- 03 Jan, 2026 *
Back in the day Enron abused "mark to market" accounting. They counted future profits they hoped to earn as actual profits today.
Now it feels like a new type of accounting is going on for AI hyperscalers. I call it: mark to belief. It’s valuation based on where you hope things are going, rather than where they are.
The big labs all have strong models, but it’s crowded at the top. Performance differences are small and the models are basically interchangeable. And anyway, open source models are now almost as good - sometimes better.
In an open market it’s hard to justify a premium when a free alternative exists at roughly the same standard.
The labs know this. They can see the "commodity death spiral" looming. Competitors have reached parity, open sour…
- 03 Jan, 2026 *
Back in the day Enron abused "mark to market" accounting. They counted future profits they hoped to earn as actual profits today.
Now it feels like a new type of accounting is going on for AI hyperscalers. I call it: mark to belief. It’s valuation based on where you hope things are going, rather than where they are.
The big labs all have strong models, but it’s crowded at the top. Performance differences are small and the models are basically interchangeable. And anyway, open source models are now almost as good - sometimes better.
In an open market it’s hard to justify a premium when a free alternative exists at roughly the same standard.
The labs know this. They can see the "commodity death spiral" looming. Competitors have reached parity, open source has set a ceiling on price, and "intelligence" is fast becoming a utility with thin margins.
So belief becomes the differentiator. Belief that this lab will be the one to crack AGI. That belief is what supports the valuation and keeps the music playing.
But as I’ve written about before, the product wrapper is the "known known." It’s where investment has a much clearer and more immediate return. If the model itself becomes commoditised, then the entire value proposition collapses down to the wrapper*. And wrappers are infinitely easier to copy than genuine breakthroughs.
Any competent startup can build a polished UI, solid integrations, enterprise features, and a sweet UX. None of this is a durable moat.
So you end up with this strange inversion: companies valued via mark to belief as if they’re on the brink of AGI, while competing day to day on features that have nothing to do with it. As was the tagline of Enron, it’s time to “ask why.”
Postscript
*When I say product wrapper, I am including hosting. I didn’t go into it here as I think smaller models with simpler hosting needs are going to increase in demand in 2026, meaning the convenience of the hyperscaler cloud matters less.
Maybe AGI is round the corner. I actually believe it’s something we should be taking seriously and preparing for, so this isn’t a commentary on the likelihood of AGI. Instead it’s simply an observation on what we’re paying for (and using) vs what we’re being told.