What are the key takeaways from this Latent Space episode?

Key insights include: **GPU Economics vs CPU Economics:** GPU customers exhibit extreme price sensitivity because scaling laws mean every additional GPU generates revenue, unlike CPU customers who hit flat capacity needs. A 10% price difference on $1 billion hardware equals $100 million in value, making customers willing to spend $50 million replicating software rather than paying margin premiums. This destroys traditional cloud provider business models that depend on high-margin software services layered on commodity hardware.; **CoreWeave's Real Estate Model:** CoreWeave succeeded by signing only long-term contracts with creditworthy customers like Microsoft and OpenAI (77% of revenue), then using these contracts to secure prime interest rates from lenders. This approach works because it treats GPUs as a banking business rather than a software company. Hyperscalers likely lose money reselling NVIDIA GPUs because they cannot capture the high margins their CPU businesses require while competing on price with specialized providers.; **The Four Quadrants of GPU Risk:** GPU providers face a risk matrix based on contract length and payment timing. Optimal position: long-term contracts paid upfront to low-risk customers (lowest interest rates). Worst position: short-term contracts at low prices paid over time (highest interest rates). Most GPU clouds died trying to charge high prices for short-term contracts, getting competed down by price-sensitive customers until margins disappeared completely.

What did Sf Compute discuss on Latent Space?

Evan Conrad explains how SF Compute built a GPU marketplace by recognizing that GPU economics fundamentally differ from CPU clouds. CoreWeave succeeded by locking in long-term contracts and treating the business like real estate rather than software. SF Compute creates liquidity through hourly reservations, automated auditing, and plans for cash-settled futures to derisk the entire AI infrastructure market. Key topics include: **GPU Economics vs CPU Economics:** GPU customers exhibit extreme price sensitivity because scaling laws mean every additional GPU generates revenue, unlike CPU customers who hit flat capacity needs. A 10% price difference on $1 billion hardware equals $100 million in value, making customers willing to spend $50 million replicating software rather than paying margin premiums. This destroys traditional cloud provider business models that depend on high-margin software services layered on commodity hardware.; **CoreWeave's Real Estate Model:** CoreWeave succeeded by signing only long-term contracts with creditworthy customers like Microsoft and OpenAI (77% of revenue), then using these contracts to secure prime interest rates from lenders. This approach works because it treats GPUs as a banking business rather than a software company. Hyperscalers likely lose money reselling NVIDIA GPUs because they cannot capture the high margins their CPU businesses require while competing on price with specialized providers..

How long is this episode of Latent Space?

This episode is 72 minutes long. SignalCast provides an AI-generated summary so you can get the key insights in about 3 minutes.

Latent Space

SF Compute: Commoditizing Compute to solve the GPU Bubble forever

April 11, 2025

72 min episode · 3 min read

Sf Compute

Episode

72 min

Read time

3 min

Topics

Investing, Startups, Fundraising & VC

AI-Generated Summary

Published Jan 31, 2026

Key Takeaways

✓GPU Economics vs CPU Economics: GPU customers exhibit extreme price sensitivity because scaling laws mean every additional GPU generates revenue, unlike CPU customers who hit flat capacity needs. A 10% price difference on $1 billion hardware equals $100 million in value, making customers willing to spend $50 million replicating software rather than paying margin premiums. This destroys traditional cloud provider business models that depend on high-margin software services layered on commodity hardware.
✓CoreWeave's Real Estate Model: CoreWeave succeeded by signing only long-term contracts with creditworthy customers like Microsoft and OpenAI (77% of revenue), then using these contracts to secure prime interest rates from lenders. This approach works because it treats GPUs as a banking business rather than a software company. Hyperscalers likely lose money reselling NVIDIA GPUs because they cannot capture the high margins their CPU businesses require while competing on price with specialized providers.
✓The Four Quadrants of GPU Risk: GPU providers face a risk matrix based on contract length and payment timing. Optimal position: long-term contracts paid upfront to low-risk customers (lowest interest rates). Worst position: short-term contracts at low prices paid over time (highest interest rates). Most GPU clouds died trying to charge high prices for short-term contracts, getting competed down by price-sensitive customers until margins disappeared completely.
✓SF Compute's Market Mechanism: The platform enables hourly GPU reservations that function like perishable inventory—prices drop continuously until compute clears the market, similar to day-old milk. Users can set limit prices (e.g., $4/hour maximum) to buy at spot rates (often $0.80-$0.90) while avoiding price spikes. This creates liquidity pockets where customers needing one month of compute can buy from others selling back unused portions of year-long contracts.
✓Automated Auditing Infrastructure: SF Compute runs LINPACK burn-in tests for 48 hours to seven days to kill dead GPUs before deployment, then implements active and passive performance checks during operation. The company controls clusters via BMC access (remote reimaging capability) and provides automated refunds when hardware fails. This standardization layer enables fungible GPU contracts necessary for a functioning marketplace, unlike bespoke cluster arrangements that lock in buyers.

What It Covers

Evan Conrad explains how SF Compute built a GPU marketplace by recognizing that GPU economics fundamentally differ from CPU clouds. CoreWeave succeeded by locking in long-term contracts and treating the business like real estate rather than software. SF Compute creates liquidity through hourly reservations, automated auditing, and plans for cash-settled futures to derisk the entire AI infrastructure market.

