Mythic, a Palo Alto-based AI chip startup, has raised $125 million to develop analog processing units that deliver up to 100 times greater energy efficiency than conventional GPUs. Led by deep tech investor DCVC, the round includes NEA, Atreides, Future Ventures, SoftBank KR, S3 Ventures, Linse Capital, One Madison Group, Catapult, and strategics like Honda Motor and Lockheed Martin. The funding follows a restructuring under CEO Taner Ozcelik, ex-NVIDIA VP, to tackle AI’s power consumption crisis.
Technology and Applications
Mythic’s analog in-memory computing fuses memory and processing in a single plane, eliminating energy waste from data shuttling between components. This brain-inspired design suits inference for large language models up to one trillion parameters without GPU-style high-speed interconnects. Internal tests show higher tokens-per-second-per-watt than top GPUs.
The company launched Starlight, a sub-one-watt platform embedding chips in image sensors for superior low-light performance. Deployments span data centres, automotive ADAS, robotics, and defence surveillance. Manufacturing occurs in the US and allied fabs using standard processes, ensuring supply chain resilience.
Ozcelik envisions APUs as GPU companions, akin to how GPUs complemented CPUs. “Energy efficiency will define the future of AI computing everywhere,” he stated, targeting inference markets over training dominance.
Investor Confidence and Market Fit
Investors see Mythic collapsing AI’s energy-cost barriers. NEA’s Aaron Jacobson highlighted scalability in power-limited settings. Future Ventures’ Steve Jurvetson praised the unified compute-memory model mimicking neural efficiency.
The 13-year-old firm will mature its software kit and ramp production for commercial rollout. This positions Mythic amid surging AI hardware demand, where power grids strain under hyperscaler expansions.
India Angle in Semiconductor and AI Ecosystem
India’s data centre capacity triples by 2026, but renewable shortages amplify efficiency needs. GCCs in Bengaluru and Pune, plus manufacturing hubs, benefit from edge AI chips reducing cloud dependency. Defence applications align with indigenous drone and border tech pushes.
ATMP investments in Gujarat and Assam gain from low-power alternatives to import-heavy GPUs. DPDPA compliance favors sovereign inference stacks. Bengaluru’s analog IC design talent could partner for localization, supporting PLI schemes and green IT goals.
Mythic’s approach aids India’s AI ambitions without infrastructure overhauls, balancing growth with sustainability in a power-constrained landscape.
