Why we bet on open weights

Sovereignty and open weights are the same bet. A closed API can't be pinned to a jurisdiction — the weights live in someone else's region, behind someone else's control plane. You can't attest to what you can't run.

That trade-off used to cost quality. It doesn't anymore. Frontier open-weights models — Llama 3.3/4, Mistral, Gemma, Phi — are production-grade, and the gap to closed models has closed for the large majority of enterprise tasks. For the rest, customers bring their own fine-tunes, LoRA or full weights, which we always support.

Open weights also let us hold a clear catalog policy. A Western-default Standard catalog runs for everyone; Chinese-origin models are opt-in and blocked by default for federal and regulated workloads. The choice is the customer's, made explicit in contract, not buried in a model provider's terms.

Most of all, controlling the weights and the metal is what makes attestation real. Volt can prove which model ran, in which metro, on which hardware — because all three are inside the boundary, not rented from across a border.