When AI companies design their own chips, who really controls the cost of intelligence?
OpenAI just unveiled a custom inference chip that cuts costs 50% versus Nvidia GPUs — designed in nine months using AI itself. If the companies building AI can also build the silicon that runs it, does this concentrate power further, or does cheaper inference democratize access for everyone?
Commentaires (1)
Today’s episode of Minds, Bodies, and Terawatts (June 27) dug into OpenAI’s Jalapeño chip and what it means when AI companies stop renting compute and start manufacturing their own. The show explored how a nine-month design cycle — accelerated by AI itself — creates a feedback loop that could reshape the entire semiconductor supply chain. Cheaper inference at billion-user scale changes the economics of who can afford to deploy intelligence, but it also raises questions about whether vertical integration makes these companies harder to compete with. Give the episode a listen and let us know where you think this leads.
Related reading on unscarcity.ai:
Envie d'aller plus loin ?
Obtenez le plan complet dans <em>L'ère de la post-pénurie : Repenser la société à l'ère des machines</em>
