Mythic AI

Market position

Mythic competes in a crowded edge AI chip market alongside startups including SiMa.ai, Axelera, Flex Logix, NeuReality, EnCharge, Hailo and Kneron, and positions its planned data-center systems against Nvidia on total cost of ownership and power consumption.

The company claims very high performance per watt from analog compute-in-memory — more than 4 TOPS/Watt on first-generation parts and, per company claims, up to 100x the performance per watt of traditional graphics cards — along with single-frame latency at batch size 1 and the ability to use more mature, lower-cost process nodes than bleeding-edge digital designs. Low power and thermal output allow 3D stacking and placement of a compute chiplet directly beneath an image sensor's photodiode array, which the company says is not feasible for a conventional DSP.

Technology

Analog compute-in-memory using Flash transistors that store analog values, operated at 256 sub-threshold levels for 8-bit compute, with heavy calibration and compensation to maintain accuracy. Data is represented as fluctuations in a continuous electrical signal rather than discrete digital signals, and model layers can be executed in parallel. Designs pair a Flash-based analog compute die with a digital control die in a 3D-stacked configuration; next-generation parts are chiplet-based and support 8-bit and 4-bit computation with an on-chip DSP for unsupported operators. Future generations are planned to add NAND-Flash-based arrays alongside the existing NOR version to fit larger models on a chiplet. The software toolkit compiles and optimizes dataflow graphs, applies quantization and can retrain models for the APU.

Go-to-market

Following its 2023 restructuring the company refocused on defense as its primary vertical, with secondary emphasis on public safety, industry and consumer applications, and increased reliance on development partners and off-the-shelf components to reduce cost and time to market. It plans to build the Starlight sensor and AI accelerator combination itself, with both the accelerator chiplet and the image sensor product scheduled to tape out in 2026.

Systems builders in defense, robotics, automotive, video surveillance, public safety, industrial machine vision and consumer devices, plus prospective data-center operators running large language model inference. Lockheed Martin has been described as the company's largest customer to date; its corporate venture arm is also an investor.

Geography

Headquartered in Austin, Texas, with an India organization headed by a VP of Mythic India and an Israel Sensor Innovation Center led by a VP; a legal executive is based in Portland. The leadership page also lists the CEO of videantis, a German-founded video processing platform company.

History

After raising more than $170 million across multiple rounds from investors including Hewlett Packard Enterprise and BlackRock, Mythic reportedly ran out of capital in November 2022 and underwent a radical restructuring and downsizing at the end of that year after failing to raise. In March 2023, existing and new investors provided $13 million and co-founder and former CTO Dave Fick, who had succeeded departing CEO Mike Henry, was appointed CEO. Under Fick the company cut costs by using more development partners and off-the-shelf components and refocused its go-to-market on defense, alongside public safety, industrial and consumer verticals, while preparing a next-generation chip called the M2000. Taner Ozcelik, previously at Nvidia and ON Semiconductor, joined as CEO in mid-2024 and led a rebuild of the architecture, roadmap, software and strategy; he has said the first-generation implementation left on-chip RISC-V processors near full utilization, lacked on-chip DACs and used ADCs of insufficient resolution, and that the first chip was taped out before the software stack was sufficiently developed. In December 2025 the company announced a $125 million round led by DCVC.

Risks & controversies

Mythic reportedly ran out of capital in November 2022 and carried out a radical restructuring and downsizing after failing to raise, before a $13 million rescue financing in March 2023. Current CEO Taner Ozcelik has publicly described prior leadership as having made significant technical and strategic errors, including taping out the first chip without a developed software stack, and said he identified 17 issues with the company's approach and technology before taking the job. The company operates in a crowded edge AI silicon market amid tighter semiconductor venture funding, and its stated data-center performance and TCO advantages are company claims rather than independently verified results.