Dec 8, 2025 · 30m · a16z
The Chip That Could Unlock AGI.
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of The a16z Show, Unconventional AI co-founder and CEO Naveen Rao discusses how shifting from traditional digital processors to physics-based analog computing substrates can solve AI's looming energy crisis and pave the way toward artificial general intelligence.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. How this is scored →
The host as informed peer 3.9 Guest teaching 4.7 Guest disagreement 0.9 The host pushing back 0.9
speaking balance: gold is the host, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 13:00 Rejecting digital vs analog binary
Naveen reframes the host's premise directly, stating that computing is not a strict digital-versus-analog binary but rather depends on workload dynamic systems.
Hardest push from the host ▶ 12:23 Highlighting transmission grid limitsMatt pushes back on the optimism around adding 400 gigawatts of AI power capacity by highlighting that the legacy 1970s transmission grid cannot support such load.
Biggest teaching moment ▶ 8:40 Explaining physical substrate isomorphismNaveen educates the host on why using deterministic, high-precision digital silicon substrates for inherently stochastic neural networks represents a core architectural mismatch.
The host holds their own ▶ 12:23 Citing energy infrastructure bottlenecksMatt displays specific hardware domain knowledge by identifying physical grid transmission limitations rather than just power generation deficits.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The host as informed peer | Guest teaching | Guest disagreement | The host pushing back | Why |
|---|---|---|---|---|---|---|
| Naveen Rao's Background in Hardware and Neuroscience | 3 | 4 | 1 | 1 | Matt asks inquisitive questions about Naveen's career spanning hardware and software across Nirvana and Mosaic. Naveen gently reframes what 'full-stack' used to mean versus today's definition. The tone is collaborative and conversational throughout. | |
| Digital vs. Analog Computing Paradigm History | 2 | 6 | 1 | 0 | Matt prompts Naveen to explain the fundamental differences between digital and analog paradigms. Naveen educates the audience and host on the historical trade-offs, precision errors, and vacuum tube scaling up to modern GPUs. | |
| Using Physical Substrates to Subserve Intelligence | 4 | 6 | 1 | 1 | Matt synthesizes the idea that computers historically complemented human precision, whereas modern AI introduces intentional fuzziness. Naveen explains why stochastic neural dynamics are poorly matched to deterministic silicon substrates. | |
| Data Center Power Demands and Infrastructure Challenges | 5 | 4 | 1 | 2 | Matt demonstrates clear technical awareness of energy infrastructure bottlenecks by pointing out that 1970s-era transmission grids will melt under AI energy demands. Naveen agrees and provides global data center energy capacity statistics. | |
| Mapping Transformers and Diffusion Models to Physics | 5 | 5 | 1 | 1 | Matt cites Yann LeCun's work on energy-based models and probes whether alternative dynamic paradigms bring us closer to AGI. Naveen explains differential equations and causality in physical systems. | |
| Industry Dynamics: TSMC, Google, and NVIDIA Relationships | 4 | 4 | 1 | 1 | Matt asks Naveen to position Unconventional relative to tech giants and specifically clarifies Google's TPU strategy. Naveen explains TSMC partnership plans and potential NVIDIA collaboration. | |
| Engineering Pragmatism and Embracing the 'Crazy' Vision | 4 | 4 | 0 | 0 | Matt references hardware industry humor around Cerebras and Jensen Huang's wafer reveals. Naveen elaborates on engineering pragmatism, hiring needs, and organizational agency. |
Statements from this episode (23)
Rao: Unconventional AI focuses on physics of learning, not traditional chips
“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 works.”
Rao: The boundary between hardware and software is an arbitrary choice
“Software and hardware is, is not really a natural boundary. It's just where we decide to draw the line and say, okay, this is something I make configurable or I don't.”
Rao: Computer architecture has remained fundamentally unchanged for 80 years
“We've been building largely the same kind of computer for 80 years. We went digital back in the 19 forties”
Rao: Computing is currently constrained by energy at a global level
“We're at a point in time where computing is, is, is bound by energy at the global level, which just was never true in all of humanity.”
Rao: Early analog computers failed to scale due to manufacturing variability
“Analog computers were actually some of the first computers, and they worked really well. They were very efficient, but they couldn't be scaled up because of manufacturing variability.”
Rao: Analog computing is inherently more efficient by leveraging physical substrates
“Analog still is inherently more efficient because you, it's actually analogous computing is the way to think about it. Like, can I build a physical system that is similar to the quantity I'm trying to express or compute over? You're effectively using the physi…”
Rao: Deterministic digital silicon is ill-suited for stochastic neural networks
“A neural network is actually a stochastic machine, and so why are we using the substrate that is highly precise and deterministic? For something that's actually stochastic and distributed in nature.”
Rao: Unconventional AI can build analog circuits isomorphic to intelligence
“So we believe we can find the right isomorphism in electrical circuits that can subserve intelligence.”
Rao: Human brains run on 20W, mammalian brains on 0.1W
“Human brains run on 20 watts of energy. That's, that is true. But if you look at mammalian brains generally, actually extremely efficient, like a squirrel or a cat, it's like a 10th of a watt.”
Rao: US holds roughly 50% of global data center capacity
“The U S is about 50% of the world's data center capacity.”
Rao: Data centers consume 4% of the US energy grid
“And today we put about four percent of the energy grid of the U S energy grid into those data centers.”
Rao: AI will require 400 GW of new power capacity over 10 years
“By some estimates, we need 400 gigawatts additional capacity over the next 10 years. To power the demand for AI.”
Rao: Traditional numeric computing lacks physical time, simulating it instead
“In the real world, every physical process has time. And in the computing world, like the numeric computing world, we actually don't have that concept. You simulate time with numbers.”
Rao: Unconventional AI will release open-source tools for its analog hardware
“And we will be releasing some open source and things around this to let people play around.”
Rao: Transformers succeeded because they maximized GPU hardware efficiency
“Transformers are really they're a big innovation because they made the constructs of the GPU work extremely well.”
Rao: Dynamic substrates with physical time and causality are required for AGI
“My intuition says that anything where the basis is dynamic, which has time and causality as part of it will be a better basis than something that's not.”
Rao: Current AI models are nowhere close to AGI
“I don't think they're anywhere close to AGI because, I mean, they still make stupid errors. They're very useful tools, but they're not what, it's not like working with a person, right?”
Rao: Unconventional AI aims to build manufacturable analog chips within five years
“A couple of things that we set out to do when we built, we were starting this company was see if we can find a paradigm that's analogous to intelligence within five years. And then at the five year mark, we should be able to build something that's scalable fro…”
Rao: TSMC will partner with Unconventional AI for chip manufacturing
“TSMC is absolutely going to be a partner forward, you know, met with them recently and, you know, we want to work closely with them to make sure we get what we need, get fast turnaround times to prototype and all of that.”
Rao: Unconventional AI is building a substrate superior to matrix multiplication
“We are trying to build a better substrate than matrix multiply.”
Rao rejects AI doomerism, calling AI the next evolution of humanity
“I, I'm the opposite of an AI doomer. I think AI is the next evolution of humanity. I think it takes us to a new level, allows us to collaborate, understand each other and understand the world in much deeper ways.”
Rao: Over 40 years of academic research proves analog AI compute viability
“There's also 40 plus years of academic research, which is showing a lot of promise here. People have built different devices, albeit not in the latest technology with professional engineering teams, but they have built proofs of concept that actually show some…”