Current AI and digital computation are hitting a wall of diminishing returns and unsustainable power consumption, creating a critical need for a new paradigm.
Naveen Rao's new venture, Unconventional AI, has raised $475M to build a new hardware stack based on analog computing, aiming for a 50-1000x efficiency improvement to solve the energy bottleneck.
The future of computation lies in exploring unconventional, 'deep tech' approaches like new lithography, superconducting logic, and biologically-inspired architectures to move beyond the limitations of current models.
Venture firms like Playground are making high-risk, capital-intensive bets on these foundational technologies, viewing the entire compute stack as a 'jump ball' open for fundamental disruption.
12 quotes
Concerns Raised
Unsustainable power consumption of AI data centers is making large-scale build-outs physically unrealizable.
Current digital architectures and semiconductor technologies (e.g., EUV lithography) are experiencing diminishing returns.
The Transformer architecture is not the end-state for AI and is computationally inefficient.
The high capital requirements for deep tech hardware innovation present a significant barrier to entry.
Opportunities Identified
Developing new analog or neuromorphic hardware to achieve orders-of-magnitude (50-1000x) improvements in energy efficiency.
Creating novel lithography techniques to 'restart' Moore's Law and overcome the limitations of current systems.
Building new computational paradigms, such as superconducting logic, to unlock massive performance gains.
Disrupting the entire AI stack, from materials to models, which is currently a 'jump ball'.