Hacker News Reader

Comment by dzonga | Hacker News Reader

80-90% of AI costs are related to inference. Once zAI served their new model entirely on Chinese chips for free then the nVidia jig is up.

who is going to pay back the money nVidia invests in AI labs, AI datacenter companies if the models are being served dirt cheap.

the Chinese are not the only competitor - Amazon with their Trainium, Google with their TPUs etc.

Nvidia might have a moat on training but on serving it's gonna be a blood bath.

But for now they're capturing 80% of all the AI spend so they gonna keep making money.

View on HN

Pretty much fanfic. The gap between premise and execution remains limiting. Nvidia has adapted to changing markets better than anyone else up to this date which is what got them to the top in the first place. What makes you think "this time it's different?"

The fallacy I see repeatedly is someone spots a legitimate issue that could hurt Nvidia and by the time they have stood up the challenge to pull that off, Nvidia has pivoted and addressed it. Or more recently, they just acquired it like Groq.

For your premise to hold, you're asserting Nvidia is blind to inference. And yet his public positioning suggests otherwise.

https://qz.com/nvidia-gtc-2026-jensen-huang-keynote-takeaway...

What makes you think they've hit some sort of metaphorical iceberg and all they'll do before they sink is rearrange deck chairs?

I don't think anyone expects nvidia to "sink". We're concerned that competitive pressures and possible reductions from sky-high demand could make it really hard for nv to justify or maintain their current valuation.

That is, revenues may be unusually cyclically high.

It's going to be hard for frontier labs to keep spending at this rate without starting to make money. And even if they do, lower cost alternatives will likely impact at least to -some- extent how much of that revenue nv gets.

Have you ever checked out the beta on Nvidia stock? Not to mention its surprisingly sane P/E ratio?

Compare and contrast with Tesla.

I know the P/E isn't completely nuts.

It still prices in revenue growth.

It is possible that nv maintains an absolutely dominant market position and total amount spent on GPUs continues to rise.

It's also possible that either of these things doesn't happen.

Yes, but the first mover advantage really is a thing. As an example, Yahoo really had to mess up to lose its spot to Google, but they were up to the challenge.

That doesn't really have anything to do with what I said.

I'm not saying the bet is against Nvidia the company as in how it executes, or enters new markets.

the bet is against the high margins in inference - when other capable players have entered the market as models become commoditized and the inference serving chips as well. Cerebrus ai etc are already showing custom silicon can make a dent while being served cheaper & faster.

> But for now they're capturing 80% of all the AI spend so they gonna keep making money.

And this is why the stock remains high. This hasn't even started to make a dent in Nvidia's bottom line yet.

These comparisons to Enron are absurd when Nvidia is generating this much cash flow.

Let's see the other alternatives come online and start taking away sales.

There is the Law of Demand. As ai costs plummet, demand increases, just nobody knows the curve shape yet. But could be that cheap inference is how all the AI supply meets the demand

> There is the Law of Demand

What is this?

Demand functions (price as a function of quantity demanded) take all kinds of shapes. Veblen goods are the silly example of wrong-sloped demand [1]. The in-vogue example of sigmoid demand, however, is hot water–make hot water (or lighting, for that matter, as another comment today pointed out for LEDs) cheaper and there is a limit to the things we want hot water for. Halving the cost of hot water doesn't induce much new hot-water demand, it increases demand for other goods and services.

[1] https://en.wikipedia.org/wiki/Veblen_good

I don't think that follows when free plans to get people interested are still a thing. Most people aren't paying anything at all.

Reply on news.ycombinator.com