Episode Summary
Executive Summary: Naveen Rao, CEO of Unconventional AI, argues that current digital computing paradigms are fundamentally inefficient for AI workloads, which require massive energy. He advocates for analog computing that more closely mimics biological neural networks to achieve dramatic efficiency gains. Rao believes now is the critical time for this shift due to the global energy crisis driven by AI data centers, and he positions autonomous computing as essential for democratizing intelligence.
Main Topics: Digital vs. Analog Computing Paradigms: Rao explains the fundamental difference: digital computers rely on precise numerical representations and general-purpose architectures, while analog computers use physical systems to directly model computations, mimicking the efficiency of biological brains. Energy Crisis and AI Infrastructure: The discussion highlights the staggering energy demands of current AI data centers—the U.S. grid faces 4% consumption now, requiring 400 additional gigawatts over the next decade—and argues this is the primary driver for rethinking computing architecture. The Essence of Intelligence and Causality: Rao proposes that true intelligence requires understanding causality and temporal dynamics, which are inherently missing from current digital models that treat time as a numerical simulation rather than an embedded physical process. Analog Computing for AI Workloads: The episode explores how diffusion models, flow models, and energy-based models are naturally suited for analog implementation because they involve dynamical systems that map onto physical dynamics, enabling massive efficiency gains. Unconventional AI's Strategy and Challenges: Rao discusses the company's approach: starting with first-principles theory, building scalable analog chips in partnership with manufacturers like TSMC, and navigating potential collabouration or competition with incumbents like NVIDIA and Google. Team and Culture for 'Crazy' Bets: Rao emphasizes the need for interdisciplinary talent spanning theory, architecture, and analog circuit design, and advocates for giving engineers high agency to pursue radical ideas despite external skepticism.
Key Arguments: Current digital computing has remained fundamentally unchanged for 80 years and is incredibly inefficient for AI workloads, consuming vast energy while biological brains achieve similar capabilities with negligible power. Intelligence is inherently a physical, stochastic process—not a numerical, deterministic one—making analog computing a more natural substrate for building intelligent systems. The global energy crisis from AI data centers is a critical driver: the U.S. already uses 4% of its grid for data centers, and projections require 400+ additional gigawatts in the next decade, which is unsustainable without architectural innovation. Achieving true artificial general intelligence (AGI) likely requires embedding causality and temporal dynamics into the computing substrate, which analog systems can inherently provide. Analog computing is not a replacement for digital but a complement: certain fuzzier, dynamical problems (like perception and motor control) are vastly more efficient in analog, while precision tasks still benefit from digital. The key challenge is scalability—previous analog attempts failed due to manufacturing variability, but modern semiconductor fabs and advanced engineering can overcome this if the theoretical and practical groundwork is done.
Data Points: Brain energy consumption: 20 watts - Human brain runs on 20 watts; a squirrel brain on a tenth of a watt. US data center energy percentage: 4% - Current percentage of the entire U.S. power grid consumed by AI data centers. Projected additional grid capacity needed: 400 gigawatts - Estimated additional capacity required over the next decade to meet AI demand. US portion of world data center capacity: 50% - The U.S. represents about half of global data center capacity. ENIAC vacuum tubes: 18,000 - Number of vacuum tubes in the first general-purpose digital computer (1945), analogous to modern GPU counts.
Pivotal Quotes: "In a brain, the neural network dynamics are implemented physically. There is no abstraction. Intelligence is the physics." — Naveen Rao: Explaining why analog computing can be vastly more efficient than digital for AI. "Squirrel brain runs on a tenth of a watt. Our AI data centers now consume 4% of the entire U.S. power grid, and we need 400 more gigawatts in the next decade just to keep up." — Matt Bornstein (host): Summarizing the extreme energy disparity between biological intelligence and current AI systems. "I think AI is the next evolution of humanity. I think it takes us to a new level, allows us to collaborate and understand the world in much deeper ways." — Naveen Rao: Rao's vision for AI, positioning himself as an 'optimist' rather than a 'doomer'.
Implications: If successful, Unconventional AI's analog approach could drastically reduce energy consumption for AI, enabling ubiquitous, sustainable intelligence. This challenges NVIDIA's dominance and forces the industry to rethink hardware-software co-design. For listeners, it signals a potential paradigm shift in computing architecture within the next decade, with profound impacts on AI accessibility and environmental sustainability.
From the Episode
The best examples we have of intelligent systems nature are brains. And it's often been said, you know, human brains run on 20 watts of energy. That's true. But if you look at mammalian brains generally, actually, extremely efficient, like a squirrel or a cat, it's like a tenth of a lot. And so there's something there that we're still missing. And not to say that we understand all of it, but part of what I think we're missing is we have lots of abstractions in a computer that are quite lossy. In a brain, the neural network dynamics are implemented physically. So There is no abstraction. Intelligence is the physics. They're one and the same. There's no OS and some sort of API and this and that. So there's some visual stimulus, for instance, that directly activates an actual neural network and produces some somatic response, that sort of thing. Exactly. And those things are mediated by chemical diffusion and the physical properties of the neuron, the physics itself. So I think absolutely it's possible to build something that's much more efficient by using physics in an analogous way.
Are these potential allies for unconventional? Are these competitors? So I think TSMC is absolutely going to be a partner. Google kind of has everything internally. And NVIDIA, of course, they built the platform that everyone programs on today. So are we going to be at odds with NVIDIA going forward? I don't know. We'll see what the world looks like, but there could be a world where we collaborate. Has anyone called you crazy yet for doing this? Oh, yeah, plenty of people. A squirrel brain runs on a tenth of a watt. Our AI data centers now consume 4% of the entire U.S. power grid, and we need 400 more gigawatts in the next decade just to keep up. Naveen Rao thinks the problem isn't power generation, it's that we've been building the wrong kind of computer for 80 years. Naveen sold his last AI chip company to Intel, and now he's back with a bet most people call crazy: analog computing, purpose-built for intelligence. In this conversation, A16Z's Matt Bornstein sits down with Naveen to discuss why. Now is the time for this unconventional bet.
I think AI is the next evolution of humanity. I think it takes us to a new level, allows us to collaborate and understand the world in much deeper ways. Naveen Rao is here, expert in AI. Naveen Rao, probably one of the smartest guys in this domain. He sees things well before anybody else sees them. You had a lot of success doing Nirvana Mosaic and Databricks. Why start a new chip company now? First off, it's not a chip company per se. Most of what we're doing is really kind of looking at first principles of how learning worked in that physical system. NVIDIA, TSMC, Google. Are these potential allies for unconventional? Are these competitors? So I think TSMC is absolutely going to be a partner. Google kind of has everything internally. And NVIDIA, of course, they built the platform that everyone programs on today. So are we going to be at odds with NVIDIA going forward? I don't know. We'll see what the world looks like, but there could be a world where we collaborate. Has anyone called you crazy yet for doing this? Oh, yeah, plenty of people.
About The a16z Podcast
The a16z Podcast discusses tech and culture trends, news, and the future – especially as ‘software eats the world’. It features industry experts, business leaders, and other interesting thinkers and voices from around the world. This podcast is produced by Andreessen Horowitz (aka “a16z”), a Silicon Valley-based venture capital firm. Multiple episodes are released every week; visit a16z.com for more details and to sign up for our newsletters and other content as well!