There is one number in the 2026 capex table that doesn't belong with the others. Oracle is spending an estimated 86% of revenue on capital expenditure. Meta is at 54%, Microsoft 47%, Alphabet 46%, Amazon 25%. Oracle isn't at the top of a range — it's in a different category.
Capital intensity, 2026 estimates
| Company | Capex / revenue |
|---|---|
| Oracle | 86% |
| Meta | 54% |
| Microsoft | 47% |
| Alphabet | 46% |
| Amazon | 25% |
What that ratio actually means
A software company converts revenue to cash at very high rates because the marginal cost of another customer is near zero. That is the entire reason software earns the multiple it earns. When 86 cents of every revenue dollar goes into land, power and GPUs, the financial profile stops resembling software and starts resembling a utility or a property developer — heavy assets, long payback, financed partly with debt.
The market can value that. It just doesn't value it at software multiples.
The mechanism
Oracle's bet is that it can win AI training and inference workloads by building capacity faster than it has demand, then filling it with large contracted customers. That's a developer's playbook: build, pre-lease, refinance. It works spectacularly when the contracts are signed and the counterparties are good. It works badly when capacity arrives into a softening market, because the depreciation and interest don't soften with it.
The difference between Oracle and a genuine hyperscaler is the balance sheet it's building from. Alphabet funds its buildout from search cash flow. Oracle is funding a comparable ambition from a much smaller base.
The risk in both directions
If AI compute demand keeps outrunning supply, Oracle's aggression is the correct call and the capital intensity is temporary — it collapses as soon as the assets fill and revenue compounds against a fixed base. That's the bull case and it isn't unreasonable.
The bear case doesn't require AI to fail. It only requires the buildout to be a year or two early, because at this capital intensity a timing miss is financed with debt at rates that are now at the year's highs.
My take
I treat Oracle as the highest-beta expression of the AI infrastructure trade among the large caps — more levered to the outcome than Microsoft or Alphabet, with less cushion if the timing is wrong. That's not a criticism, it's a description of what you're buying. Anyone holding it as a stable enterprise-software compounder should look at that 86% and decide whether that's still the position they intended to own.
What I'd watch: contracted backlog converting to recognized revenue, and the cost of the debt funding the gap.
Bottom line: at 86% capex-to-sales, Oracle is an infrastructure developer wearing a software brand. The upside is real if the capacity fills. The financing risk is real if it fills late.
This is analysis, not investment advice.