Of course? On average, companies that are still standing after ~15 to 20 years are going to be a lot larger than the companies who started more recently.
they're a replacement for cargoships. If done right, they're faster than a cargoship, but could (maybe) carry lots of capacity for cheap, while also not requiring them to take the same routes as a ship.
This seems broadly good. If you told me a democratic admin had recruited these people, I would think "wow! what a positive signal for the current admin!"
This only makes sense if you think scaling laws won't hold.
If someone gets something to work with 1k h100s that should have taken 100k h100s, that means the group with the 100k is about to have a much, much better model.
San Francisco Compute | Multiple Roles | San Francisco | ONSITE
We're building a new, regulated, commodity market for large scale GPU clusters. More like the Kalshi, less like craigslist. To accomplish this, we make a VM orchestrator called "Fog" that understands InfiniBand, GPUs, and heterogeneous hardware setups.
We make money by taking a flat fee of a few cents per GPU hour upon delivery. Up until recently, we've been "running a market" via a spreadsheet, by selling bursts on very large clusters (think 6 to 8 figures deal sizes).
SFC started because Alex (https://alexgajewski.org/) & I went to go train an audio model and none of the vendors at the time would sell us a month-long contract. So we bought a year-long one, and tried to sublease it at cost (https://news.ycombinator.com/item?id=36933603), so that way we could buy just one month.
The goal of the company is the same: we want to make it possible to buy a big, giant training cluster for a short time period. We think a liquid market will let you spend $40m for a month instead of $300m for a year. If we can't make that happen, then only the big labs will get to make AI.
Most of our code is in rust, some typescript. We're about ~20 people. We raised $12m a bit ago. There are 7 ex-founders on the team.
We just go down a lot. It's VERY beta at the moment; we literally take the whole thing down about once a week. So if we know of some major problem, or we're down, we just don't let people on (since they'll have a bad experience).
You're right though that the prices are probably lower because of this. That's why we have a thing on our website that says "*Prices are from the sfcompute private beta and don’t represent normal market conditions."
If you'd like on anyway, I can let you on, just email me at evan at sfcompute, but it may literally break!
> Tech giants and beyond are set to spend over $1tn on AI capex in coming years,
with so far little to show for it.
Regardless of whether or not the implicit claim here is true (the claim being "all this spend won't produce an ROI"), the explicit claim here is nonsensical.
Of course the $1tn in capex has nothing to show for it! The spend has not happened yet! Of the spend that _has_ happened, most of the chips are not physically in data centers yet. Of the chips that _are_ in data centers, most of the models are not yet trained!
And of the models that _have_ been trained, many have clearly had a significant ROI. GPT-4 cost $100m, and OpenAI's revenue is now reported to be $3.4 billion a year.
Saying there's "little to show for it" is an absurd claim; the products are _printing_ cash! We beat the turing test! You can drive around in a self-driving car!
It's perfectly reasonable to say "where does the ROI come from when you spend $1tn on capex", but it's hard to argue against the success of the spend of the last generation of models.
> Whether (people) get insurance, or what the rate for their insurance is, or legal decisions or employment decisions, whether you get fired or hired, could be up to an AI algorithm
This is a bit like trying to regulate horseshoes while everyone else is talking about speed limits & seat belts. Both parties say the word "carriage" and "passenger", but they have completely different ideas in their heads about what is about to happen.
I would ignore everyone who's repeating the same sort of standard startup advice or saying "blah blah doesn't xyz already do this". A lot of startup advice is bad / actively harmful, your thing is pretty good, and you should just do it if it's fun.
Because you're on HN, you're about to get a good bump. In a few days, that bump will go down, and you might start to feel sad! Ignore this sadness and keep pushing through. The time from when people first hear about a product and when they actually start using it can be like a few weeks to a month. If, in a few days, all the numbers are going down, know that this is how even successful products look after a product launch. (See the trough of sorrow) Keep going!
Keep shamefully posting it in places, consider a hacker news launch, or even putting up posters in high-foot-traffic areas (if you live in or close to a walkable area). I know social-media bad, but it's a pretty good place to reach out to folks on. My guess is making a little tiktok thing is not a bad idea.
If you've got cash to blow, consider ads. Lots of people will tell you ads are bad, because they "don't scale" or some other over-optimized thing. But if you just want people to see your thing, ads are a reasonable way to drive a little traffic and get a beginning user-base to experiment with.
Have fun! The design and explanation of the site is great!
Nat & Daniel’s cluster is great, and we fully recommend startups seek out this option as well. Nat & Daniel are some of the best investors one can have
Ah, we're running a medium amount of compute at zero-margin. The point is not to go sell the Fortune 500, but to make sure a grad student can spend a $50k grant.
Right now, it's pretty easy to get a few A/H100s (Lambda is great for this), but very hard to get more than 24 at a reasonable price ($~2 an hour). One often needs to put up a 6+ month commitment, even when they may only want to run their H100s for an 8 hour training run.
It's the right business decision for GPU brokers to do long term reservations and so on, and we might do so too if we were in their shoes. But we're not in their shoes and have a very different goal: arm the rebels! Let someone who isn't BigCorp train a model!