uber burned a years worth of ai budget in four months

this story made me sit up.

a major tech company allocated their entire annual ai budget at the start of the year. a real budget. nine figures.

they burned through it in four months.

not because they were wasteful. because the workload exploded faster than the budget could expand to meet it.

every product team had a use case. every customer flow had an ai opportunity. every micro decision in the product became a model call. and model calls add up at thirty cents per inference.

thirty cents.

doesnt sound like much until you realize ai-native products make millions of these calls per day. tens of millions at scale.

this is the wall that almost nobody is talking about. the wall isnt model performance. its not infrastructure. its not training data.

its the cost of every individual transaction.

traditional payment rails cost too much per call to power an ai-native economy. visa cant settle a third of a cent. stripe cant settle a millisecond api ping. paypal isnt designed for an autonomous agent buying compute by the second.

so a new layer is being built right now. micro settlement infrastructure. sub penny payments. machine to machine commerce. agents transacting with other agents at the speed and price points that humans cant match.

and heres where it gets interesting.

this layer doesnt run on traditional rails. it runs on stablecoins and increasingly on bitcoin lightning.

lightning was built for payments humans dont care about. coffee. content tips. small in-app purchases. the use case never quite materialized at scale.

but what if the user isnt a human.

what if the user is a software agent making a hundred thousand decisions a day, each requiring a tiny payment to access an api or a model or a piece of data.

lightning becomes obvious infrastructure for that.

and the elegance is that bitcoin doesnt need to win the human payments war to win the agent payments war. agents dont have brand loyalty. agents dont care about onboarding ux. agents care about settlement finality, low fees, and global availability.

those are bitcoins three best features.

this is why the contrarian bet on bitcoin is shifting from store of value to settlement layer for ai. the people who hold this view are quiet about it because the narrative still belongs to the digital gold story.

but the actual demand for fast cheap final settlement is about to ten x.

and the source isnt humans.

so heres a question worth chewing on.

when the ai agents start transacting on their own.... whose rails are they running on?