Why Mythic's Shocking $125M Fundraise Could Change Tech Forever—Don't Miss Out!

Chip startup Mythic Inc. has made headlines today by announcing a substantial $125 million funding round led by DCVC. The venture capital firm was accompanied by notable investors including NEA, Softbank KR, Honda Motor Co., and several others, bringing Mythic’s total outside funding to over $175 million.

In a world dominated by standard processors that represent information as a series of discrete electrical signals—known as digital data—Mythic has taken a revolutionary step forward with its development of an analog processing unit (APU). This innovative technology diverges from traditional methods by using fluctuations in a single, continuous electrical signal to represent data, rather than relying on multiple individual signals.

While analog chips are typically utilized for simpler tasks like distributing power to components or filtering Wi-Fi interference, Mythic's APU is designed specifically for running artificial intelligence (AI) models. The company boasts that its chip can deliver performance that is 100 times more efficient per watt compared to conventional graphics cards, a claim that has significant implications for power consumption in AI applications.

Central to Mythic's APU is a design known as compute-in-memory architecture. This cutting-edge chip integrates memory circuits that perform dual functions: storing and processing information. The processing is executed via resistors, which inhibit the flow of electrons and facilitate calculations within the chip.

AI models consist of artificial neurons grouped into layers that carry out various calculations to analyze prompts and pass outputs to subsequent layers. A distinctive feature of Mythic's APU is its ability to run these layers in parallel, drastically speeding up processing times. This could revolutionize how complex computations are handled in real-time applications.

Mythic is poised to transform its technology into a commercial offering with a device called Starlight. This device houses multiple APUs, collectively consuming less than one watt of power, making it suitable for embedding in edge systems like robots. This feature promises to enhance the quality of data collected by image sensors within these systems.

Looking ahead, Mythic envisions its silicon being used in data centers as well. Prior to the funding announcement, the company tested its APU's capability to operate large language models (LLMs) and found that APU-powered servers could process up to 750 times more tokens per second than traditional graphics cards. This stark comparison underscores the potential of Mythic's technology in handling intensive computational tasks.

To facilitate the integration of its chips, Mythic provides a software toolkit that enables developers to adapt their LLMs for its architecture. The toolkit employs quantization—a method that compresses neural network parameters to minimize their memory footprint—while also enhancing performance through retraining specifically for the APU.

Additionally, Mythic has disclosed that it is working on a new version of the chip that aims to empower low-power devices like smartphones to run LLMs comparable to GPT-3. This advancement could usher in a new era of efficient, high-performance AI applications across a range of consumer devices.

This funding round not only highlights the growing interest in innovative AI hardware but also signals a significant shift toward more energy-efficient solutions in a market that increasingly demands sustainable technology. As Mythic continues to push boundaries with its unique approach to AI processing, the implications for industries reliant on AI are profound—ranging from robotics to data analysis, potentially reshaping the landscape of artificial intelligence as we know it.

As we watch Mythic Inc. navigate this promising trajectory, the tech community will undoubtedly keep a close eye on how its advancements play out in practical applications, particularly in enhancing AI performance while minimizing energy consumption.