San Francisco Compute

sfc is at gtc from march 15th through 19th, swing by in the inception startup area at booth 4031

We're launching a new series on the Village Global Podcast: Worldbuilders. Guest hosted by @sumeet724 — Founder & Managing Partner of Worldbuild — Worldbuilders explores rabbit holes that might turn out to be tunnels into the future. First up is a conversation with @evanjconrad , Founder & CEO of @sfcompute. SF Compute started as an audio model company. They bought a year-long GPU contract, used it for one month, and had to sublease the rest or go bankrupt. Now they're building supercomputers. A few highlights from the episode: "The current scale of the GPU buildout is the largest infrastructure project in the history of the world. It is larger than the cost of wars. It is multiple times the Apollo project." "The bubble isn't in the GPU cloud. The bubble is in whoever owns the equity of those companies that just raised a whole big round. Because if they don't succeed, they can't get that capital back." "We are the company that takes no bullshit. You will never see us say we're gonna democratize compute. We sell GPUs. We rent them to you. The reason why you use us — price, reliability."

Computer use models shouldn't learn from screenshots. We built a new foundation model that learns from video like humans do. FDM-1 can construct a gear in Blender, find software bugs, and even drive a real car through San Francisco using arrow keys.

When sfcompute started, we were this ai lab that had bought a GPU cluster for a year and could only afford a month. Every month, we’d need to find a subleaser or we’d would go under. That forced us into being this accidental cloud. A lot of folks that worked at SFC, rightfully, left during that time. It was more than half the company. The people who stayed formed the seed of what came to be a very gritty, very long-term minded, first principles culture. In Brandon Sanderson’s “Way of Kings” book, there’s a small band of folks who are forced to carry big wooden bridges on their backs, position them over a ravine, so that their army can raid the enemy and then retreat back to safety. Every battle, when they do this the enemy shoots the bridge crews with arrows. And battle after battle, the bridge crews would fall to their death, until eventually they escape from their torment. A lot of the early folks at sfcompute were really into this book and so those early days of the company got referred to by the crew of us remaining as “bridge runs”. However, in the past year, we deleted our old infra, got out of our initial cluster, rebuilt our team, ended up gaining enough of a perspective on the industry to have a real strategy, and started growing. It took 2 years to really do it, but we finally dug ourselves out of the mud. So anyway, now we do “bridge runs”: a few times a week, lots of folks from SFC will go run to Golden Gate Bridge after dark in a group. Anyway, we’d love if you join us, especially if you’re down in a chasm right now. We're doing one tomorrow! Life before death, link below.

Working from @sfcompute today in Levi's Plaza Same location where @alecqfong and I first set up our office 6 years ago!!! Since then went thru YC, startup acquired by NVIDIA, got married, grew a mustache... and Alec's patagonia is still trucking along 😂🤙

@evanjconrad Our writeup on SF Compute: thedeepview.com/articles/how-s…

RT @evanjconrad: What SF Compute does. When you finance a GPU cluster, you need to get an "offtake" agreement. Basically, someone has to…

15x

13x

12x

the san francisco offtake company cannot be stopped. @sfcompute @evanjconrad

San Francisco Compute is the off-take company.

11x

We keep adding clusters and they sell out over the hour

9x

8x

Got off a customer call with a YC founder who reached out mentioning “There’s a class of companies that now that they exist, idk how we lived without them. @lighthousehq_ is one and the other is @sfcompute"

sfcompute is hiring for systems & networking engineers we build the software stack of a gpu cloud, but without the financial burdens of a gpu cloud if you want to make QoS work good or mess with bluefield nics on massive supercomputers, we'd love to work with you our comp includes big chunks of equity in a company that operates more chips than many neoclouds and is currently growing revenue faster than startups like cursor or together we're based in san francisco (remote sometimes ok), with a team of kind people from tesla, lambda, voltage park, & red hat DM or check the link below for more details!

7x