All-In Pod: Cerebras Has $25B in Backlog and the Open Source AI Map Is Broken - Two CEOs on What Comes Next
Cerebras disclosed a $25 billion demand backlog. Hyperscalers order chips before production finishes because demand outruns the industry’s ability to build.
Feldman’s real pitch is not price. Run a reasoning model 15x faster and 24 hours of compute becomes weeks of thinking. Cost-per-token misses this.
Both CEOs say AGI is already reached by any historical definition. The debate has moved to how fast recursive loops converge on superintelligence.
Black Forest Labs, which invented latent diffusion, is building one architecture that makes images, video, and audio today and predicts robot actions next.
Andrew Feldman brought a half-billion-dollar Cerebras wafer-scale chip to the interview. His company has $25 billion in unfilled orders for more of them.
That is not a financing story. It is a demand signal.
Then came Robin Rombach, CEO of Black Forest Labs, whose team wrote the latent diffusion paper behind every major image and video model. His thesis: the architecture that makes a photograph will make a robot move.
This breakdown is for paid subscribers. Below: how the $25B backlog converts to revenue and why cost-per-token misprices Cerebras, the open source gap handing sovereign demand to Chinese models, and whether Black Forest Labs’ one architecture reaches the robot layer. Join to get full access.
