GPU Perpetual Futures Prototype
GPU rental prices are super volatile but there's no derivatives market to hedge. I built a perpetual futures platform to see what this could look like. The idea is airlines hedge jet fuel, starbucks hedges coffee beans - as GPU compute becomes critical infrastructure the same hedging tools should exist. Not sure if anyone actually needs this but it was interesting to build. How it works: - Pulls live H200 spot prices from Vast.ai every 15s into a tradeable index - Full perp mechanics: funding rates, mark price calc, real-time P&L - Event-driven Rust backend with supervisor pattern and…
What it does
In the maker’s words, at launch
GPU rental prices are super volatile but there's no derivatives market to hedge. I built a perpetual futures platform to see what this could look like. The idea is airlines hedge jet fuel, starbucks hedges coffee beans - as GPU compute becomes critical infrastructure the same hedging tools should exist. Not sure if anyone actually needs this but it was interesting to build. How it works: - Pulls live H200 spot prices from Vast.ai every 15s into a tradeable index - Full perp mechanics: funding rates, mark price calc, real-time P&L - Event-driven Rust backend with supervisor pattern and circuit breakers - Next.js frontend with TradingView charts, real-time WebSocket updates What's real vs simulated: - Real: Index construction, funding rate engine, forward curve, state persistence - Simulated: Order book depth and trade matching (its a single-client demo) The backend is the part I'm most proud of - isolated tasks coordinated by a supervisor, each has it's own state machine so if one component fails it doesn't take down the others. Tried to build it with production patterns in mind even though its just a demo. Also made a 15-page derivatives pricing doc that covers the economic model and hedging scenarios. Basically: rental prices = f(CAPEX, utilization, depreciation) so futures pricing reveals market expectations about GPU supply/demand. GitHub: https://github.com/zacharyfrederick/compex Would love feedback on the architecture or if the market mechanics actually make sense. First time building something like this.
Does the same job
all alternatives →
- GMGPU market is absurd! So I built a dashboard of pricing/restock trends2025 · gpuisfine.singhkays.com · ▲6
Hey HN! This idea started with me not being able to buy a GPU and constantly losing to bots/scalpers. I figured I'd use this as way to see how far I can get with 'vibe-coding and designing'*. The end result was pretty far! Here are more details of behind the scenes. In a future blog post, I'll detail behind the scenes process of building this. - The landing page is React/Typescript/Tailwind.css (which I've never used before) - The dashboard is based on Evidence.dev - which is SQL queries in Markdown + little bit of custom Javascript for chart formatting (again never used…
- AGA GPU group-buying layer to get enterprise rates (~50% off H100s)2025 · trysaga.ai · ▲5
We built, saga[1] a layer that lets smaller teams access enterprise GPU discounts through collective buying power. How it works: 1. Aggregate GPU spend across hundreds of ML teams 2. Get enterprise rates through combined volume 3. Pass savings to users, monetize via provider partnerships Technical notes: - Works at billing layer only (no access to code/data) - Supports existing cloud setups or managed GPUs - Private beta running since January, opening more spots for March - Currently seeing ~50% savings on H100s/A100s [1] https://trysaga.ai
- CECategorised eBay GPU Prices2024 · gpu-prices.com · ▲9
I've made a GPU comparison site: https://gpu-prices.com/US/ Yes, there was one yesterday - I got beaten to it. This one is different in that I've put a lot of work into categorisation, so you can filter down quite precisely. For example, VRAM-per-dollar, limiting to Nvidia, and a minimum performance score allow you to find good ML GPUs. It currently supports Australia, Canada, Ireland, the UK, and the US. The tech stack is Python for data pull and static site generation then Cloudflare Pages for actually serving the site. It updates three times a day, but I could increase…
More ai this month
the category →
I trained a 125M-parameter transformer to autocomplete piano performances in real time (~108 notes/sec on an iPhone 15). The idea is basically GitHub Copilot or Tabnine, except instead of prompting it with code, you prompt it by playing a few notes on a MIDI piano. The model then continues what you played, entirely on-device. The app is free if anyone wants to try it. Happy to answer questions about the model, training, Core ML, or the many things that didn't work.
AI · 16d ago · simedw.com
Astute▲585Automate your B2B brand going viral, with new media creators
AI · 18d ago · company-app.joinastute.com
Hey HN, Henry from Cactus here! We previously released Cactus Needle, a 14MB agentic LLM for tool call, device use, and structured extraction for phones, wearables, smart homes, small robots and microcontrollers. We got really great feedback here, and have now incorporated the suggestions to release Needle 2. The whole model is a single 14MB binary that runs a full session in 28MB of RAM; 45m parameters at 2bit compression. Needle hits 500 tokens/sec decode speed on a Raspberry Pi 5, sits between 400-1,500 tokens/sec on VR devices like Meta Quest 3S and Apple Vision Pro, and ranges…
AI · 26d ago · cactuscompute.com
