Economies of scale is an advantage in which the cost of each unit falls as an operation grows, because large fixed costs are spread over more output and size buys better terms from suppliers. Smaller rivals pay more for each unit, so the largest company can charge less, earn more, or do both.
StrengthenedAI strengthens it: AI capacity takes capital spending on a scale only a few companies can fund, which widens the gap between the largest operators and everyone else.
How it makes money
Scale is the oldest moat in business. Spread a fixed cost over more units and each unit gets cheaper; buy in bulk and suppliers cut the price. The biggest operator can then undercut everyone, or match their price and keep the difference.
AI computing is a fixed-cost business of an extreme kind. Data centers, chips and power are paid for up front; the cost of answering each request afterward is small. That favors whoever can afford the up-front bill.
The bill
The size of the bill is in the cash flow statements. Over the latest twelve months, Microsoft, Amazon, Alphabet, Meta and Oracle together spent $601 billion on capital, against operating cash flow of $710 billion: 85%. In fiscal 2016 the same companies spent $32 billion, 28% of their operating cash flow. Meta alone expects capital expenditures of about $135 billion to $145 billion in 2026 "to support our AI efforts and core business."
The cost of carrying it
Spending on this scale reaches the income statement as depreciation. Meta's depreciation and amortization rose 48% in the first nine months of 2026 against a year earlier. Microsoft says the same thing in plain words: "The investments we are making in cloud and AI infrastructure and devices will continue to increase our operating costs and may decrease our operating margins." Its Microsoft Cloud gross margin percentage has already decreased, to 66%.
How it fails
Scale fails as a moat when the thing built at scale stops being needed, or gets cheaper to build. A data center bought for demand that does not arrive is a fixed cost with nothing to spread it over. And if the price of computing falls fast, the newcomer who buys later may pay far less for the same capacity.
Our read
AI has turned capital back into a barrier to entry. For most of the software era, the advantage of the largest companies was customers and data, not buildings; a startup could rent the same servers. Now the frontier needs capacity that only a short list of companies can pay for out of their own cash, and they are paying for it while their margins come under pressure.
That is a moat, but an expensive moat to dig. It works if demand grows into the capacity before the equipment ages. For everyone else, the lesson is not to try to match the bill. Rent the capacity, pick a layer the giants do not want, or compete on something scale does not buy.
Questions
- What are economies of scale?
- An advantage in which the cost of each unit falls as an operation grows, because fixed costs are spread over more output and size buys better terms. The largest company can then charge less or earn more than smaller rivals.
- How much are big tech companies spending on AI infrastructure?
- Microsoft, Amazon, Alphabet, Meta and Oracle spent $601 billion on capital over the latest twelve months, 85% of their operating cash flow. Meta alone expects $135 billion to $145 billion of capital expenditures in 2026.
- Does AI spending hurt margins?
- It can. Microsoft says its investments in cloud and AI infrastructure will increase its operating costs and may decrease its operating margins, and Meta's depreciation and amortization rose 48% in the first nine months of 2026.
Sources
- Stratrix calculation, From SEC XBRL company facts, company filings and BEA data (2026-10-25)
- SEC EDGAR, Meta Platforms, Inc., Form 10-Q for the quarter ended September 30, 2026 (2026-10-22)
- Microsoft Corporation (SEC EDGAR), Form 10-K for the fiscal year ended June 30, 2026 (2026-07-29)
Every figure and quotation on this page traces to these filings. “Our read” is Stratrix’s judgment, drawn from them. Analysis and data, not advice.
More from Stratrix
“How long can you carry the capacity before the demand has to arrive?”
See the pack — $499 →The Business Model Wizard
Make the seven choices that define your model, test your numbers against real companies, and meet the rival AI would build against you.
Free · about twelve minutesBuild your Model Thesis →Test a decision, free
Set a live decision against the companies that made it before you.
AcquisitionsThe Distance TestDivestituresThe Divestiture TestDiversificationThe Diversification TestMake or buyThe Make-or-Buy TestStrategic Decisions
The big calls companies keep facing, told through the 819 companies on record that made them.
24 decisionsFind your decision →New analyses, by email
The defining moves in business, checked against the record, as they publish.