News for hackers. Stuff that matters.
Re: by jon128 on Sunday November 15, 2026 @09:21AM (#50507136)
Nvidia wants all ML/AI running on CUDA.

Amazon Web Services, Google Cloud, and those two other clowns LARPing as a cloud, give away cloud credits.

As a big player, you want startups on your "platform", using your tooling, speaking and coding in your vernacular, in your flavors of things and mindsets on ways of doing things.

The first sample is free. Feels great the first time around.

But migrating isn't trivial. Vendor lock-in. A few quarters later, the user is scraping for more. Some startups sought to be acquired to be able to continue to fund their AWS costs because they can't do their own IT.

Check out the licensing on CUDA code as it pertains to silicon, it's gnarly. Stallman would double plus disapprove that one!

How many ML teams do you know, sans the one legit DevOps guy, have a genuinely agile mindset (as programmers define it) to gracefully pivot to non-Nvidia silicon?

"Once a customer, always a customer", loosely something like that.

Did Nvidia manipulate its own market, going far beyond the vendor imperative to give out free samples of something designed holistically for vendor lock-in? Maybe it did. But on the surface, Nvidia is acting out the nature of a vendor. As more details come out, we'll learn how much of existing insidious (though not illegal) behavior and how much of alleged behavior was at play.

Re: by retwakm4 on Sunday November 15, 2026 @10:36AM (#50507706)
Well, Google does and OpenAI does. They both contracted Broadcom to build them TPUs.

People renting GPU servers for their own little pet projects? Probably want Nvidia. But if you are a large scale cloud provider trying to provide SaaS AI models, then you probably are going to want to dump Nvidia at some point and go for ASICs. You can be way more efficient and make more money that way.

The whole Nvidia owning the cloud AI space thing was never going to last.

Re: by jon128 on Tuesday November 17, 2026 @04:06AM (#50529146)
Google is a special place of high talent density, of highly competent people in one place, a partner/investment portfolio of critical talent density startups, and in enough parts of Google itself the right vibe and culture for those people to synergize and shine.

OpenAI is another unicorn in its own right.

_

> own little pet projects? Probably want Nvidia.

Personally I'd reach for non-Nvidia and strive to port CUDA code to run on non-Nvidia. But I have enough of the relevant skills to attempt such a thing.

The inertia of CUDA is vast and heavy.

Most dinosaurs (aka "enterprise") will generally want Nvidia too.