Keep pulling the thread on Naveen Rao.
Data centers in the United States consume approximately 4% of the U.S. energy grid's output.
Naveen Rao predicts that increased energy demand from data centers led to power brownouts in the Southwestern United States during the summer of 2025.
Unconventional AI's core technological goal is to create a computing substrate superior to the matrix multiplication-based approach used in current AI hardware.
Unconventional AI's first prototype chip is expected to be one of the largest, or potentially the largest, analog chips ever constructed.
Unconventional AI plans to partner with TSMC for manufacturing its chips.
The United States holds approximately 50% of the world's data center capacity.
Unconventional AI's initial hardware will be designed to support current state-of-the-art models, including Transformers and diffusion models.
Unconventional AI plans to release open-source software related to its analog computing approach.
Unconventional AI has a goal to discover a computing paradigm analogous to intelligence within a five-year timeframe.
Unconventional AI aims to have a manufacturable and scalable product design by its five-year mark.
Naveen Rao believes Google's internal hardware development strategy for products like TPUs focuses on lower-risk, continual improvements rather than high-risk architectural changes.
Unconventional AI's analog chips are expected to be highly energy-efficient, operating at low temperatures and not requiring large heat sinks.