NewsUnverified
FluidStack provides over 18,000 Nvidia B200 GPU hours for GTC hackathon
Thursday, March 5, 2026 at 04:39 PM
A hackathon event was held at GTC involving thousands of B200 GPU compute hours provided by FluidStack, focused on leveraging Nvidia's latest architecture for developer optimization.
Context
In a major hardware-focused demonstration at GTC 2026, FluidStack and SemiAnalysis collaborated to host the "Power to Prefill, Dirt to Decode" hackathon, providing 296 developers with access to Nvidia Blackwell infrastructure. The event featured 18,432 GPU hours on Nvidia B200 chips and $180,000 in compute credits. This collaboration highlights the rising profile of FluidStack, which has recently achieved Gold tier status in the ClusterMAX 2.0 GPU cloud ratings and was selected by Anthropic to deliver custom data centers in New York and Texas.
The deployment is significant as available compute remains scarce; SemiAnalysis reports that on-demand prices for older Hopper units continue to rise while Blackwell clusters are firmly locked up. SemiAnalysis social media confirmed the event's scale, stating, "296 hackers. 18,432 B200 GPU hours. $180K in compute credits. 60,000mg of caffeine." This intensive use of B200 hardware, which features 192GB HBM3e memory and 9,000 TFLOPS of FP8 performance, underscores the industry's shift toward prioritize AI infrastructure over consumer electronics at TSMC’s N3 nodes.
Sources (8)
The Great AI Silicon ShortageJoin Exclusive Tech Events with SemiAnalysis!Happy Sunday and happy @nvidia #GTC Week here in #SanJose ...not much happened today | AINewsDylan Patel — Deep dive on the 3 big bottlenecks to scaling AI computeNvidia B200 vs Nvidia H200Fluidstack: Leading AI Cloud Platform for Training and InferenceBlackwell B200
Related Companies
Nvidia
NVDA