Hyperscaler capex and the AI bubble: how much is being spent

What is hyperscaler capex and why does it matter for the AI bubble?

A hyperscaler is a cloud provider that runs data centres at very large scale. Capex means capital expenditure: money spent on assets that last for years, such as buildings, servers, and networks. Hyperscaler capex for AI is the part of those budgets aimed at AI computing. It matters because the whole AI supply chain sells into these budgets, so a change in them moves many other companies.

For the AI bubble question, capex is the clearest measure of conviction. Companies commit cash to AI before customers prove demand, and that bet sets the size of the revenue they must later earn. This guide gives you a method to read the bet. It leaves out specific budget figures, which change every quarter and which you should take from each company’s own filings.

What does hyperscaler capex for AI actually buy?

AI capex buys 5 broad things. Land and buildings house the equipment. Servers and accelerator chips do the computing. Networking links thousands of chips together. Power and cooling keep the site running. Leases give access to space built by others. Each item has a different life, a few years for servers and decades for buildings, which changes how fast its cost reaches profit.

Ask what share of a budget goes to short-life items. A budget heavy in chips and servers ages faster than one heavy in buildings, so the same dollar amount creates a larger yearly cost. Company filings split capex by category, and the split tells you how quickly the spending must earn its return.

How do you read hyperscaler capex against revenue and cash flow?

Two ratios put capex in proportion. Capex intensity divides capex by revenue. The cash-flow share divides capex by operating cash flow. In an illustration with revenue of 100 units, capex of 25 units, and operating cash flow of 40 units, intensity is 25% and the cash-flow share is 62.5%. The same company has little room left for buybacks, dividends, or a downturn.

Track both ratios each quarter. A rising intensity with rising cash flow shows a company that funds growth from its own business. A cash-flow share that climbs toward 100% shows a company that needs debt or new equity to continue. The second case is the one to watch, because spending that depends on outside money ends when that money stops.

How does hyperscaler capex reach profit with a delay?

Capex reaches the profit statement slowly, through depreciation. If a company spends 100 units on equipment that lasts 6 years, the yearly cost is about 17 units. Profit looks healthy in the first year because only a sixth of the bill appears. As spending continues, yearly depreciation climbs, and profit margins must absorb a rising cost.

That lag is why a bubble can build quietly. Strong profits during the build-out hide the cost still to come, and the cost shows up when growth in spending slows while depreciation keeps rising. Read the depreciation line in each filing next to capex. A depreciation line that grows slower than capex signals that more cost is queued behind current profit.

How reliable are hyperscaler capex plans as an AI bubble signal?

Companies give capex plans in advance, and they revise them during the year. Revisions carry the information. Raised plans show that management still sees demand. Cut plans show the opposite. Compare each plan with the actual spending reported later to learn how reliable a company’s guidance has been, and keep a record for each company you follow.

Include commitments outside capex. Leases, purchase agreements for chips, and power contracts can bind a company before spending shows up in the cash flow statement. These commitments appear in the notes to the accounts. When commitments grow faster than capex, the company has committed to more than the current budget suggests.

Which suppliers depend on hyperscaler capex in the AI chain?

Hyperscaler capex is revenue for others. Chip designers, memory makers, network equipment suppliers, construction firms, and power providers all depend on it. A cut at one cloud provider reaches those suppliers within a few quarters, usually through orders first and revenue second. Mapping who sells to whom shows which companies hold the most exposure to a single customer’s budget.

In Singapore, ASEAN, and Hong Kong, the chain includes data centre builders, power and cooling suppliers, and chip supply networks. The EQOS Asia AI study records about $200 billion a year invested in AI worldwide (April 2026), and its φ(S) score of 0.337 describes a system that works but that a shock can tip. Concentrated customers raise that fragility.

What should a decision maker watch in hyperscaler capex each quarter?

Build a watch sheet with 6 columns per cloud provider: capex plan, capex actual, intensity, cash-flow share, depreciation growth, and commitments. Update it after each earnings report, mark any column that worsens, and review the sheet once a quarter. This page gives a reading method and takes no view on any single company.

If your business sells into AI data centre budgets, or buys AI capacity from a hyperscaler, a 30-minute pre-diagnostic with EQOS Asia scores your exposure with the same method as the EQOS Asia method page. You bring the customer concentration and the contract terms, and the session returns a score and a ranked set of options.

Frequently asked questions

About the author

Stéphane Jambu

President, EQOS Forecast Asia

Stéphane Jambu leads EQOS Forecast Asia and works as a consultant in e-commerce, search visibility, and AI. LinkedIn profile