March 31, 2025

What is "good" AI revenue?

"Only when the tide goes out do you discover who's been swimming naked."
—Buffett

There have been recent reports of highly capitalized AI startups earning questionable revenue:

  • 11x, an outbound SDR agent platform, has allegedly been counting revenue from cancelled contracts and claiming customers they don't have (TechCrunch).
  • Builder.ai, a no-code software platform, revised down its 2H24 revenue by 25% and hired auditors (Bloomberg).
  • There are even some questions about Perplexity coming from its users, though their CEO has attempted to debunk those theories (TechCrunch).

"Everyone wants their boat to float on the AI river," to quote a recent Books with Legs podcast.

That said, we also saw two incredible fundraises today proving that AI startups are setting records for revenue growth:

  • OpenAI raised $40B at a $300B valuation, expecting revenue to rise to $29.4B next year (Axios).
  • Cursor raised $625M at a $9.6B valuation, having just hit $200M in ARR, mere months after setting the record for fastest startup to achieve $100M ARR (Twitter).

Recently, we asked whether Cursor is a good business. The answer is probably yes. While there are some reasons to be bearish – the switching costs among AI products are low – it looks like there is a huge brand moat for the companies that become the clear leader in their category. ChatGPT is a household name; Cursor is the ChatGPT for software engineering.

The nuance is that you need to be out of strike range of ChatGPT. Back in 2022, Jasper was all the rage because it was emerging as the first category leader for "AI for marketing." But its business apparently dried up after ChatGPT launched, effectively supplanting Jasper's basic text-generation capabilities.

I suspect something similar is happening with Perplexity. Perplexity was the first category leader for "AI for search," but ChatGPT and Claude have both launched their own web search capabilities. Anecdotally, I completely stopped using Perplexity and switched to ChatGPT's search feature after it launched.

Perplexity is now working on their own browser, training their own foundational model (Sonar), and the CEO is tweeting about funding the creation of a new Mahabharata. These suggest desperation and would worry me if I were an investor.