# AI Overbuild and ROI — retained record

> Retained reddit record from reddit:wallstreetbets.

Canonical: https://highsignal.app/data/records/d68c9659aef8153f

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:wallstreetbets)  published 2026-10-23 02:00 ingested 2026-10-24 02:39

[original source ↗](https://www.reddit.com/r/wallstreetbets/comments/1xd1dxg/ai_overbuild_and_roi/)

## retained text



```
**AI Overbuild and ROI**

This is to answer the question of whether AI a bubble. Two things can happen that will make it a bubble: overbuild and poor ROI.

I know none of you all care about the numbers so I’ll keep it simple and feel free to skip to the summary.



ROI Answer

Total AI Capex since 2022 is 5T. Lets assume it doubles by the time buildout is complete: 10T. Now assume it will be paid off over 20 years. That means AI profit must be 500B annually.

Assume total US population between ages 7-70 pay $20/month = $64.8 Billion annual

Assume top 10% of US companies pay $650/month per employee and every other company pays $50/month = $177 Billion annual

Both figures combined = $243 Billion annual revenue from AI

That equals to around $710 per person ages 7-69 per year.

Apply that to the world with income adjusted (Western Europe, Japan, South Korea, Australia, Canada, South America and the Middle East)

That equals approximately = $565 Billion/ year. 

Assuming a 50% margin, that equals to about half of the necessary 500 Billion/year needed for ROI. Even if all AI spending stopped today, it would only just break even.



Overbuild Answer:

Even if compute demand keeps increasing, it doesn’t mean there won’t be overbuild. I wont bore you with the math, just know I took into account: Current demand growth rate, plateau timing and length, breakthrough demand surge, today's compute shortage, chip capacity growth (packaging, memory, power), compute efficiency gains per chip and 5 other factors

The answer is a compute overbuild hitting 3rd or 4th quarter 2027, likey lasting at least a year until demand outpaces it again.



Summary

ROI is going to be nearly impossible to achieve in aggregate, so it will have to be winner-take-all to justify the spend, or at the very least some sort of uneven distribution with many players losing money. So pick your horse well. Regarding overbuild: AI demand will continue to grow but nothing in life is perfectly linear, and there will likely be a compute overbuild hitting late 2027 and last for 1 to 1.5 years.



Positions

Long GOOGL, AMZN, HOOD. I’m going to ride these till around April 2027 then trim and load up on cash/diversify. (HOOD is a proxy for S and P which will likely dip too with the overbuild).



Have fun calling me a retard for the next 10 minutes before this post is deleted
```



document fetched 2026-10-24 00:35 · canonical https://reddit.com/r/wallstreetbets/comments/1xd1dxg/ai_overbuild_and_roi
