OpenAI's Sarah Friar publishes essay on the economics of cheaper AI
Published alongside the first Jalapeño chip benchmarks, the essay frames custom silicon as one layer in a stack whose combined gains OpenAI says compound over time.
- Ideas & essays
- Compute & infrastructure
- Minor
OpenAI chief financial officer Sarah Friar published an essay arguing that the falling cost of a given amount of useful AI work comes from gains compounding across the whole technology stack — chips, data centres, model architecture, software and the products built on top of them — rather than from any single breakthrough. She described the layers as one integrated system, in which improvements at each level fund, and are funded by, improvements at the others.
The essay published the same day as OpenAI’s first disclosed benchmark results for Jalapeño, the inference chip built with Broadcom, which Friar cited as the stack argument in practice: a chip designed in-house, partly using OpenAI’s own models to speed the design process, meant to lower the cost of running OpenAI’s own models rather than to be sold as a general-purpose product.
The essay was one of several OpenAI published through 2026 under an “abundant intelligence” framing that positions falling unit costs as the rationale for its continued infrastructure spending, at a time when the scale of that spending had drawn sustained scrutiny from investors and commentators.
In the commentary
What people were saying around this time — external links, from the record's commentary rail.