# NVDA and TSMC: What if AI demand keeps growing but infrastructure spending slows? — retained record

> Retained reddit record from reddit:ValueInvesting.

Canonical: https://highsignal.app/data/records/9e3aaae73db900a6

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


source: [reddit](https://highsignal.app/data/reddit) (reddit:ValueInvesting)  published 2026-10-21 14:50 ingested 2026-10-22 02:43

[original source ↗](https://www.reddit.com/r/ValueInvesting/comments/1xbp98u/nvda_and_tsmc_what_if_ai_demand_keeps_growing_but/)

## retained text



```
I've been looking at Nvidia and TSMC's earnings forecasts through 2028, and I keep wondering whether we're confusing two different things: the growth of AI and the growth of spending on AI infrastructure.

One thing surprised me. Even if Nvidia misses its 2028 earnings estimate by 20%, it could still deliver around 12.5% annualized share-price returns through 2035, assuming 10% annual earnings growth afterward and a 20x terminal P/E.

Of course, those assumptions are far from guaranteed.

At October 20 prices, Nvidia trades at around 10.5x expected 2028 earnings, while TSMC trades at 12.8x. That sounds fairly cheap, until you look at how much growth analysts already expect.

One comparison I found interesting is the telecom boom.

Internet traffic continued growing rapidly in the early 2000s, even as telecom capital spending collapsed. Demand was real, but companies had built infrastructure faster than it could be justified.

Could something similar happen with AI?

Inference is getting cheaper, models are becoming more efficient, and companies are finding ways to get more work out of existing hardware. Meanwhile, AI agents could create enormous new workloads.

I can see both things happening: a period of disappointing chip orders, followed by several more years of demand growth.

I also compared Nvidia and TSMC. Custom accelerators could weaken Nvidia's pricing power while still benefiting TSMC, which manufactures chips for competing designers.

I wrote up the full research, including the telecom comparison, inference economics and several valuation scenarios for both companies. It's free, with charts and sources:

[https://michaelhillaert1.substack.com/p/could-ai-infrastructure-growth-last?r=6vawcp](https://michaelhillaert1.substack.com/p/could-ai-infrastructure-growth-last?r=6vawcp)

I'm curious what others think. Is the bigger risk that the 2028 earnings forecasts won't materialize, or that investors won't pay anything close to 20x earnings for these businesses in 2035?
```



document fetched 2026-10-22 00:47 · canonical https://reddit.com/r/ValueInvesting/comments/1xbp98u/nvda_and_tsmc_what_if_ai_demand_keeps_growing_but
