Materials Inc. (NASDAQ:AMAT) | $507.67 | $720 | 42% |
| Broadcom Inc. (NASDAQ:AVGO) | $389.28 | $530 | 36% |
| Advanced Micro Devices Inc. (NASDAQ:AMD) | $476.15 | $620 | 30% |
Marvell tops the screen with 95% potential upside, followed by Micron at 88%.
But the list stretches across almost every critical layer of the AI infrastructure stack.
Nvidia and AMD provide accelerated compute. Broadcom and Marvell sit at the intersection of custom silicon and networking. Credo addresses connectivity.
Micron supplies the memory needed to keep accelerators fed with data.
KLA and Applied Materials sit even further upstream, selling the equipment needed to manufacture increasingly complex chips.
That breadth is important. The investment case is no longer dependent on one chipmaker capturing AI spending.
It depends on the physical complexity of building AI infrastructure itself.
Scarcity Is Becoming The Investment Thesis
The clearest example may be memory.
Micron shares have already experienced a spectacular re-rating as AI demand collided with constrained supply. The memory cycle became one of the strongest trades in semiconductors this year.
Connectivity has followed a similar path.
Credo controls roughly 75% of the active electrical cable market, according to earlier BofA research. That market could exceed $7 billion by 2030 from about $1.2 billion this year.
The same mechanism keeps appearing across the AI supply chain: hyperscalers want more compute faster than suppliers can comfortably provide it.
And scarcity can spread.
More accelerators require more memory. More servers require faster networking. More complex chips require additional wafer fabrication and testing equipment.
Even higher computing density creates second-order spending on power and cooling infrastructure. The transition toward liquid cooling, for example, is already creating larger order books across the data-center thermal supply chain.
The $1.2 trillion number therefore matters less as a forecast than as evidence of where the AI cycle currently stands.
The market spent much of the recent semiconductor correction asking whether Big Tech had already spent too much.
Bank of America’s latest estimates suggest investors may soon have to confront the opposite problem: what happens if hyperscalers still cannot build fast enough?
