What happened
tiny corp, developer of the tinygrad framework, launched and ships its "tinybox" deep learning computers. The $12,000 Red v2 offers 778 TFLOPS (FP16) with 4x 9070XT GPUs and 64GB RAM. The $65,000 Green v2 delivers 3086 TFLOPS (FP16) with 4x NVIDIA RTX PRO 6000 Blackwell GPUs and 384GB RAM, per tiny corp. Both models are in stock, shipping weekly. A future Exabox is projected for 2027 at ~$10M. tiny corp claims "best performance/$", citing MLPerf Training 4.0 benchmarks.
Why it matters
tinybox systems shift the cost curve for deep learning infrastructure, increasing high-performance compute accessibility for smaller teams. Procurement teams face a new build-versus-buy decision; tiny corp's direct-to-consumer model and wire transfer-only payment streamline acquisition but limit traditional vendor relationships. Platform engineers gain a compact, high-density compute option, potentially reducing data centre footprint. This commoditises petaflop-scale compute, challenging established hardware vendors and lowering AI development entry barriers.
