Keep pulling the thread on Naveen Rao.
AI inference and training currently consume many gigawatts of electrical power.
Naveen Rao predicts that the world will exhaust its available energy supply for AI within the next two to four years.
Current computing technology is approximately three orders of magnitude less energy-efficient than what is theoretically achievable with 2D lithography.
Unconventional AI plans to build its first chip prototype, based on nonlinear dynamics, in the summer of 2026.
Unconventional AI developed a full chip prototype in six months, starting from no team in January 2026, by leveraging AI in its design process.
Unconventional AI's dynamical systems architecture is a non-Von Neumann approach that eliminates the separation between state and computation by using the time axis of physics to compute.
Startups can achieve chip tape-outs in months, a process that takes years for traditional chip companies, due to having no legacy baggage.
The human brain operates on approximately 20 watts of power.
The total power consumption of all 8 billion human brains on Earth is 160 gigawatts.
The world's total electricity generation capacity is approximately 9,000 gigawatts.
The United States has approximately 1,000 gigawatts of electricity generation capacity.
The Landauer principle defines a fundamental thermodynamic limit to the energy efficiency of computation.