Meta and NVIDIA Team Up for AI Hackathon
Meta and NVIDIA are hosting a hackathon at GTC on March 14 focused on building high-performance GPU kernels. Participants will use PyTorch's Helion DSL, with prizes including a DGX Spark and Meta glasses, highlighting the companies' push to cultivate talent in high-performance AI.
This hackathon is a glimpse into a massive, multi-year partnership between Meta and NVIDIA. Meta is undertaking a colossal AI infrastructure build-out, deploying millions of NVIDIA's current Blackwell and future Rubin GPUs to power its data centers for both AI training and inference. The event's focus, Helion DSL, is a Meta-led PyTorch project designed to simplify writing high-performance code. It acts as a higher-level language than NVIDIA's own Triton (which it compiles to), automating complex tasks like memory management and hardware-specific tuning to bridge the gap between developer productivity and raw performance. The hackathon's venue, GTC, is NVIDIA's premier global AI conference, attracting over 30,000 developers, researchers, and business leaders to San Jose. The conference, which originated in 2009 with a focus on GPU computing, is now a key event for major AI announcements, centered around a highly anticipated keynote from CEO Jensen Huang. The grand prize, a DGX Spark, is a desktop AI supercomputer powered by NVIDIA's GB10 Grace Blackwell Superchip. Aimed at developers for local AI model creation, NVIDIA recently increased its MSRP by $700 to $4,699, citing constraints in the global memory supply. This collaboration extends beyond GPUs to the entire data center stack. Meta is adopting NVIDIA's Arm-based Grace and future Vera CPUs to improve performance-per-watt, along with Spectrum-X Ethernet for AI-optimized networking at scale. A concrete outcome of their joint work is already live within Meta's products. NVIDIA's Confidential Computing technology is being used to enable AI-powered features in WhatsApp while ensuring user data remains private and secure.
