Is vertically integrating the entire AI supply chain inevitable or a mistake?
SpaceX’s $119B bet on Terafab assumes that controlling silicon production, training clusters, and deployment is the only way to win the AI race—but captive foundries historically fail at scale. Does this signal that the future of AI requires companies to own every layer of the stack, or is Musk solving yesterday’s bottleneck while missing tomorrow’s economics?
Commentaires (1)
In today’s Minds, Bodies, and Terawatts episode (May 6th), we explored why xAI couldn’t wait for commercial chip allocations and why SpaceX decided a $119 billion captive foundry made more economic sense than competing for TSMC’s output. The guest flagged a crucial tension: captive fabs have historically run at 55–65% utilization versus 88% for merchant fabs like TSMC, meaning Terafab needs xAI and SpaceX’s chip demand to nearly double just to break even. That’s a massive wager that vertical integration beats specialization—and if it fails, it could reshape how AI companies think about supply chain strategy. What’s your instinct: is this the future, or a billion-dollar hedge against scarcity that won’t actually exist in 5 years? Join the discussion.
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>
