ElevenLabs is taking its voice models off the cloud. The company opened early access for on-premise and on-device deployments, aimed at enterprises that can't or won't route audio through its SaaS stack. A company blog frames it as filling in the gap between cloud and VPC.
Two paths. On-Premise runs on standard GPU-enabled servers with Confidential Computing, pitched at government agencies and regulated sectors where cloud isn't available in-region. On-Device is built for offline inference on constrained hardware: entry-level GPUs, NPUs, modern CPUs, and ARM-based chips. Think vehicles, wearables, embedded systems.
Both variants claim 30+ language support, per the deployment page. Inference and audio processing stay inside the customer's environment. License validation and telemetry are optional and configurable, including fully air-gapped setups. Fine-tuning for specific languages or dialects is supported, the company says.
Pricing isn't listed publicly. ElevenLabs says it's case-by-case, typically a license fee plus a usage-based component. Also unlisted: exactly which models ship at launch. The company notes only that its local lineup won't mirror the full cloud portfolio, and that updates arrive on a controlled cadence.
The waitlist is open now. Initial releases are slated for the first half of 2026. VPC deployments on AWS SageMaker and GCP Vertex, the middle option between cloud and fully isolated, are already available.
Bottom Line
ElevenLabs is targeting the first half of 2026 for initial on-premise and on-device voice AI releases, with pricing negotiated case-by-case.
Quick Facts
- Two deployment paths announced: On-Premise and On-Device
- Initial release window: first half of 2026
- On-Device runs on entry-level GPUs, NPUs, ARM chips, modern CPUs
- 30+ languages supported on both variants (company-reported)
- Pricing: license fee plus usage-based component, case-by-case
