AI21 Labs

AI21 sells model access the way infrastructure is sold: metered. Enterprises and developers pay per usage to run the Jamba family through AI21's API, and the models are also distributed through cloud-platform partnerships, letting buyers spend committed cloud budgets — a channel suited to AI21's enterprise focus. On top of consumption, AI21 signs enterprise agreements for private deployments of Jamba — the long-context, memory-efficient architecture is pitched precisely at customers running models inside their own environments — and those arrangements layer contract and license economics over usage. Maestro, the planning-and-orchestration system introduced in 2025, extends the model up the stack: AI21 charging for reliability and orchestration of complex multi-step tasks rather than raw tokens, its clearest move from commodity inference toward system-level enterprise deals. The open-weight Jamba releases function as top-of-funnel distribution rather than revenue. The recurring base is therefore usage-dominated with enterprise contracts adding committed floors, and the strategic intent behind Maestro is to shift mix toward higher-margin platform agreements.