Mythic is an AI chip company based in Palo Alto, California, which has recently raised $125 million to develop an analog processing unit (APU) that could be up to 100 times more energy efficient than existing GPUs. This round was led by DCVC, a deep tech venture capital firm, and the funds will be used to advance Mythic's product development, software improvements, and commercial scaling.
Other investors include NEA, Atreides, Future Ventures, Softbank KR, S3 Ventures, Linse Capital, One Madison Group, and Catapult, as well as Honda and Lockheed Martin. Mythic said the funding was raised after CEO Taner Ozcelik, who previously worked at NVIDIA and has extensive industry experience, restructured the company. Ozcelik noted that energy efficiency will be crucial for future AI computing.
Mythic plans to apply its chips across data centers, automotive systems, robotics, and defense. As a 13-year-old company, Mythic will also use the funds to rebuild its architecture, software platform, and strategy.
Additionally, Mythic has launched a sub-watt sensing platform called "Starlight," which integrates its chips into image sensors. This system enhances signal extraction in low-light conditions, primarily targeting applications in defense, automotive, and robotics.
Mythic's chips are manufactured using standard semiconductor processes in the United States and its allied countries. The company plans to use the new capital to expand production, refine its software development kit, and seek commercial deployment in the AI inference market. Mythic's chips employ analog memory computing technology, combining storage and processing on a single plane, thereby reducing energy loss during data transmission. Mythic claims that its current architecture can deliver 120 trillion operations per watt.
Ozcelik emphasized that the company does not intend to replace GPUs but hopes to complement them. He said, "Just as GPUs became accelerator computers next to CPUs, our APU will become an accelerator computer next to GPUs." According to internal benchmark tests conducted by Mythic, its chips can run large language models with up to one trillion parameters without high-speed interconnects, and the number of tokens processed per watt exceeds that of current high-end GPUs.
Aaron Jacobson, a partner at NEA, said that Mythic's platform breaks through current limitations in energy and cost, providing the company with space for scaling. Steve Jurvetson, a future investor, said that Mythic's approach "unifies computation and storage like a brain," improving efficiency.
Key Points:
🌟 Mythic raised $125 million and plans to develop an ultra-energy-efficient analog processing unit to challenge GPUs.
🚀 The newly launched "Starlight" platform enhances signal extraction in low-light environments and is widely applied in multiple fields.
💡 Mythic's chips reduce energy loss from data movement, claiming 120 trillion operations per watt.
