1
1TL;DR: The biggest tech companies are spending way more money on AI infrastructure than anyone expected. Investors hoping for a slowdown might have to wait until 2028.
Imagine you’re building a massive, super-powered brain. You need special buildings, endless electricity, and millions of advanced computer chips. That’s exactly what hyperscalers (giant cloud companies like Google, Meta, and Microsoft) are doing right now for AI.
The problem? The price tag keeps going up, and the companies are burning through their cash to pay for it.
New research from Goldman Sachs strategist Ben Snider reveals just how much estimates have jumped since earnings season started:
| Metric | Old Estimate | New Estimate | Change |
|---|---|---|---|
| Total Spending | $929 billion | Over $1 trillion | +$100B+ |
| Annual Growth | 23% | 33% | Massive acceleration |
Important Callout: For the first time, capital expenditures (capex) will exceed cash flow from operations from 2026 through 2028.
Translation: They’re spending more than they’re making from their core business. They’ll need to borrow or use savings to fund the difference.
| What Investors Want | What Companies Are Doing |
|---|---|
| Free cash flow (money left after spending) | Spending every dollar (and then some) on AI |
| Predictability | "Dynamic" plans, no 2027 numbers |
| Profits now | Building for profits later (maybe 2028+) |
Right now, Wall Street only cares about one thing: How much cash does the company keep after buying all those chips?
A hyperscaler is a massive cloud computing company that operates huge data centers worldwide. Think: Amazon (AWS), Microsoft (Azure), Google (Cloud), Meta. They "scale hyper-fast" to handle global internet traffic and now AI workloads.
Capex = Money spent on long-term physical assets — buildings, servers, chips, networking gear, power systems. It’s not day-to-day expenses (like salaries); it’s investing in the "factory" for future profits.
It means the company can’t fund its growth from profits alone. It must:
CFO Susan Li said planning is "highly dynamic" — meaning demand, chip supply, and technology are changing too fast to predict. It’s honest but makes investors nervous.
Goldman Sachs suggests 2028 — when today’s massive build-out finishes and AI revenue (hopefully) ramps up enough to cover costs. Until then, expect volatile stocks.
Follow the money. The AI race isn’t about who has the best model — it’s about who can afford to build the biggest computer.