Scuttlebutt

Scuttlebutt

Robotics will be 100x bigger than coding agents, says Unconventional AI CEO

13 calls: why US spare power runs out within a few years, why pure software investors are in trouble, and why inference becomes a $10 trillion market within a decade

Scuttlebutt's avatar
Scuttlebutt
Oct 25, 2026
∙ Paid
Robotics will be 100x bigger than coding agents, says Unconventional AI CEO

Naveen Rao, founder and CEO of Unconventional AI, told Harry Stebbings on 20VC that the inference market will be worth $10 trillion within 10 years.

Most AI investors are betting on software, from SaaS tools to coding agents. Naveen said coding agents are overrated as a market, robotics will be 100 times larger, and the margin is moving to the chips that turn energy into intelligence.

"I actually think the hardware is the place to be over the next 10 years."

Naveen has sold two AI companies. He founded Nervana Systems, which Intel bought, and co-founded MosaicML, which Databricks bought for $1.3 billion. Unconventional AI, his third company, builds AI hardware around energy efficiency.

The full interview is covered here so you can skip it. 73 minutes of audio, 10 minutes of reading.

Here are the 13 calls that matter.

The Twenty Minute VC (20VC) · Oct 24, 2026 · 1h 13m · 63m saved

With Naveen Rao, Founder and CEO at Unconventional AI


This Scuttlebutt Summary is available in full to premium subscribers. If you're not one already, please consider subscribing to keep reading.


Key Takeaways

  • A gigawatt of new power costs roughly $50 billion, more than a semiconductor fab

    • The US has been using spare grid capacity, which Naveen expects to run out within a few years

  • Chip efficiency now comes from smaller data types, from 32-bit down to 4-bit, not from denser transistors

  • AI moved the bottleneck from producing work to verifying it, and agents are not yet trustworthy enough to replace human judgment

  • Pure software investors are in trouble because SaaS will no longer be where margin is made

  • Coding agents are overrated as a market because the value goes to the model provider, not the harness or the IDE

  • Robotics will be 100 times larger than coding agents, and inference a $10 trillion market within 10 years

  • About 80% of AI tokens will run on open models within three years, with only 20% going to frontier models

1. Spare Power Runs Out Soon

Naveen opened on energy, which he called the real limit on AI. Building a gigawatt of power capacity costs roughly $50 billion in his estimate, more than a semiconductor fab. Most of the buildout so far has used spare capacity on the existing grid rather than new plants.

The US grid's cushion will be used up within a few years at current AI growth rates

"I think that excess capacity is going to start getting very scarce... we're going to be out of energy pretty soon, in the next few years." — Naveen Rao

Naveen put average US grid output at roughly 900 gigawatts, against peak capacity near 1.2 to 1.3 terawatts. That gap is the cushion the industry has been drawing on. He added that China passed the US in total power capacity around 2023 by building about six times faster, largely through central planning.

Even 10 new gigawatts in a year would be an achievement, at about half a trillion dollars

"Maybe we build 10 gigawatts [this year]. If we build 10, that's amazing. That's half a trillion dollars, just to put it into perspective." — Naveen Rao

Some forecasts call for 40 gigawatts a year. Naveen treated 10 as the realistic ceiling, which puts the physical buildout well short of the most aggressive plans.

2. A Blip, Not a Bubble

User's avatar

Continue reading this post for free, courtesy of Scuttlebutt.

Or purchase a paid subscription.
© 2026 The Golden Age LLC · Privacy ∙ Terms ∙ Collection notice
Start your SubstackGet the app
Substack is the home for great culture