All the businesses in the AI industry chain—from TSMC to NVIDIA to HBM to nuclear power—are ultimately driven by one thing: hyperscaler capital expenditure.
How much Microsoft, Google, Amazon, and Meta pour into AI infrastructure every year determines the revenue ceiling for every downstream pick-and-shovel company. So the most direct, least likely-to-lie signal for where we are in the AI cycle is the Capex of these four companies.
First, look at the scale for 2026.
The four hyperscalers are guiding to total capital expenditures of about $725 billion, up roughly 77% from $410 billion in 2025.
Looking at them separately: Alphabet’s Q1 quarterly capex was $35.7 billion, up 107% year over year; both Microsoft and Meta have significantly raised full-year guidance; and Amazon has also continued to lift its 2026 capital expenditure target.
Historically, no industry has ever seen a capital expenditure expansion this fast or this large.
When looking at these numbers, there are four signals to watch.
First, the direction of Capex guidance, not the absolute number.
In every earnings call, the CFO gives full-year Capex expectations. What really matters is whether that expectation is being “raised” or “cut.” Over the past five quarters, the Capex guidance of these four companies has been raised in nearly every quarter, which is the strongest bullish signal.
Conversely, if any one of them starts to cut guidance—especially leaders like Microsoft or Google—that is a signal that needs to be taken seriously. Because hyperscaler capital expenditure decisions are the leading indicator for orders at all the pick-and-shovel companies, and they show up in downstream revenue six to twelve months in advance.
Second, the composition of Capex, not just the total.
AI Capex is not one thing; it is a bundle of projects—data center construction, power hookup, server purchases, GPU purchases, network equipment, software licenses. In some quarters, construction is the main item; in others, GPU procurement is the main item.
If a company starts shifting Capex from “new data center builds” toward “filling existing campuses,” that means expansion is moving into a slowdown phase; if everything is still being poured into new land, then we are still in an acceleration phase. This detail often comes up in the Q&A section of earnings calls, and it is worth listening for.
Third, the customer mix in NVIDIA’s data center business.
In NVIDIA’s Q2 2026 earnings report, the CFO clearly said that roughly half of the data center business comes from hyperscalers, and the other half comes from other customers (neo-cloud, sovereign AI, enterprises). Changes in this ratio are crucial.
If that half coming from hyperscalers starts to slow, while other customers can pick up the slack, the main trend is still intact; if both sides slow together, then that is the real top.
Fourth, the ratio of Capex to free cash flow.
In the past, hyperscalers were cash cows, and capex was only a portion of cash flow that could be built out gradually. Today, AI Capex at these four companies is already eating up a large chunk of free cash flow. Microsoft has already started financing through debt issuance, and Amazon has been squeezing spending in other businesses. That means today’s Capex is “spending ahead of future cash flow,” and there is a limit to it.
If macro liquidity tightens, or if AI monetization fails to keep pace, these companies will slow down immediately. The ratio of capital expenditure to free cash flow is a very good leading indicator.
From an investment perspective, hyperscaler Capex is the master switch for the entire AI industry chain.
You don’t need to track every pick-and-shovel company’s quarterly revenue, but you absolutely must track the quarterly capital expenditures of these four companies. It won’t tell you whether today’s stock price is right, but it will tell you six to twelve months in advance whether next year’s orders for the pick-and-shovel companies will still be there.
The real top of the AI cycle will not first show up in NVIDIA’s stock price; it will first show up in the hyperscalers’ Capex guidance.