French Startup Kog Claims 30x Faster LLM Inference on Existing GPUs Through Low-Level Reverse Engineering

TechCrunch·medium signalXBlueskyLinkedInCopy link

TechCrunch profiled Kog on August 14, 2026, a French startup arguing the belief that GPUs are poorly suited to agentic workflows is a misconception — its Kog Inference Engine claims up to 30x decoding speedups on standard NVIDIA and AMD datacenter GPUs with no new hardware. The approach is low-level reverse engineering and hardware-aware optimization of memory bandwidth utilization rather than a new architecture; the company emerged from stealth in May 2025, is backed by Bpifrance and French Tech 2030, and reports 200+ business leads. Treat the 30x as a vendor claim on selected workloads until independently benchmarked, but the thesis directly contradicts the custom-inference-silicon narrative.