CluenexWhat is capex, and why does it connect AI to stock prices?
Capital expenditure (capex) is spending on long-lived assets such as buildings, servers and equipment. It is recorded on the balance sheet and written off gradually (depreciation) over years, not all at once.
AI runs in physical buildings full of accelerator chips, which need electricity, cooling and fast network links. Every dollar a cloud company commits to that infrastructure leaves its books and becomes revenue for a supplier. That is an accounting fact: capex out, revenue in. It is not a claim about how good AI models are.
For the spender, the cost keeps showing up as depreciation whether or not the new capacity earns a return. For the supplier, the spending is future sales. Capex and free cash flow are closely related, so the free cash flow guide is a useful companion.
How does the AI capex chain work?
The chain has three links: compute, power and networking. Each turns a customer's announced spending into supplier revenue, but on different timelines.
| Link | What it supplies | Typical revenue lag | Main bottleneck |
|---|---|---|---|
| Compute | Accelerators, CPUs, memory, storage, advanced packaging, foundry capacity | Shortest, often quarters | Memory and packaging supply |
| Power | Generation, transmission, transformers, switchgear, backup, cooling | Longest, often years | Grid connection approval |
| Networking | Optical parts, switches, cabling, interconnect chips | Medium, follows build schedule | Cluster size and design |
Compute. Accelerators are the largest single cost, but each one also needs high-bandwidth memory, packaging, foundry wafer capacity and substrates. Whoever holds the scarce part at a given moment tends to capture the profit. Our semiconductor analysis guide covers inventory and lead-time cycles.
Power. Large AI campuses can draw as much electricity as a small city. Utilities and equipment makers benefit, but grid connection queues and regulatory approvals run on multi-year timelines that money cannot easily speed up.
Networking. Training a large model means thousands of chips exchange data constantly, so links between them are a performance limit, not a commodity.
How much are the big cloud companies spending?
Amazon, Microsoft, Alphabet and Meta are the four largest buyers, and their capex guidance is the best single signal for the whole chain. Their plans are stated in earnings calls and filings, and they change from quarter to quarter.
Because figures are revised often, check the latest numbers in each company's most recent quarterly filing (Form 10-Q) and earnings release, not in a headline from months ago. Look at three things:
- The full-year capex range the company guides to, and whether it was raised or cut.
- Capex as a share of sales. A company spending close to half of its revenue on physical assets has bet heavily on a future return.
- Any comment on component price inflation, which means part of a rising budget buys the same capacity at a higher price, not more capacity.
A worked example of the share-of-sales math: a company with $300 billion of annual sales that plans $120 billion of capex is spending 120 / 300 = 40% of sales on assets. The depreciation from that spending will flow through its income statement for years.
Why do supplier stocks move before the orders arrive?
A stock price reflects expected future profit, so it reacts when the forecast changes, not when revenue is booked. A capex guidance raise changes the outlook for every company in the chain on the same day.
This is why the group can jump on a single earnings call and drop on a single canceled project. The price trades the promise, not the shipment. The same idea explains why a good report can still produce a falling price, covered in our guide to expectations and multiple compression.
A practical test is to compare what a supplier has contracted (backlog, remaining performance obligations) with what its customers have only announced. Announcements get revised. Contracted backlog is harder to cancel without penalty. See order backlog explained.
What happens when the chain runs in reverse?
When a project is canceled, the expected revenue for every link disappears from forecasts, even though no order was ever booked. Prices can fall before the reason is understood.
An example is the Virginia Digital Gateway, a very large data center campus proposed in Prince William County, Virginia. Public reporting indicated that the developer abandoned it after long-running litigation over local land-use approvals. The cause was legal and local, not weak demand or a lack of financing.
The lesson is to read the stated cause. A land-use defeat tells you little about total AI demand, while a cut in a big buyer's capex guidance tells you a lot. Our guide on how markets price probability instead of headlines explains why the first reaction is often too broad.
Where does the capex-chain idea fail?
The idea is a useful map, not a prediction. These are the main ways it misleads.
- Announced capex is not committed revenue. Guidance is a forecast that is revised. Projects also stall because of permits, power availability and local opposition.
- A single cancellation does not prove the boom is ending. Announced pipelines have always been larger than what gets built, so counting cancellations against announcements can overstate the shortfall. Treat any one headline tally as an estimate.
- Power is not a safe shortcut. Utilities face slow interconnection queues, regulators who control their rates, and public pushback when load growth raises household bills.
- Heavy capex is not automatically bullish. It shows commitment, which cuts both ways. If the return disappoints, depreciation still lowers reported profit for years. For more on judging whether the spending pays off, see capex today, revenue tomorrow.
- Not every link benefits equally. A scarce part gives its maker pricing power while raising costs for the buyer, so two companies can react to the same news in opposite directions.
What this means for you
If you own several AI-linked stocks, the main risk is hidden overlap. A chip designer, a memory maker, a networking supplier and a utility with data center customers can all rest on one assumption: that the largest cloud companies keep raising spending.
- List what each holding depends on. If the answer is the same for several, count them as one bigger position. The concentration risk guide shows how.
- Read the buyers' guidance each quarter. Note whether capex ranges rose, held or fell, and what share of sales they represent.
- Check each supplier's contracted backlog, not just the headline about a customer's plans.
- Know each company's place on the lag curve. Compute can show results in quarters, power in years.
- Do not react to a single project headline. Find the stated cause first.
Cluenex gives each of about 1,000 large US-listed stocks a free public page with business quality scores and the main risk, which can help you compare the companies in a theme: https://cluenex.com/stocks/