Hacker News Reader

Comment by gamegoblin | Hacker News Reader

The one caveat here is that little companies can compete in particular niches in which they can provide a more tailored solution.

For instance, Lambda Labs is "AWS for ML". They have GPU cloud instances with pre-configured ML packages and a network attached filesystem. That's all you need.

The other day, I wanted to run Stable Diffusion, but I don't have a GPU. I went to spin up a GPU instance in AWS, but got hit with a quota limit. By default, AWS accounts have a quota limit of 0 for GPU instances. Presumably to prevent fraudulent crypto mining. You have to file a support ticket that is manually looked at by humans to get this limit raised. In my case, it took 4 days for them to raise it.

I had an A100 instance running in minutes on Lambda Labs. It's significantly cheaper than AWS, too.

I'll be using Lambda Labs for all my future GPU needs, significantly better experience.

View on HN

Niches like this are often ephemeral and poor uses of capital in the long run. Early on they prove a need exists and can be profitably satiated with the right solution. As long as the market stays minuscule the company introducing the solution has a moat to be a lifestyle business. Once the market is sizable enough though a big guy just offers the same service and consumes the market share.

I get your point about small companies occupying a niche as their distinct competitive advantage, but in this case, one has to ask the obvious question: why doesn't Lambda Labs have a problem with fraudulent crypto mining, if the bar for entry is so low and it's so much easier to be up and running in minutes?

And whatever the answer is, one hopes that there is some particular reason why Lambda Labs doesn't think AWS could replicate it... otherwise, it wouldn't be a durable enough moat to defend their business model long-term.

That's quite easy to explain. They don't have a problem because the people that contract them do it explicitly to run GPUs. They don't have to protect a massive number of people that don't exactly know what they are hiring.

That makes sense under a certain set of assumptions I suppose, namely that you might think AWS's problem arises mostly from normal customers not realizing that someone has hijacked their account to run GPU crypto mining without their knowledge, because GPU instances were not their intended use case. So "people that don't know what they are hiring" is the failure mode you're concerned with.

But I wonder more about, say a fraudster using stolen credit cards or gift card abuse, to open an account, or abuse a free trial, and if Lambda Labs makes it super easy and low friction to start up a GPU, then the fraudster can open an account, mine crypto immediately, rack up as much crypto as they can for days or a month until billing fails or trial expires, and just try to squeeze Lambda for as much computing power as they can subject to Lambda’s ability to detect fraud.

Low friction for customers, usually translates directly to higher risk of abuse. AWS adds its limits and friction for a reason, not because they enjoy making life harder (esp. since, easier for customers == more business/revenue).

There's Vercel, there's Netlify, there are more. They all have their niche offerings, which sometimes beat what AWS does. However, they're not a _generic_ cloud hosting provider as Rackspace is. Seems like specialisation is the only way out for them.

Reply on news.ycombinator.com