San Francisco Compute

We build large scale gpu clusters and sell them on contracts you can sublease.

[Nikta Khani @_nikta_khani]

So excited to present Market Price, a dinner series bringing together operators, investors, and traders working on compute, and how it's priced, financed, and moved. @sfcompute is partnering with @Midnight_Captl @computeexchange for our first one next week.

DMs are open for the last few spots at the table and for those who are interested in partnering on future events. Venues, dinner co-hosts, and new friends welcome!

[Alex Konrad @alexrkonrad]

AI has a credit problem. And it's maybe not the one you think.

I spoke with SF Compute CEO Evan Conrad (@evanjconrad) to talk about how the current compute market pushes AI labs to put down huge deposits upfront -- helping to spark some of the sky-high valuations of $10B, $20B or $30B you're seeing in tech today.

Conrad's own startup got into this business accidentally when he realized he was on the hook for $400K in GPUs each month, with only $500K in the bank. Now it's built what he calls the first physically-settled market for compute, recently signing a $245M contract with a lab.

But that's just a drop in the bucket in the huge and growing compute industry -- one where data centers are getting dropped into chicken coops, and not every H100 or B300 is created equal.

On The Upstarts Podcast, Conrad talks about why credit, not land or power, is AI’s true bottleneck; why upfront compute costs are driving eye-popping valuations from VCs; and why data centers could become geopolitical bargaining chips.

Plus, he shares his Upstart Moment: pitching early investors on a new “power grid” for AI they just couldn’t understand.

This season's founder conversations are presented by @Atlassian and @chargebee 🫡

CHAPTERS:
00:00 Introduction
01:25 Inventing A Market For Supercomputing
04:56 The Stripe For GPU Clusters
07:27 Chicken Coops, Soybeans And Monsoons 11:01 “Anyone Who Competes With Us Is Stupid”
13:47 Arming The Rebels, Not Just The Big Labs
18:00 Getting Into YC As A Therapy Exercise
19:46 Dalton Caldwell’s Advice: “Don’t Die”
23:01 When $245 Million Isn’t A Big Deal
26:10 Upstart Moment: “There Isn’t An Elevator Pitch”
28:42 Why Compute Futures Don’t Work Like Wheat 31:55 No More Massive Deposits: SF Compute’s Vision
34:47 GPU Prices, Geopolitics And A Market Bigger Than Oil
37:39 Where The Most Money Can Be Made In Compute Today

Thanks for watching!

[brian @brianzhangg3]

life in 2026:
>> AI companies need GPUs
>> GPUs cost a lot of $$$
>> the lenders who finance them have gotten picky, so startups are getting asked to prepay an ungodly amount upfront, or don't even qualify as a customer because they don't have the credit

tmr night in SF @sfcompute is hosting a Q&A session led by @evanjconrad on what this means for startups, VCs, and anyone buying GPUs / needs compute

starts at 5:30pm w/ food and drinks! register here: https://t.co/sEwjtDXWlk

[Alex Konrad @alexrkonrad]

coming tomorrow: the Konrad / Conrad crossover episode 👀

@evanjconrad and I talk GPUs, wild AI valuations and why you would have to be "stupid" to want to compete with @sfcompute's business, on The Upstarts Podcast

[MTS @MTSlive]

SF Compute CEO @evanjconrad advises AI founders to raise 10X more capital than planned, potentially $1 billion, just to secure compute:

"I would raise as much money as I can. I would take your round and I would add 10X to it. And if I was an investor, I would buy compute instead."

"I would either try to find a big investment grade normal company that I can partner with and use their balance sheet to co-sign on the lease with me, or I would try to raise the biggest round I possibly can, so I’m above the bar."

"If you’re trying to raise a round, I think you should try and raise like a billion dollars. I know. That’s the problem. This is why the article was called End of Startups. I don’t think many people can raise a billion dollar round."

"If you don’t fix the debt problem, most of the portfolios of various venture firms are gonna consolidate or get wiped out, or they will convert into basically GPU clouds."
@sfcompute

[evan conrad @evanjconrad]

if you care about startups, have a balance sheet, and you would like to backstop this at a scale that is larger than what we can, we would like to work with you

if we don't fix this, we believe the industry will consolidate into a few large credit worthy players

dms open

[Dean W. Ball @deanwball]

This seems right to me, unfortunately, and it is essentially a species of interest-rate rise (the opportunity cost of capital is increasing). It is true that forward and futures markets for compute, and making compute into a financial commodity, would help here.

