AI token prices just got a reality check | Lattice Finance Blog

AI token prices just got a reality check thumbnail

The AI bill comes due

Anthropic dropped Claude Fable 5, its most powerful publicly available model, and the sticker shock is the plot twist: $10 per million input tokens and $50 per million output tokens. That’s double the price of Claude Opus 4.8, which is basically the AI version of your streaming bundle quietly adding a few bucks until one day you’re somehow paying for six apps and still missing the show you wanted.

Why Wall Street suddenly cares

Wells Fargo strategist Ohsung Kwon says the real risk isn’t just a wobble in AI stocks — it’s that token costs are inflating right when the market has gotten a little too comfy. His argument: labs have stopped subsidizing inference, and companies like Walmart and Uber have already warned that AI budgets can get eaten alive in months.

That matters because the AI trade has two giant pressure points:

Higher demand costs: if using AI gets pricier, companies may slow adoption or squeeze usage.
Higher build-out costs: Microsoft, Meta, and Alphabet are still pouring money into AI infrastructure, so rising CapEx plus pricier tokens is a nasty combo.

The names in the crosshairs

Kwon’s warning lands hardest on the usual suspects. Microsoft, Meta, and Alphabet are all front and center in the AI arms race, while Nvidia is the obvious “picks and shovels” name if the infrastructure spending boom cools off. In other words: if the gold rush gets less goldy, the shovel seller doesn’t exactly throw a party.

The bigger market shrug-or-panic question

Kwon argued last Friday’s selloff was more about positioning than fundamentals, which is Wall Street-speak for “everyone rushed for the exit at once.” He thinks the cleanest way to play it is hedging — puts, call selling, that whole sophisticated side-quest — rather than bailing on AI entirely.

Big picture: the AI story is still alive, but the easy-money phase may be fading. If token inflation keeps climbing, investors may need to start thinking less about the hype cycle and more about who can actually make the math work.