Capping a dramatic turnaround, Mythic has raised $125M in an oversubscribed funding round led by DCVC to solve the biggest problem of AI—its insatiable, ruinous energy consumption—at both the data center and the edge, with an architecture that is 100x more energy-efficient than today’s top-of-the-line GPUs and all competing AI ASICs.
Mythic has the only semiconductor architecture that approaches the energy efficiency of the human brain—and the company can serve all parts of the AI market. Its analog AI chiplets support AI models in the robotics, automotive, and defense industries, as well as 1T+ parameter Large Language Models (LLM) in data centers. In addition to offering massive energy savings, Mythic’s chiplets can dramatically lower cost and latency, while achieving massive throughput advances. With the world’s first analog computing silicon in production, validated by DoD, major auto OEMs, and major defense partners, Mythic gives the United States an enormous advantage in the global competition for AI leadership, because victory will no longer come down to which country can produce, at disastrous economic and environmental cost, the greatest amount of electricity soonest.
Read More – Culture Biosciences Closes Series C Funding
The new funding follows a period of intense reinvention for Mythic. Under new leadership, the company has completely rebuilt its architecture, roadmap, software, and strategy. That progress convinced a consortium of investors led by deep-tech venture-capital firm DCVC (which has backed Mythic since 2016) and joined by NEA, Atreides, Future Ventures, Softbank KR (SBVA), S3 Ventures, Linse Capital, One Madison Group, Catapult, UDC, and many others. These top venture funds were joined by Honda Motors and Lockheed Martin, each ranked in the top 10 in the world in their respective industries, automotive and defense, turbo-charging Mythic in two vitally important, trillion-dollar markets that are being transformed by AI.
Mythic APUs to augment GPUs, solving the AI energy crisis
Massive energy consumption has become an existential problem for GPUs in AI: by the end of the decade, one-tenth of the electricity produced by the US power grid is expected to be consumed by data centers running AI workloads powered by GPUs. While GPUs fueled the rise of generative AI, they now face a power wall, because they are the relic of a 1945-era chip design architecture, called Von Neumann, in which memory and compute are physically imagined, architected, and built separately. This divide forces modern chips to move data back and forth between the memory and the processor, consuming three orders of magnitude more energy than necessary, which results in 90% of the energy used in AI being wasted. Incumbent GPU-based AI accelerators attempt to patch this with High Bandwidth Memory (HBM), an expensive and marginal improvement that cannot prevent an imminent crash of the current AI systems against the power wall.
Mythic’s computer architecture does not delineate between the processor and the memory, instead treating them as one and the same, much like they are in the human brain. What is more, Mythic’s APUs perform the most intense part of the AI workloads, i.e., matrix multiplications, in analog— much as a human brain operates. The result is unprecedented levels of energy efficiency in AI—yielding 120 trillion operations per second (TOPS) per watt in Mythic’s current APU architecture, which is 100x better than today’s top-of-the-line GPUs (including memory transfers), while yielding accuracy levels higher than those of Von Neumann architectures operating in higher precision.
Mythic has essentially found a way to scale the human brain. The company unlocks AI that is as impressive for its parsimonious use of energy as it is for its exceptional powers of inference. Mythic’s APUs are the closest thing to the human brain in silicon in energy consumption, where a single multiply accumulate (MAC) math operation, i.e. a giant matrix multiplication, which comprises 95% of today’s AI math workloads, consumes only 17 femto joules. That is 1,000x more energy-efficient than the same operation performed on today’s GPUs.
Led by an ex-NVIDIA executive, Taner Ozcelik, who founded NVIDIA’s automotive business and grew it consistently over a decade, Mythic intends to go after four trillion-dollar industries: data centers, automotive, robotics, and defense, with a singular strategy of being the unambiguous leader in AI in performance per watt, and 100% R&D leverage, much like NVIDIA does for its GPUs.
Read More – Radiant Raises Over $300M in Funding
