ElevenLabs expands into enterprise agents, tuning and partnerships

ElevenLabs plans to add 15 megawatts (MW) of NVIDIA Vera Rubin compute capacity and reports more than 550 partners across 44 countries, according to its November 11, 2026 announcement at the New York Summit.

At the summit, ElevenLabs previewed fine-tuning (adapting an existing model to a customer domain or workflow) for selected open-source language models and its voice models using data from live deployments. It also introduced the ElevenAI workplace app, a Partner Program, a Marketplace and an NVIDIA model integration.

Together, the package extends ElevenLabs beyond voice generation and transcription into software for managing customer conversations and employee work, the company said.

CEO Mati Staniszewski said conversational agents would become “critical infrastructure for every enterprise, as essential as cloud or payments” within a few years.

His comparison sets a high bar for enterprise adoption: ElevenLabs must turn live interaction data into reliable models while routing routine work away from frontier models. The announcement also ties the strategy to security, data residency and workflow integration, requirements for large enterprise deployments.

What can ElevenLabs customize?

Customization work is scheduled to begin in the coming months, ElevenLabs said. Businesses will work with the company’s researchers to tune models for their own customers and operating context, selecting optimization metrics such as customer satisfaction or reliability.

ElevenLabs says live-deployment data lets clients keep refining models as interactions accumulate. That creates an ongoing feedback loop, rather than leaving companies with only a general industry model or a one-time customization.

ElevenLabs already applies this industry-specific approach through Scribe v3 Medical, a healthcare transcription model built to recognize specialized terminology. Staniszewski said a domain-optimized financial-services model is also in development.

The medical example shows the intended level of specialization: specialized terminology rather than generic transcription. The financial-services model would extend that approach to another domain with distinct language and workflows, although ElevenLabs disclosed no release date.

How does the Interaction Engine route work?

The previewed Interaction Engine addresses a separate problem: deciding how a conversation should run. ElevenLabs says it manages turn-taking and interruptions and can assign different language models to different parts of an exchange.

A system can use a cheaper model for triage and a frontier model for complex tasks. This design keeps routine exchanges from requiring a frontier model at every step and is intended to limit token use.

For high-volume deployments, this turns model selection into a routing decision at each stage. ElevenLabs presents cheaper triage as the way to reserve frontier models for complex work, although the announcement provides no benchmark scores or cost figures.

What do Memory and Routines retain and automate?

ElevenLabs says its ElevenAgents platform for deploying voice and chat agents will add two forthcoming features: Memory and Routines. Memory will recognize returning authenticated customers and distill each conversation into a structured record of preferences, constraints and future occasions.

The customer relationship management system will remain the primary customer record, ElevenLabs said. A customer who prefers Spanish and can talk only after 5 p.m. would not need to repeat those details on the next call.

Memory’s design also limits duplication of the customer record: ElevenLabs positions it as a conversational layer, not a replacement for the CRM. The 5 p.m. example makes that retention claim concrete.

Routines will automate multi-step work before, during and after conversations, according to ElevenLabs. The feature will connect with existing tools and systems, including these workflows:

• Pre-screen a mortgage applicant and call the applicant back within minutes.
• Remind patients about visits and offer canceled slots to waitlisted patients.
• Email reports, write to a customer relationship management system and initiate calls.

Both features remain forthcoming. These examples span mortgage pre-screening, healthcare scheduling and routine service work. Together, they move ElevenAgents from answering inquiries toward completing the work surrounding a conversation, according to ElevenLabs’ own summary.

What does ElevenAI add for employees?

Staniszewski previewed ElevenAI, an employee app that listens to spoken conversations, including in-person meetings, and produces notes, summaries and action items. It can also provide live assistance and pass context between people and agents.

ElevenLabs says the app takes meeting notes with participants’ consent and includes notice and consent mechanisms.

Free early access opened with the announcement. Dictation and meeting notes use Scribe v3, which ElevenLabs describes as its recorded-audio transcription model covering 102 languages. The separate Scribe v3 Live variant targets low-latency, real-time audio.

How will partners deploy and distribute agents?

ElevenLabs reports more than 550 partners across 44 countries, including Google Cloud, IBM, Deloitte, DXC and Genesys. The new Partner Program lets consultancies, systems integrators, software vendors and agencies integrate ElevenAgents into enterprise technology stacks. Participants can also work with ElevenLabs’ forward-deployed engineers on complex deployments.

An alliance with McKinsey and QuantumBlack, AI by McKinsey, pairs the platform with McKinsey’s enterprise-transformation work. The accompanying Marketplace initially offers a curated catalog of partner-built applications for healthcare, financial services, insurance, education and retail.

Enterprises can browse the applications and contact their developers. Purchasing through existing ElevenLabs commitments is planned to start in 2027.

The partner count establishes distribution reach, but ElevenLabs does not present it as an active-deployment total. The Marketplace adds a second route to market through partner-built applications rather than only direct ElevenLabs deployments. Browsing therefore precedes a later purchasing test.

What does the NVIDIA expansion enable?

ElevenLabs and NVIDIA announced that Nemotron 3.5 Super VL is available in ElevenAgents. The planned 15 MW addition of NVIDIA Vera Rubin capacity sits on top of ElevenLabs’ existing Blackwell Ultra systems.

ElevenLabs says the addition is intended for large deployments with security or data-residency requirements and will give researchers access to optimized compute. For enterprises, data residency (keeping data within a required jurisdiction) now sits alongside model quality as an infrastructure criterion.

The planned commitment would give ElevenLabs more capacity for research access and security-sensitive workloads. Its practical effect still depends on when organizations move the previewed tools into production and put the added infrastructure to use.

What does Poland’s deployment show?

The expanded collaboration also covers an on-premises voice agent (one deployed inside the customer’s own infrastructure) for Centralna e-Rejestracja, the central appointment system in Poland’s national e-Health platform. EasyCall supports delivery.

The agent calls patients who have only landlines to remind them about appointments and help them reschedule. ElevenLabs says the central system handles 40 million appointments annually and has no-show rates of 10% to 20%. Those figures define the operational problem.

The deployment uses NVIDIA Confidential Computing to keep citizen data and audio within government infrastructure and protect them during processing. The Polish deployment joins high service volume to a strict data-governance requirement, placing data protection inside the deployment architecture rather than in a separate compliance layer.

What has to happen next?

The Summit package makes ElevenLabs’ enterprise strategy depend on more than model performance: tuning and Memory draw on live-use data, Routines connects conversations to business systems, and the Partner Program and Marketplace create routes into enterprises. Two execution tests follow:

• Move customization, Interaction Engine, Memory, Routines and ElevenAI from preview or early access into broader production use.
• Convert Marketplace browsing and developer contact into purchases through existing ElevenLabs commitments, beginning in 2027.

Production remains the test. The stakes are measurable: ElevenLabs has disclosed feature status and partner scale, but not production-adoption figures for the previewed tools or benchmark results for the industry-specific models.

The next proof points are production adoption of the previewed tools and Marketplace purchases through existing commitments in 2027. Those results will test whether ElevenLabs can convert live conversation data, more than 550 partners and planned compute capacity into sustained enterprise use.