Key Questions Answered

•GPU Economics vs CPU Economics: GPU customers exhibit extreme price sensitivity because scaling laws mean every additional GPU generates revenue, unlike CPU customers who hit flat capacity needs. A 10% price difference on $1 billion hardware equals $100 million in value, making customers willing to spend $50 million replicating software rather than paying margin premiums. This destroys traditional cloud provider business models that depend on high-margin software services layered on commodity hardware.
•CoreWeave's Real Estate Model: CoreWeave succeeded by signing only long-term contracts with creditworthy customers like Microsoft and OpenAI (77% of revenue), then using these contracts to secure prime interest rates from lenders. This approach works because it treats GPUs as a banking business rather than a software company. Hyperscalers likely lose money reselling NVIDIA GPUs because they cannot capture the high margins their CPU businesses require while competing on price with specialized providers.
•The Four Quadrants of GPU Risk: GPU providers face a risk matrix based on contract length and payment timing. Optimal position: long-term contracts paid upfront to low-risk customers (lowest interest rates). Worst position: short-term contracts at low prices paid over time (highest interest rates). Most GPU clouds died trying to charge high prices for short-term contracts, getting competed down by price-sensitive customers until margins disappeared completely.
•SF Compute's Market Mechanism: The platform enables hourly GPU reservations that function like perishable inventory—prices drop continuously until compute clears the market, similar to day-old milk. Users can set limit prices (e.g., $4/hour maximum) to buy at spot rates (often $0.80-$0.90) while avoiding price spikes. This creates liquidity pockets where customers needing one month of compute can buy from others selling back unused portions of year-long contracts.
•Automated Auditing Infrastructure: SF Compute runs LINPACK burn-in tests for 48 hours to seven days to kill dead GPUs before deployment, then implements active and passive performance checks during operation. The company controls clusters via BMC access (remote reimaging capability) and provides automated refunds when hardware fails. This standardization layer enables fungible GPU contracts necessary for a functioning marketplace, unlike bespoke cluster arrangements that lock in buyers.
•Path to GPU Futures Market: SF Compute builds toward cash-settled GPU futures to derisk the entire AI supply chain. Currently, startups absorb hardware risk, pushing it to VCs who write inflated pre-revenue valuations to fund $50 million GPU purchases. A futures market would let data centers lock in prices without forcing customers into year-long commitments, reducing systemic risk similar to how agricultural futures stabilized farming economics by separating price risk from physical delivery.

Notable Moment

Conrad reveals SF Compute started accidentally when they signed a year-long GPU contract to train music models, planned to use one month and sublease eleven months, then owed $500,000 monthly with only $500,000 in the bank. For twelve months, failing to sell out the cluster meant bankruptcy. This forced survival mode transformed into understanding GPU brokerage economics better than anyone, eventually building the market infrastructure they originally needed as desperate customers.

Know someone who'd find this useful?

Episode Transcript

Hey, everyone. Welcome to the Living Space podcast. This is Alessio, partner and CTO at Decibel, and I'm joined by my colleague Swyx, founder of Small AI. Hey. And today, we're so excited to be finally in the studio with Evan Konrad from SF Compute. Welcome. Hello, how good is it? How are you? I've been fortunate enough to be your friend before, you're famous, and also we've hung out at, like, various social things. So it's really cool to see that SF Compute, like, is is coming into its own thing and it's, you know, it's it's a significant presence, at least in the San Francisco community, which, of course, it's in the name. So you couldn't help but be. Indeed. Indeed. I think we have a long way to go, but yeah. Thanks. Of course. Yeah. One way I was thinking about kicking off this conversation is we will likely release this right after Core Reef IPO. Uh-huh. And and, I was watching I was looking doing some research on you. You did a talk at the curve. Yeah. I think I may have been viewer number 70. It was a great talk. More people should go see it, Evan Conrad at The Curve, but we're we have, like, three orders of magnitude more people. And I just wanted to to highlight, like, what is your analysis of what Core Eve did that that went so right for them? So locked in long term contracts and don't really do much short term at all. I think, like, a lot of people had this assumption that GPUs would work a lot like CPUs. And the, like, standard business model of any sort of CPU cloud is you buy commodity hardware, then you lay on services, that are mostly software, and that gives you high margins. And pretty much all your value comes from those services, not really the underlying compute in any capacity. And because it's commodity hardware and it's not actually that expensive, most of that can be, sort of on demand compute. And while you do want locked in contracts for folks, it's mostly just to sort of de risk your situation. It helps you plan revenue because you don't know if people are gonna scale up or down. But, fundamentally, people are, like, buying hourly, and that's how your business is structured. And you're gonna make 50% margins or higher. This, like, doesn't really work, in GPUs. And the reason why it doesn't work is because you end up with, like, super price sensitive customers. And that isn't because necessarily, it's just way more expensive, though that's totally the case. So in a CPU cloud, you might have, like, you know, let's say, if you had a million dollars of hardware, in GPUs, you have a billion dollars of hardware. And so your customers are buying, at much higher volumes than you'd otherwise expect. And it's also smaller customers who are buying at higher months of …

Get the full transcript (15,719 words) + summary by email — free

One-time email with the complete transcript and AI summary of this episode. No account needed.

One email, no spam. We’ll also show you what SignalCast does.

Browse all Latent Space transcripts →