frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

Open in hackernews

Story of How Im Running an Unlimited $6/Month AI Provider on 4x RTX 3090s

7•yolo-auto•2d ago
This submission is a tale about how I launched an unlimited LLM provider to about 60 hyped people on the waitlist, then immediately served them a fully dysfunctional death-loop model, and how most people, very reasonably, disappeared, but thanks to a few extremely nice people stuck around anyway, we kept the project alive and its still pretty chaotic but gaining traction.

To back up a little bit-- I believe that the whole point of AI agents is that they should keep working. They should read files, retry, search, code, summarize, run tools, and loop until the job is done. When your employer is paying for it, who cares about cost, but when it comes to my personal money/hobbies, if every loop feels like a tiny financial event, you start babysitting the agent instead of using it, and its not fun.

On the other hand, metered pricing makes me worry about using too much. Usage subscriptions make me feel like I need to use every last magical % or I'm are "wasting it". If only an unlimited provider existed....

Then I joined the AMD developer program - I got some credits to spin up my own MI300x and started tinkering with vllm/sglang inference serving on AMD.

After learning about AMD MI300x , i did some napkin math:

Renting MI300x at 2.00 an hour = ~$1500 a month . It can probably support about 150 users using a small MOE model, like qwen-35b-3a , maybe more.

1500 / 150= $10.00 per month, and we all get to play with agents for a small price.

You can oversubscribe a bit, so i landed on $6 per month, per user, for 2x generation slots, 128k context, no token limits, no rate limits.

I built the site, router, made a waitlist, and then over-optimized the MI300x to the point where vllm bench had like 3k+ output and 40k+ throughput.... But i didn't test the final config/serve commands... And that's where i did a disaster launch. You couldn't prompt the thing without it looping or bugging out, it was cursed. And that's where we lost alot of people.

Luckily, my buddy had a few 3090s, so he threw me a life boat and began hosting qwen for us on 2x 3090s and we finally had an operational model that wasn't costing $2.00 an hour for our whopping 3 users.

We started gaining a more users, so we moved up to 4x 3090s. Which we have plenty of room for more users, but even so, since then:

we've configured vllm wrong like 15 times a GPU died we lost power I made a bunch of one-click starts for openclaw,hermes,pi-mono and none of them really work right and that probably drives people away. Those are still on our site right now.

...but people that know what they are doing seem to really be liking the price point. All in all we have like 98% up time. Its been about a month. We've both learned a ton, even already having backgrounds in SWE/SE/AI , being on the hook for a couple paying users forced us to really focus on delivering them a good product. And now i think we might be close to paying the power/hosting bill so we're not operating at a loss (if u include 3090 capex were still at aloss).

Our break-even point is moving to the cloud to max out a MI300x, which is now tuned and ready to go once we get the users.

And im finding in some areas, subscribing to our service is cheaper than running the model (but as someone who loves local models, i totally get it).

Since then, I've been working on a desktop agent that actually works with small models like qwen -- thats going to replace the broken 1 click starts. It's barebones, but its something out of the box that just works. I made it open source, you can see what im talking about here: https://github.com/yolo-auto-org/yolo-auto-desktop , we're at yolo-auto.com and we have an abysmal free tier to prove it works!

Anyway, hope you got a laugh or found it interesting! Drop a question if you have any.

Comments

b--l•2d ago
`qwen-35b-3a` is really garbage (I know well because it's what I run locally). What quant are you running? Would people really pay $6 a month for it even if unlimited?

That said, nice looking site and wish you the best for getting it off the ground.

yolo-auto•1d ago
So on mi 300x we run FP8 and that was original plan.

We are doing some weird q4 on 3090s now that is surprisingly good (for qwen) .

will people pay for it? Yeah, sometimes. Some dude just burned 80mil tokens in 24 hr and posted it to our discord. I'd say he's getting 600 pennies worth.

I think the best part is that no one is watching what your doing, you don't worry, and it's fun.

Also just as divided as people are to whether it's a good deal or not, people are divided on liking/hating the name and brand.

hyde0395•1d ago
Interesting! i've never just used claude code and coded with local model before. is it smooth with a local model?

Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?

925•cloudking•17h ago•419 comments

I indexed 669 GB of my GoPro videos using my M1 Max computer and local ML models

425•iliashad•1d ago•112 comments

Anthropic pauses credit change for Claude Code

24•fabianlindfors•11h ago•6 comments

Ask HN: What are you working on? (June 2026)

302•david927•1d ago•1077 comments

Ask HN: How are thinking efforts implemented?

101•simianwords•1w ago•30 comments

Newer macOS runs slower on Intel (undeniably) – on purpose or "accident"?

12•srevenant•13h ago•10 comments

Ask HN: How long have you been looking for a job?

10•baddash•8h ago•3 comments

Ask HN: What Jobs (Roles) are in the best position to take advantage of AI?

2•CWhiting•3h ago•1 comments

I'm Eric Ries, author of "The Lean Startup" and new book "Incorruptible" – AMA

801•eries•5d ago•576 comments

Ask HN: Why hasn't there been a real competitor to Ticketmaster yet?

262•mdni007•1w ago•239 comments

Ask HN: Do you remember when you gained consciousness? What was it like?

10•kelseyfrog•8h ago•10 comments

Ask HN: Favorite text heavy blogs that are a joy to read?

115•joshmarinacci•6d ago•30 comments

AWS Bedrock to require sharing data with Anthropic for Mythos and future models

425•TomAnthony•5d ago•254 comments

Ask HN: Why does LLMs love the usage of –?

4•reimertz•13h ago•4 comments

What I have done with Claude Code in the last 60 days being a non tech person

7•sahiltll•1d ago•15 comments

Ask HN: Want to build something open source on nights and weekends together?

39•vira28•5d ago•18 comments

Notes on DeepSeek

208•vinhnx•5d ago•140 comments

Ask HN: Would it be useful to have a slop button in addition to flag?

37•BugsJustFindMe•5d ago•23 comments

Ask HN: What agentic directory structure do you use?

7•dominiek•18h ago•1 comments

Ask HN: Are most corporate SWE jobs performative?

248•hnthrow10282910•5d ago•282 comments

Ask HN: What are tools you have made for yourself since the advent of AI?

442•aryamaan•1w ago•779 comments

Ask HN: How are you adapting technical interviews in this agentic era?

8•jcgr•12h ago•0 comments

Tell HN: Claude is completely unusable for biology

13•Protostome•22h ago•3 comments

Is there a name for the type of comments agents add where they leak the prompt?

18•xdennis•2d ago•8 comments

Ask HN: How do you get into a flow state when using AI to code?

95•kilroy123•4d ago•122 comments

I procrastinate by building tools to stop me from procrastinating: A sad story

20•thisislorenzov•4d ago•10 comments

Ask HN: What does your local LLM setup looks like?

6•the-mitr•1d ago•7 comments

Ask HN: Are you still using a Vision Pro?

171•y1n0•6d ago•215 comments

Ask HN: Year of Linux Desktop is fun with LLMs

5•mirekrusin•1d ago•3 comments

Ask HN: Is anyone shorting the overspend in AI yet?

18•ggm•5d ago•15 comments