Show HN: Computable – Buy, sell, and redeem GPU for the exact weeks you want

Posted by yuansong98 3 days ago

Counter45Comment26OpenOriginal

Hey we are Computable. We spent years building trading infrastructure at Jump Trading and Coinbase. From that point of view, compute looks like energy markets before 2000: everything trades through private bilateral leases, there’s no visible price, and nothing can be resold. The same H100 rents at a 2x spread depending on who’s asking, and once you sign a 24-month lease, it can never change hands.

So we built a market for GPU nodes, sold by the calendar week. Here are three things you can do on it that you can’t do anywhere else:

- Buy exactly the weeks you need. Three nodes for the last two weeks of October means you pay for those two weeks and walk away when the run is done. You don't have to commit to 6-24 months, and there's no flexibility premium.

- Sell back what you don’t use. If your plans change, unused weeks turn back into cash: we keep a market quote posted on every position, so there’s always a price to hit.

- Buy (or sell) the future. Lock January in July at a price you set, so the cost of a future run is known before another 40% H100 rate jump makes it not.

The first auction is live now at https://market.getcomputable.com: a block of nodes, August through the end of January. Bidding closes July 31. The bids are sealed. You bid how many nodes, which weeks, and your price. Every clearing price gets published after settlement, because nobody knows what a GPU week is worth and we want the prints to exist in public.

Happy to go deep in the comments on anything, including the clearing mechanism, which is a packing problem and was fun to design!

Comments

Comment by hazard 3 days ago

Didn't sfcompute launch with a vision of something like this and then cancel it because CFTC decided it was running an unregulated futures exchange? How are you getting around regulatory issues like this?

Comment by yuansong98 3 days ago

Good question. this is something we built around from day one. Users pay 100% up front for specific nodes on the exact weeks they want. No leverage, no margin, no cash settlement. This is honestly not that much different from you booking a hotel.

Comment by teraflop 3 days ago

What does "no cash settlement" mean? If I sell a reservation, do I not get cash?

Comment by 3 days ago

Comment by yuansong98 3 days ago

So yes, you do get cash. when i said cash settlement, i was referring to how the contract itself ends, unrelated whether you can sell it. A cash-settled contract means it never delivers anything and at expiry it just pays out the difference between your price and an index price, which makes it a pure price bet.

on computable, whoever holds the reservation when the week arrives gets the actual nodes when they redeem for usage. so selling your reservation to someone else beforehand is no different from reselling a concert ticket. and when you sell, the buyer takes over the claim, you get their money, and the contract still ends with someone getting the GPUs.

hope that clears up.

Comment by 3 days ago

Comment by rekpero 1 day ago

I work on a GPU cloud, so this is close to home. You mention “buy” and “sell,” - but from what I can tell the GPU node is effectively just being rented. What’s the practical difference here? Aside from price volatility, as a buyer, if I need GPU access I’d usually just go to a cloud or on-demand provider and rent a dedicated server for 1–2 weeks. What’s the real advantage of bidding and going through this process?

Also, can buyers choose the hardware location they want? And once the auction ends, how are shipping and the other operational details handled?

Comment by flowerbreeze 3 days ago

All the verifications without being able to see what it even does fully feels like pretty a big barrier, even more so than being login gated. Gives it a big banking/gambling vibe too.

Then again, I was only curious how it compares to other GPU rent options in general. Somebody going to bid on something right away might not care.

Comment by spenceryang 2 days ago

Will imagine the team want to make sure they stay more compliant. Google login was easy enough for me though.

Comment by flowerbreeze 2 days ago

[dead]

Comment by paigexu522 3 days ago

Interesting approach. How do you ensure liquidity if buyers and sellers want different weeks? Is there a market maker, or is everything matched through the auction?

Comment by leoqa 3 days ago

I imagine they’re building the supply side as well, where you can bid your fixed GPU. My question with this stuff is how do I know my data is secure on these for-rent boxes? What steps are taken to sanitize the GPU between runs etc.

Comment by 3 days ago

Comment by yuansong98 3 days ago

we take this pretty seriously. we lock down the nodes themselves so that only the hypervisor has direct access. and between customers, we don't just sanitize disks. we actually delete them entirely and spin up a fresh VM from a clean qcow2 base image, so no prior customer data persists.

Comment by Retr0id 3 days ago

How do you prevent sellers from fraudulently selling future capacity they don't actually have?

Comment by yuansong98 3 days ago

Great question. When we list a capacity from a provider, the capacity lists only after it's contracted and attested, and the provider guarantee travels with the claim. If delivery fails, that's on us to remedy, not on the buyer to chase a stranger.

Comment by Retr0id 3 days ago

So in other words, you're screwed?

Comment by yuansong98 3 days ago

we are as screwed as going to a taylor swift concert if she doesnt show up

Comment by Retr0id 3 days ago

More like if you resold tickets that turned out not to be valid, at the ticket gate. At scale.

Comment by yuansong98 3 days ago

providers directly list on us and/or authorize us to resell and redeem. in other words, listing is not permissionless. hence the taylor swift anaolgy.

Comment by carterEthan 2 days ago

Interesting approach. Who do you see as the primary users—AI startups, researchers, or individual developers? How do you ensure GPU availability matches reservations?

Comment by aubreyzhang 3 days ago

curious how you'd keep all parties winning at once on the platform (supply&demand& traders maybe?)

Are you planning to make money on float / spread / flat fee?

Comment by yuansong98 3 days ago

tldr is that the game is not zero sum, it's against the waste in the current setup. Today capacity sits idle on one side while someone overpays for flexibility on the other. An opaque and fragmented market today creates a lot of dead weight loss. here' what we want to see: Providers can sell weeks forward at wholesale instead of letting them sit unsold. More of the calendar monetized, revenue known in advance. Great working capital. Buyers can pay for exactly the weeks you use, resell if they change their mind anytimes. providers want long commitments, buyers want short ones. So on Computable, whoever values January less sells it to whoever values it more, and both do better than the lease they'd have signed.

Traders can ofc buy and sell too. They are great for providers and buyers because they help the market run more efficiently.

Comment by 3 days ago

Comment by zuzululu 3 days ago

how do you deal with data security

Comment by yuansong98 3 days ago

we lock down the nodes themselves so that only the hypervisor has direct access. we take this pretty seriously.

Comment by kevin_thibedeau 3 days ago

State level actors will snoop the bus.

Comment by Retr0id 2 days ago

You don't need to be a nation state to defeat this even if it's set up "properly", you can use low-cost interposer attacks like https://batteringram.eu/ to alias the guest's memory into another guest that you control.

Comment by zeroday404 2 days ago

nice brother

Comment by Lethalman 3 days ago

I need to login to see anything. I walk away.

Comment by 3 days ago

Comment by tinyattackcat 3 days ago

same, I closed that tab so fast

Comment by justacatbot 2 days ago

[flagged]

Comment by TokenLat 3 days ago

[flagged]

Comment by TokenLat 3 days ago

[flagged]