Fluidstack is partnering with @AnthropicAI to deliver their $50B investment in American AI infrastructure.
We are proud to work with them to create American jobs and bolster American competitiveness.
We're singularly focused on delivering 10 to 100s of GWs of compute faster than anyone else, rethinking every layer of the stack. Speed and scale are our key differentiators.
Join @fluidstack to build civilization-scale infrastructure for AI.
Software Engineer:
It was an unforgettable event that wouldn't have been possible without the support of @SemiAnalysis_ 🖤
Special shoutout to everyone that joined us for the afterparty at the bowling alley next door. A well needed celebration for everyone.
See you at the next one!
Honorable Mentions:
Best Tokens — Tinker-Nomics by Simon Guo (RL cost calculator, reverse-engineered Tinker's stack)
Best GPU — Speculative Prefill by Kunal Menda (6.2x latency reduction on voice agents, single B200)
Best Slurm — Bridge Engine VLA by Terresa (2.2B VLA robot
3rd Place: Tuvaluan LLM
Luc Chartier and Wiktoria Leks built the largest Tuvaluan-English corpus, fine-tuned Qwen3-30B, shipped a live Tuvaluan football news app, and beat GPT-5.4 on the Tuvaluan benchmark.
An LLM for an island nation of 11,000 people. It's awesome.
2nd Place: GPU-accelerated OLAP database
Ryan Yang built it on multi-GPU + NVLink. Beat every ClickBench submission by 24%. Better price-performance than every top CPU entry.
1st Place: quectoGPT
Distributed GPT training across browsers using WebGPU. Visit a website, pull weights, run forward/backward prop, send gradients back. BPE tokenizer over Shakespeare data. GPT-2 implemented entirely in WebGPU.
Built by Vivek and Sreejith.
300 people showed up and started building. The room was loud, focused, chaotic in the best way.
60,000 mg of caffeine and 9 hours later, submissions closed. We talked to every team we could. The technical level across the board was mind-blowing. The judges picked their winners.
The event was sold out. 300 registered, 100+ on the waitlist. Nvidia and OpenAI committed to sponsor. Two DGX Sparks signed as prizes. One by Jensen himself.
The next morning, the team set up the venue. It was show time.
Reflecting on the “From Silicon to Scale” hackathon at GTC 2026.
With 296 builders and 48 hours of intense innovation, we unleashed 18,432 B200 GPU hours and $180k in credits!
Plus, we had speakers and prizes from @nvidia, @OpenAI, @thinkymachines, and @GPU_MODE. Here's what
Winners have been crowned. 🏆
🥇 [1ST PLACE] 🥈 [2ND PLACE] 🥉 [3RD PLACE]
Incredible builds from an incredible group. Thank you Fluidstack for making this happen at #GTC2026🔥 @dylan522p
Let the hacking begin.🔨
Live from San Jose — kicking off NVIDIA GTC with @fluidstack🚀
Big thanks to our speakers for getting things started: @dylan522p | @marksaroufim (GPU Mode) | @cHHillee (Thinking Machines) | Thomas Raoux (OpenAI) | Gary Wu (FluidStack)
This is just