This is not financial advice. Figures come from company releases dated Oct 15–21, 2026. Anything we calculated ourselves is marked as our math.
"Is AI a bubble?" gets asked every time a big tech company reports. Hot takes don't settle it. What you can do is check, each quarter, whether the spending is being paid for, and whether it's turning into revenue for someone. Every number below is in a standard earnings release.
We'll use Meta's Q3 2026 results (quarter ended Sep 30, reported Oct 21, 2026) as the worked example, because it reported a week ahead of Microsoft and Alphabet (Oct 28) and Amazon (Oct 29). You can run the same five checks on those when they come out; our AI earnings calendar has the call times.
1. Capex as a share of revenue
How to find it: capital expenditures (in the cash flow statement or highlights) ÷ revenue.
Meta, Q3 2026: $34.91B capex, including finance-lease principal, on $64.60B revenue = ~54% (our math). A year earlier it was about $19.37B on $51.24B, or ~38% (our math).
How to read it: a rising share means spending is growing faster than sales. On its own that isn't a red flag, but it raises the bar for future revenue.
2. Capex versus operating cash flow
How to find it: compare "net cash provided by operating activities" with capex. Many companies report the difference as free cash flow.
Meta, Q3 2026: operating cash flow of $32.79B against $34.91B of capex gives free cash flow of −$2.12B, versus +$10.63B a year earlier. For the first nine months, free cash flow is still positive at $11.05B.
How to read it: this is the cleanest single test. While capex fits inside operating cash flow, the business is funding the buildout itself. When it doesn't, the money comes from reserves or borrowing.
3. How fast depreciation is growing
How to find it: "depreciation and amortization" in the cash flow statement.
Meta, Q3 2026: $7.15B, up from $4.96B a year earlier (+44%, our math). Net property and equipment rose to $253.2B from $176.4B at the end of 2025.
How to read it: capex doesn't hit profits right away. It shows up later as depreciation, spread over the assets' useful lives. Here's an illustrative example, not Meta's numbers: $100B of servers depreciated over five years adds about $20B a year of cost. Growing depreciation is the bill for past spending arriving. Revenue has to grow to absorb it.
4. Debt and buybacks
How to find it: the financing section of the cash flow statement.
Meta, first nine months of 2026: - $24.91B of new long-term debt. Long-term debt was $80.92B on Sep 30, 2026. - No share repurchases, compared with $26.25B in the same period of 2025. - Dividends continued. - A new non-voting Class C stock is planned for January 2027, which Meta says is partly for capital raising.
How to read it: pausing buybacks and adding debt shows a company choosing to put cash into infrastructure over returning it to shareholders. That's a strategic bet, not proof of a bubble. Still, these are the numbers to watch if growth slows.
5. Are the suppliers still seeing orders?
The other side of big-tech capex is supplier revenue. As of their mid-October 2026 releases:
- TSMC (Q3, reported Oct 15): revenue of US$46.72B, +41.2% year over year. It guides Q4 to US$52.3–54.3B.
- Lam Research (quarter ended Sep 27, reported Oct 21): record revenue of $8.22B. Its CEO said "AI driving higher semiconductor demand".
- Texas Instruments (Q3, reported Oct 21): revenue +28% year over year, with "continued broad growth in industrial, automotive and data center."
- Memory: TrendForce (Sep 30, 2026) expects DRAM and NAND contract prices to keep rising in 4Q26, driven by AI servers.
How to read it: if supplier guidance turns down while big tech keeps saying it will spend, take note. As of this week, both sides still point up. Details are in chip earnings from TSMC, Lam and TI.
The other side of the ledger: is it earning anything?
Spending is only half of it. The other half is whether AI drives revenue. In Meta's case, ad revenue grew 26% and average price per ad 14% in Q3 2026. Meta credits AI for "substantial incremental advertising revenue". It also says its superintelligence-lab investments "will take longer to generate returns than those in our core business." Both statements are worth tracking against the numbers each quarter.
A one-line checklist
Each quarter, ask: Is capex growing faster than revenue? Is it still inside operating cash flow? Is depreciation outrunning revenue growth? Is debt rising while buybacks stop? Are suppliers still guiding up? The more yeses to the first four, and a no on the fifth, the more stretched things are.
If you build with AI, this matters even if you never trade a stock. These spending decisions drive GPU rental prices and memory costs. For the full Meta numbers, see Meta's Q3 2026 earnings.
