Invest Like the Best with Patrick O'Shaughnessy

Eric Vishria - A Decade of Lessons Investing in Software & Hardware - [Invest Like the Best, EP.486]

1h 7min

Eric Vishria of Benchmark discusses the competitive frontier in AI, comparing it to the cloud era. Key insights include how AI model inference is surprisingly hard and differentiated, why enterprises are adopting AI faster than cloud, and how product development must invert to leverage the jagged edge of model capabilities. He shares lessons from investing in Cerebras (hardware) and Sierra (AI agents), and explains why high cash-on-cash returns now exist beyond early-stage investing.

Summarized by Podsumo

🎧 Listen 🎙️ Ask about this episode

✨ Key Takeaways

1

Inference is not a commodity: Running large models efficiently is extremely hard, with 5x performance differences between providers using the same hardware and models, creating sustainable differentiation.
2

Market size and oligopoly: Just like cloud, AI won't be won by one player. Expect an oligopoly of winners across infrastructure, apps, and specialized models, with unlimited demand for intelligence constrained mainly by energy.
3

Inverted product development: Success requires understanding the jagged edge of AI capabilities—what models are good/bad at—and bridging that to customer problems. Product managers can no longer be non-technical.
4

Business model innovation is key: AI enables outcome-based pricing (selling value, not just subscriptions), which could be as transformative as the SaaS subscription model itself.
5

Hardware lessons from Cerebras: Investing in hardware requires productive naivete and working with crazy, special people on unbounded opportunities. Success depends on factors outside your control, like geopolitics and supply chains.

💬 Notable Quotes

"The difference now feels profound to me because while it is a hundred percent the case that blue chip enterprise AI is not well absorbed... They want it. They want to figure it out."
"The whole notion of a quota capacity model... it's definitely not the first order thing or constraint. In a lot of these companies, you have reps doing 10 or 20, 30 million. I saw 50 recently. It turns out that's different."
"If it doesn't work, it'll be for all of the reasons that your partner said. If it does work, it will be because those reasons didn't matter."