When Goldman Sachs tells you the infrastructure spending will hit $1.2 trillion next year, they're not forecasting a boom—they're telling you the foundation is already being poured.
The Summary
- The five largest US hyperscalers will increase AI infrastructure spending by more than 50% to $1.2 trillion next year, Goldman strategists project
- Goldman's Timothy Moe says AI stocks remain attractive even as bond yields rise, positioning himself in the "stronger-for-longer camp" on AI
- This is cement-mixing, datacenter-building, power-grid-upgrading money—not vaporware hype
The Signal
Goldman Sachs just put a number on what many in the agent economy already feel: the infrastructure build is accelerating, not plateauing. A 50% year-over-year jump to $1.2 trillion from the big five hyperscalers (think Microsoft, Google, Amazon, Meta, and likely Oracle or Tesla in the mix) is the kind of spending that rewrites energy grids and semiconductor roadmaps. This isn't R&D. This is buying land, pouring concrete, and signing 10-year power contracts.
The timing matters because bond yields are rising, which traditionally kills high-multiple tech stocks. But Goldman's Timothy Moe isn't flinching. His "stronger-for-longer" positioning suggests the bank sees revenue justification coming, not just continued spend. That's the shift. We're moving from "will AI pay off?" to "who captures the value?"
"When strategists stop hedging and start using phrases like 'stronger-for-longer,' they're reading order books, not tea leaves."
The hyperscaler spending breakdown tells you where the agent economy is headed:
- Compute infrastructure: more GPUs, TPUs, and custom silicon for training and inference
- Datacenters: physical footprint expansion in places with cheap power and cooling
- Networking: the pipes connecting distributed AI workloads and edge deployments
This capital is creating the substrate for autonomous agents that don't just chat, they transact. Every dollar here makes it cheaper to run an agent that negotiates contracts, manages supply chains, or trades assets. The economics flip when inference costs drop another order of magnitude, which is exactly what this buildout enables.
The real tell is that Goldman isn't saying "if AI works out." They're modeling capex growth as a given and telling clients the stocks still work at higher rates. That means they're seeing revenue acceleration in cloud AI services, enterprise AI adoption, or both. The hyperscalers don't spend $1.2 trillion on hope. They spend it on signed contracts and usage curves that justify the return.
The Implication
If you're building in the agent space, this is your tailwind. Inference costs will keep falling, deployment will get easier, and the infrastructure to run millions of concurrent agents is being built right now. The constraint isn't compute anymore, it's imagination and go-to-market.
For everyone else, watch what gets built on this foundation in 2027. The capex happens in 2026. The agent economy that runs on it goes live 12-18 months later. By the time you read about the "AI breakthrough" in mainstream news, the infrastructure spending will have already told you it was coming.