Are Software Investors “Done For”? Naveen Rao’s Seven Contrarian Calls

Harry Stebbings talks with Naveen Rao in the latest episode of the 20VC podcast. Naveen is the founder of Nervana Systems (sold to Intel for $400 million), co-founder of MosaicML (sold to Databricks for $1.3 billion), and now runs Unconventional AI, which raised a $450 million seed round at a $4.5 billion valuation. That résumé gives him the credibility to say some uncomfortable things.

Naveen discussed seven topics on the show.

1. Software investors are “done for”

“Pure software investors are kind of done for.” SaaS is no longer where the profit is. Since 2010, cloud services have let founders launch products quickly and chase $100 million–$200 million markets. As applications shift to the intelligence layer, software won’t be a huge market within 20 years.

2. Agentic coding is overhyped

The value is with model providers. Harnesses and IDEs aren’t very sticky; people move between tools. Apart from model providers, he wouldn’t invest in this category.

3. Will companies own their own models?

“That’s a bit of technological optimism.” You have to be at a certain scale to build and own your own intelligence. Big players will customize; small companies will buy out-of-the-box solutions. Hamel Husain replied under the tweet: Agreed, fine-tuning requires a lot of expertise, and only at a certain scale can you amortize the cost.

One commenter said: “Most companies can’t even update their wiki, so their own model is even further away.” Another said: “What they really want is their own data in the prompt; that’s a retrieval problem.”

4. Where is the value in robotics?

The value of robots isn’t in the model itself. Naveen invests in robotics model companies, but expects them to eventually make end-to-end devices.

5. Most memorable investor meeting

Steve Jurvetson. In 2014 no one was investing in hardware, but he was interested because he was thinking about it himself.

6. Differences between European and American founders

American founders are allowed to think bigger. European investor networks are conservative; crazy ideas get you treated like a lunatic.

7. The inference market will be $10 trillion

AI models will be a trillion-dollar market within a few years, and inference will be $10 trillion within 10 years. Token costs fall by an order of magnitude, and the market expands more than 10x—the Jevons paradox. A commenter added: GPT-3-level models went from $60 per million tokens in 2021 to $0.06 in 2024, roughly 10x per year.

Steve Jurvetson commented under the tweet: “Naveen is a national treasure. I’ve invested in his AI companies three times. He’s a polymath spanning neuroscience to chip design, and he’s also an LMP2 race car driver.”

Naveen’s views may not all be right, but they’re worth hearing. What do you think?

Full conversation: 20VC - Inference Will Be $10TRN Market