Mistral's free Apache 2.0 weights still drew Samsung's €3B Series D lead and on-prem fab AI deal
The AMW Read
Reframes the already-reported Samsung-led €3B Series D as a governance purchase over free Apache weights, updating Mistral’s case study on layered sovereign enterprise AI.
Mistral's free Apache 2.0 weights still drew Samsung's €3B Series D lead and on-prem fab AI deal
Samsung Electronics led Mistral's €3 billion Series D at a valuation above €21 billion—described as Europe's largest tech equity raise—and tied the stake to a partnership in which Samsung's device solutions unit will bring models including Mistral Large into chip manufacturing as customized on-premises systems for defect detection, equipment optimization, and yield control. ASML led the prior Series C and added about €400 million in this round to hold roughly 11% economically, while the three founders keep dual-class voting control. Mistral Large 3, a 675-billion-parameter mixture-of-experts model with about 41 billion active parameters released under Apache 2.0, can be downloaded, modified, and redistributed commercially at no charge—yet Samsung still paid for equity and co-development rights.
That gap is the load-bearing fact for enterprise AI buyers. Open weights grant the right to run and tune a checkpoint; they do not supply fine-tuning on proprietary fab logs, continuous production support, a path to the next generation, or a voice when roadmap and license terms shift. Consecutive semiconductor strategics—ASML then Samsung—are buying operational capacity and governance on top of a permissive license, not the weight file. Cap-table geography (Korea, U.S., Netherlands, Luxembourg; no French holder above 10%) sits uneasily with nationality-framed "sovereign AI," and Mistral's own stack still depends on Nvidia accelerators—Large 3 was trained on 3,000 H200s—while its European compute plan (up to 1 GW by 2030; first site roughly 40–100 MW) remains partial. Sovereignty here is a scorecard across data, model, compute, auditability, and persistence, not a binary flag. Per the AI Market Watch index, Mistral's total funding is about $4.0B including $830M of debt financing as of March 2026 (coverage of ~5,000 indexed companies).
For builders and investors, start with continuity questions, not benchmark tables: whether data and logs stay inside the boundary, whether weights can be operated and audited on chosen infrastructure, and whether those terms survive policy and ownership changes three years out. Equity and committee seats are how buyers lock the last question when no vendor can contractually guarantee its own permanence.
