Key Takeaways

  • Persistent consumer AI environments like Instinct and Muse burn between $20 and $50 per user each month just to run required virtual machines and compute.
  • Jason Lemkin argues that fixed-fee consumer subscriptions cannot survive this burn because venture subsidies inevitably dry up.
  • Rory O'Driscoll counters that compute costs fall 50% to 60% year over year, making temporary negative gross margins an acceptable price for capturing distribution.
  • Waiting for gross margins to look clean before launch allows competitors backed by aggressive venture capital to capture mindshare 12 months earlier.

The Disagreement

Building persistent AI agents creates an immediate math problem. If your agent runs continuous background tasks or holds an open virtual machine, server costs add up every minute the user stays active.

Lemkin warns that founders are walking into a trap by charging standard software subscription rates for products with variable infrastructure costs. “I think Instinct could cost as much as 20 bucks a month per user to service Instinct the way it's supposed to work today,” (play from 20:00) Lemkin points out. When a customer pays $20 a month and consumes $20 to $50 in raw compute, the company loses money on its best users. Lemkin sees this as an artificial market propped up by capital: “We've also seen this in AI. The subsidies just can't last. It's too expensive, right?” (play from 21:55)

O'Driscoll takes the opposite view, framing high initial server bills as a classic tech distribution play. “The bet you take over and over again is digital goods get, you know, reduce in price 50 60% year on year,” (play from 21:32) O'Driscoll says. “So if people like the product and they get hooked on it over two or three years, the math gets better.” (play from 21:40)

To O'Driscoll, trying to protect gross margins on day one guarantees defeat. “When things are moving so quickly you can't wait until the economics work perfectly to launch a product cuz someone else armed with venture dollars will launch that product 12 months earlier when the economics don't quite work, grab all the mind share, and then they'll basically lean into the cost curve and hope that the improvement economics bails them out.” (play from 23:40) He argues that in aggressive venture environments, “almost everything that's new and digital will look crazy for a while in terms of economics.” (play from 24:40)

Who's Right (and When They're Wrong)

Both investors describe real dynamics, but they apply to entirely different balance sheets.

O'Driscoll is right if you have secured tens of millions in venture capital and your product creates genuine lock-in. If you build high switching costs, letting users consume $30 of compute on a $20 subscription lets you buy distribution before incumbents wake up. In two years, that same compute profile will cost $5 to $7.50 if the 50% to 60% deflationary curve holds. You survive the negative margins, keep the users, and collect the margin expansion later.

O'Driscoll is wrong when products lack retention. Cheap compute down the road cannot rescue a business if users churn after forty days. If your agent is a toy, you subsidize curiosity, not habit.

Lemkin is right for every founder without a massive venture war chest. If you cannot afford to burn $30 per active seat every month while waiting for hardware makers to reduce inference prices, negative margins will kill your company before the cost curve helps you. Uncapped usage plans on flat subscriptions have wiped out dozens of consumer wrapper apps for this exact reason.

What to Do With This

Audit your top 10% most active users this week. Calculate their exact inference and server spend against the subscription revenue they generate. If servicing your power users costs more than what they pay you, implement usage caps, introduce credit tiers, or gate continuous background tasks behind consumption-based billing immediately.