Flourish Inc., a startup developing artificial intelligence models inspired by the human brain, has raised $500 million in funding at a $2.5 billion valuation.

Wired reported today that about a fifth of the capital was provided by Amazon.com Inc. founder Jeff Bezos. The rest came from a consortium that included Alphabet Inc.’s GV startup investment arm, Lux Capital and healthcare-focused fund Catalio.

Flourish is looking to develop AI models that require significantly less power than large language models. According to the company, a server-grade graphics card uses about 30 times more energy to process information than the human brain. Flourish hopes to reduce that power draw by more than an order of magnitude.

The company was founded last year by former Amazon executive Rob Williams and Thomas Reardon, a prominent neuroscientist. Reardon played a key role in the development of Internet Explorer. He later co-founded a startup called CTRL-Labs that Meta Platforms Inc. bought in 2019. CTRL-Labs’ technology underpins the Meta Neural Band, a wristband that enables users to control smart glasses with hand gestures.

Flourish hopes to discover ways of building more power-efficient AI models with the help of an in-house neuroscience lab. According to Wired, one of the facility’s focus areas will be studying cortical columns. Those are neuron structures that are believed to play a key role in information processing.

The startup plans to equip its lab with electron microscopes. Such instruments, which can cost millions of dollars apiece, are capable of observing significantly smaller structures than a standard optical microscope.

A microscope’s resolution depends on the wavelength of the light that it shines on the object being studied. The shorter the wavelength, the better the resolution. Electron microscopes replace light beams with electrons, which have a wavelength many orders of magnitude shorter than photons.

Life science researchers use three electron microscope varieties. The first type, which is known as a SEM instrument, collects data about objects by studying how they reflect electrons. Scientists also use TEM microscopes, which generate a high-energy electron beam that passes through objects instead of getting reflected by them. The third popular electron microscope variety combines TEM and SEM techniques.

Flourish plans to carry out a broad range of experiments to identify a path toward more energy-efficient AI. The company is reportedly expecting a breakthrough within five years. In the more immediate future, it plans to make a “model that can learn continuously” available on consumer devices.

It’s reportedly in talks with an unnamed chipmaker to ship a processor that can run its model. Additionally, the company is developing an AI memory management system that will reduce the amount of data needed for model training.

Flourish is one of several companies pursuing new AI architectures. Startup Cartesia AI Inc. is building so-called state space models, which process data using mathematical methods originally developed for space navigation systems. Meta, in turn, has developed an AI architecture called JEPA that is specifically optimized to power robots. 

Image: Unsplash