[evan conrad @evanjconrad]

https://t.co/FS9whe3h3L

[Latent.Space @latentspacepod]

The Programmable Data Center

@martin_casado’s Nicira virtualized networking by separating the control plane from the hardware. Cisco offered $600m; VMware bought it for $1.3b.

Now that idea is coming to GPU infrastructure.

@evanjconrad, CEO of @sfcompute, joins @swyx at @aiDotEngineer NYC to explain how SF Compute went from subleasing a cluster to making data centers programmable.

We get into:
• Why delivering a GPU trade means running the whole “hotel”
• Interruptions, hardware generations and the GPU order book
• Why physical settlement comes before compute derivatives
• Why Evan thinks NVIDIA can’t build an independent market
• Programmable InfiniBand and the data center as software

The full conversation:

[Haight Street Compute @hscompute]

Introducing the Haight Street Compute Market.

A one-day trading floor where vetted buyers and sellers of compute meet and sign deals in person. Nov 17, at the old Federal Reserve building in San Francisco.

Request your invite: https://t.co/g2evZ8I3hp

[Ethan W. Anderson @Ethan_is_online]

Everyone wants to be a “compute market” until it means sending your CTO into a monsoon to vet a data center

Eric park is a beast btw. https://t.co/yKhBHyOANX

[San Francisco Compute @sfcompute]

SF Autoresearch
- has capacity
- for training runs
- with IB & cluster local storage
- without a long term contract
- without a big deposit
- without a sales conversation
- on an easy to use PaaS

[evan conrad @evanjconrad]

SF Autoresearch (formerly givemeanode) now supports jobs with full InfiniBand fabrics!

That means you can queue hero runs into all of SFC's available capacity, without a long term reservation, a long lead time, or a big downpayment.

https://t.co/82OFH8coMO

[Matt @mmoderwell]

New .cube viewer on Ouro. Explore 3D or 2D slices of charge density, spin density, bonding (deformation) density, and orbitals.

Heavily inspired by @gp_0013's work visualizing electron (+spin) density and s/o his latest post with @sfcompute, the blog aesthetic is beautiful.

[TBPN @tbpn]

SF Compute CEO @evanjconrad compares today’s compute market to electricity before the power grid.

He explains that before utilities pooled demand, factories had to build enough generation for their own peak usage, leaving expensive capacity idle the rest of the time.

“Every factory ran its own generator. And they had to size that generator for peak capacity. Then during idle hours they would just sort of burn that money.”

“What happened in the power markets is that utilities came in and they pooled demand between all of these different factories.”

“By doing that, you made the economics of all the factories better.”

“This is how we scaled up electrification in the country.”

“This is what people are trying to do with compute markets.”

[evan conrad @evanjconrad]

We're announcing Elastic InfiniBand Partitions in SFC.

You can now get RDMA connected nodes reserved, change plans, and sell your whole contract on the fly. That’s not “sell to spot”, that’s “sell the whole contract”.

Launch the 30-day hero run on Monday, hit the snag on Tuesday, get out of the contract on Wednesday.

Running a PaaS and your customers need their own partitions? Create thousands of partitions, one for each customer. Lots of providers give IB, but all on the same partition, which means it’s not secure enough for production.

https://t.co/Bk7yho9Ldq

[Nikta Khani @_nikta_khani]

Hiring 2 people to join my team at @sfcompute. One person to help me with supply-side compute deals and another to manage cluster stand-up. If you're bullish on AI infra, compute markets, sovereign AI, like to win, and get back up quickly when you don't, drop me a line.

https://t.co/yfBRKk7t3a

[evan conrad @evanjconrad]

We sat down with Rich Jaycobs, the former president of Cantor Futures Exchange & legendary creator of 7 exchanges & 6 clearing houses, and 1/5 of all approved DCMs since 2000.

Here's what it really takes to build a compute market.

https://t.co/fJGvFVq2Jx

[ComputeSDK @computesdk]

Sandbox Dax has just finished running and the results are in!

🥇@isorun
🥈@namespacelabs
🥉@sfcompute

[Sebastien La Duca @sebastienladuca]

We just put out a writeup on an in-house ledger we built to keep track of compute reservations as they're resold! Check it out :)

https://t.co/CF9nkgMZaY

[San Francisco Compute @sfcompute]

@StepanVC sfcompute specifically works to solve the problem you're describing

it does not scoop up idle capacity and resell it for a margin