Worth adding a data point from outside the US conversation. When Jensen Huang announced the Korea Blackwell allocation in October 2025, the headline number was 260,000 GPUs, but the distribution is what matters for your thesis. Roughly half went directly to manufacturing conglomerates like Samsung, SK, and Hyundai for fab process optimization, battery yield prediction, and production line digital twins. The remainder split between cloud providers and government. No equivalent industrial AI allocation has been announced for Japan or Germany at that scale.
The implication is that the “AI-native manufacturer” future you are describing is already the present tense for Korea’s top-tier chaebols. Hyundai’s Singapore Innovation Center has been running this operating model since 2023, and Samsung and SK Hynix have had MES-integrated process optimization feedback loops for years. The Blackwell allocation is the next layer on top of an existing stack, not the beginning of one.
The twist is that Korea’s version does not look like the new-entrant story. It is the opposite. Vertical integration went to the extreme decades ago, and the top ten manufacturers already do something close to Zane’s “2% doing 30-50% of the work.” But that concentration is now the source of structural rigidity, not agility. The small suppliers around the chaebols are slower to adopt AI than US mid-market manufacturers, because the scheduling, pricing, and specs all flow down from one large customer. The learning loop stays inside the integrator and rarely diffuses out to the ecosystem.
Which may suggest that the wedge question matters even more than the piece argues. Korea shows what happens when the wedge is already fully claimed.
Annelies, AI native factories are med/long term customer profile for my pre-seed startup. It seems like you know a ton about this space. Is there a way we could talk more about this? I’d like to align my vision with the one laid out in your article.
Worth adding a data point from outside the US conversation. When Jensen Huang announced the Korea Blackwell allocation in October 2025, the headline number was 260,000 GPUs, but the distribution is what matters for your thesis. Roughly half went directly to manufacturing conglomerates like Samsung, SK, and Hyundai for fab process optimization, battery yield prediction, and production line digital twins. The remainder split between cloud providers and government. No equivalent industrial AI allocation has been announced for Japan or Germany at that scale.
The implication is that the “AI-native manufacturer” future you are describing is already the present tense for Korea’s top-tier chaebols. Hyundai’s Singapore Innovation Center has been running this operating model since 2023, and Samsung and SK Hynix have had MES-integrated process optimization feedback loops for years. The Blackwell allocation is the next layer on top of an existing stack, not the beginning of one.
The twist is that Korea’s version does not look like the new-entrant story. It is the opposite. Vertical integration went to the extreme decades ago, and the top ten manufacturers already do something close to Zane’s “2% doing 30-50% of the work.” But that concentration is now the source of structural rigidity, not agility. The small suppliers around the chaebols are slower to adopt AI than US mid-market manufacturers, because the scheduling, pricing, and specs all flow down from one large customer. The learning loop stays inside the integrator and rarely diffuses out to the ecosystem.
Which may suggest that the wedge question matters even more than the piece argues. Korea shows what happens when the wedge is already fully claimed.
Fab read! 💪🏽
Annelies, AI native factories are med/long term customer profile for my pre-seed startup. It seems like you know a ton about this space. Is there a way we could talk more about this? I’d like to align my vision with the one laid out in your article.