Naveen Rao's new venture, Unconventional AI, is developing analog computing hardware to address the unsustainable energy consumption of current digital AI systems.
The company's core thesis is that by using the physics of a physical substrate, analog chips can be orders of magnitude more energy-efficient than the matrix multiplication-based digital approach dominated by GPUs.
Unconventional AI is taking a long-term, research-focused approach, aiming to discover a new computing paradigm within five years, rather than just building an incremental chip improvement.
The massive growth in AI is creating an energy crisis, with U.S.
data centers already consuming 4% of the grid's output, creating a strong market incentive for radically new, efficient hardware.
12 quotes
Concerns Raised
The massive and growing energy consumption of AI data centers is unsustainable and threatens to bottleneck industry growth.
The existing US power grid infrastructure is insufficient to meet the projected energy demands of AI over the next decade.
Scaling analog computing has historically been a major technical challenge due to manufacturing variability.
Opportunities Identified
Developing a highly energy-efficient analog computing substrate could fundamentally disrupt the AI hardware market.
A new computing paradigm based on dynamic systems may enable breakthroughs in AI, potentially leading to models with a better understanding of causality.
The clear and present energy crisis creates a strong market pull for radical hardware innovation.