How are ppl modeling out the hyperscalers capex? (MSFT, AMZN, GOOGL, META)

All the noise Im seeing on YT/blogs is ignoring the actual financial models.

How would you model it out? For example, I think that in aggregate the four major hyperscalers (MSFT, GOOGL, AMZN, META) will end up with about $750B CapEx for CY2026. I think only about 2/3rds of that is chips and 1/3rd is buildings. Assume chips depreciate over 5 year and buildings depreciate over 20-30 years. So we can math a bit. Chips Depreciation Contribution = $750B \* 2/3 \* 1/5 = $100B. Buildings Depreciation Contribution = $750B \* 1/3 \* 1/25 = $10B. Total Dep Contribution = $100B + $10B = $110B. So that $750B Capex will have to generate about $110B in EBITDA per year for the next 5 years to break-even. For context, in CY 2026, those 4 companies are projected to have just under $2T in aggregate *revenue* . So it doesn't seem too crazy. Thoughts?

Caveats: That's just for one year. There will also be capex in 2027, 2028, etc. Also, my breakeven implies a 0% ROI, which is obviously bad. This is just as far as Ive gotten so far. Still thinking about how to think about it.

Thoughts or good sources?