Unconventional AI raises $475M to build brain-inspired chips that solve AI's energy crisis

Unconventional AI Secures Record-Breaking Seed Funding

Unconventional AI emerged from stealth with $475 million seed funding at a $4.5 billion valuation, marking one of the largest seed rounds in tech history 1

1

. Lightspeed and Andreessen Horowitz led the investment, with participation from Sequoia Capital, Lux Capital, DCVC, Future Ventures, and Jeff Bezos 2

2

. Naveen Rao, the company's founder and former Databricks VP, also invested $10 million of his own money, likely derived from his $1.3 billion sale of MosaicML to Databricks in 2023 3

3

.

Tackling AI Scaling Energy Constraints

The San Francisco-based startup is addressing what it describes as an imminent energy bottleneck in AI development. According to the company's analysis, if current projections hold, computation will become constrained by global energy supply within the next 3-4 years 3

3

. Rao explained the fundamental challenge: "AI has exponential demand but is limited by (linear) energy build-outs. We can't scale beyond a certain number of inferences per unit time because of the energy problem. We can't produce that much more energy in the next 10 years" 1

1

. The startup's goal is achieving biology-scale efficiency within 20 years, drawing inspiration from how the human brain operates on just 20 watts of power 3

3

.

Biology-Inspired Computing Approach

Unconventional AI is developing a novel hardware and software system that takes cues from biological intelligence. The company's approach challenges conventional AI chip design by arguing that neural networks today run on deterministic abstractions layered on analog circuitry, creating inefficiencies 3

3

. Rao emphasized that "natural learning systems never used numerics. They didn't simulate the dynamics of learning. They use the intrinsic physics of whatever substrate they're on to build a learning system" 1

1

. The startup plans to create a software layer that directly taps into the physical properties of silicon circuits 3

3

.

Beyond Neuromorphic Computing

While Unconventional AI draws inspiration from neuromorphic computing, Rao clarifies the company isn't strictly bound to replicating brain function. "The problem with neuromorphics is it has to work like the brain. But, why does it have to work like the brain," he stated 1

1

. Instead, the lab is exploring several different approaches to improving AI energy efficiency. The company's technology will likely involve analog AI chips rather than digital devices, leveraging nonlinear dynamics of circuits 1

1

. Intel and IBM have pursued similar concepts for years, but progress has been slow, with only a handful of working prototypes built 1

1

.

Analog Computing for Machine Learning

The startup's approach centers on the idea that machine learning workloads don't necessarily require deterministic compute platforms. Rao envisions scenarios where a combination of non-deterministic analog computing and deterministic digital logic accelerate different aspects of AI tasks 1

1

. According to him, certain models like diffusion models, flow models, and energy-based models are particularly amenable to the kinds of non-linear dynamics Unconventional is targeting 1

1

. A job posting reveals the company hopes to build AI hardware accelerators with 1,000 times better efficiency than current silicon 4

4

.

Founder's Track Record and Vision

Naveen Rao brings extensive experience in AI hardware and neuroscience to Unconventional AI. He previously founded Nervana Systems, which Intel acquired for more than $400 million, and MosaicML, which Databricks bought for $1.3 billion 4

4

. Rao studied electrical engineering at Stanford and earned a PhD in neuroscience at Brown University 1

1

. He told Axios in October that he doesn't plan to sell Unconventional AI, signaling a long-term commitment to solving AI's energy problem 2

2

. Databricks, his former employer, also invested in the seed round, and Rao maintains an advisory role there 5

5

.

Long-Term Research Focus

Rao is transparent about the timeline for Unconventional AI's ambitions. "We're not going to have a product in two years," he stated, describing the effort as "largely a research effort for the next several years" 1

1

. However, the company plans to share findings along the way, potentially as soon as next year. "This is not something that we go off in a lab for four years and emerge with the solution. Over the next several months, we're going to start releasing things," Rao explained 1

1

. His long-term aspiration extends beyond research to building a full systems company that co-designs hardware and AI models 4

4

.

Technical Architecture and Hiring

Job postings reveal that Unconventional AI's chip will be based on a system-on-chip (SoC) design that places several different types of compute modules on a single die 4

4

. The SoC will include a central processing unit for preliminary tasks like organizing raw sensory data, along with circuits optimized for linear algebra calculations used by large language models 4

4

. The company is also exploring emerging non-volatile memory technologies such as RRAM 4

4

. Unconventional AI is now hiring across hardware, software, and algorithm design roles, seeking talent capable of challenging assumptions and reasoning from first principles 3

3

.