technologyScore 75/100Watch
George Hotz building TinyGrad to run LLMs on any device; positioning against NVIDIA GPU monopoly via software stack commoditization
“I started Tiny as a like a toy project just to teach myself... If Nvidia becomes a monopoly here, um, how long before Nvidia is nationalized?... It's to make sure power stays decentralized. And here it's uh computational power. And you Nvidia is kind of locking down the computational power of the world.”
Why it matters
TinyGrad reduces the engineering complexity of porting ML models to non-NVIDIA accelerators by 10x (25 ops vs 250 in PyTorch), lowering the barrier for AMD, Intel, and other GPU competitors to capture workload-specific AI inference.
Investment implication
Success of TinyGrad as a neutral ML framework weakens NVIDIA's software moat and could accelerate adoption of alternative accelerators (AMD 7900 XTX, Intel Arc, Qualcomm). This threatens NVIDIA's high-margin software licensing and bundled ecosystem lock-in.