Mythic has engineered the Analog Processing Unit (APU) — a ground-up rethink of AI compute that moves memory inside the processor itself, eliminating the energy-draining data movement that makes conventional AI hardware unsustainable at scale.
The Problem Data transport accounts for 90% of the energy consumed by AI inference — a structural flaw baked into chip architectures dating back to the 1940s. As AI workloads explode, data centers face multi-gigawatt grid upgrades, ballooning costs, and regulatory pressure. Meanwhile, edge deployments — from autonomous vehicles to industrial sensors — are throttled by power envelopes that today’s GPUs routinely blow through.
What Makes It Different Using patented Analog-In-Memory Computing (AiMC), Mythic executes calculations directly inside flash memory using Ohm’s and Kirchhoff’s Laws — no data ping-pong, no cloud latency, no external memory bottleneck. The result: 120 TOPS per watt and a 100x performance-per-watt advantage over digital benchmarks. At 4 cents per million tokens for 1T parameter models, it undercuts Blackwell’s optimized 10 cents by more than half. The CAMP SDK ensures drop-in compatibility with PyTorch, TensorFlow, ONNX, and NVIDIA Jetson — no workflow overhaul required. Mythic is also Honda’s strategic partner for next-generation intelligent vehicle platforms.
Unlike competitors still simulating analog, Mythic ships production-ready silicon today.
When compute lives where the data lives, the energy bill — and the power crisis — shrinks with it.
