Will the price per AI token/compute increase significantly in the next 6-12 months, and how should firms think about that risk?

The group's view was that AI is currently underpriced relative to the value it delivers — one participant argued that if an AI company is saving a firm $100,000/year by replacing labor, they'll eventually charge closer to $80,000-$90,000 for it rather than pennies, since the current market is a land-grab for market share funded by VC money rather than sustainable pricing. This was compared to the 1998 B2B internet buildout, where companies gained share while losing money per client. One person expects a split to emerge: expensive frontier models (like Opus) for generative, open-ended work, alongside cheaper 'utility class' models (mentioning Llama and DeepSeek as examples) that are more restrained, more technical to deploy as consistent agents, but cheaper for repeatable tasks. No one claimed certainty on the timeline; one participant explicitly said 'I don't know when it is.'