ElevenLabs plans to add 15 MW of Nvidia Vera Rubin computing capacity on top of its existing Blackwell Ultra systems. It says Nvidia’s open Nemotron 3.5 Super VL model, now available in its ElevenAgents platform, completed its customer-support benchmark tasks “at 70% lower costs, compared to other leading open models”. The two companies announced the expanded partnership on 11 November at ElevenLabs’ New York Summit.
The first named deployment is Poland’s national appointment system, which handles 40 million appointments a year with no-show rates of 10 to 20 percent, ElevenLabs says. That is 4 million to 8 million missed appointments a year, or about 77,000 to 154,000 a week, by our arithmetic.
Business Pill · Open Weights Explained
An evergreen Business Engineer explainer on open-weight models: renting a model or running it yourself. It teaches the general idea only and does not discuss ElevenLabs or Nvidia.
The key insight:Nvidia’s model inside ElevenAgents is open, and ElevenLabs’ voice models are its own: the release pairs the two, while ElevenLabs adds 15 MW of Nvidia Vera Rubin capacity to run them. Its first named public customer, Poland’s appointment system, loses 4 million to 8 million appointments a year to no-shows, by our arithmetic.
The release lists four parts. Nvidia’s Nemotron 3.5 Super VL is now available in ElevenAgents. ElevenLabs is adding 15 MW of Vera Rubin capacity for its largest customers, including deployments with strict security and data-residency requirements. Its models and Nvidia infrastructure power a voice agent for Poland’s e-Health platform. And the two are exploring on-device voice AI for robots, vehicles and industrial machines.
Mati Staniszewski, ElevenLabs’ co-founder and chief executive, said: “As voice agents become critical infrastructure, businesses need to be able to operate securely, efficiently, reliably, and with control over their data”. Jensen Huang, Nvidia’s founder and chief executive, said in the release: “For decades we adapted to computers — keyboards, screens, menus. Now computers are learning to speak our language”.
Poland’s Centralna e-Rejestracja handles 40 million appointments a year with no-show rates of 10 to 20 percent, per ElevenLabs’ 11 November release; the missed-appointment counts are our arithmetic.
The 70% Cost Claim
The figure comes from ElevenLabs’ own benchmark for realistic customer-support tasks. On it, the release says, Nemotron 3.5 Super VL “achieved leading end-to-end task-completion accuracy”, including in financial calculations and billing analysis, at the 70% lower cost. With reasoning turned off for simpler tasks, ElevenLabs says, it is among the three fastest models on the platform.
The comparison is with other open models, not with closed ones, and the release does not name the models compared or give prices per call. ElevenLabs pairs these open language models with its own voice models.
The Poland Deployment
The agent serves Centralna e-Rejestracja, the central appointment system in Poland’s national e-Health platform, with EasyCall delivering the solution. It calls patients who only have landlines to remind them of appointments and help them reschedule. The release says “even a small reduction in missed visits can significantly improve access to care”.
It runs on-premises on Nvidia Confidential Computing, so citizen data and audio stay within government infrastructure, ElevenLabs says. By our arithmetic, every percentage point off the no-show rate is 400,000 fewer missed appointments a year on 40 million bookings.
The Speech-to-Text Side of the Summit
At the same summit ElevenLabs described Scribe v3 Medical, a transcription model for healthcare that it says tells apart drug names such as cycloserine and cyclosporine. Staniszewski said a model tuned for financial services is in the works, and that in the coming months ElevenLabs will help businesses fine-tune certain open-source language models and its own voice models.
Scribe v3 itself cut word errors 27% against Scribe v2 at the same $0.20 per hour of audio, as we reported. Voice agents are a crowded field: on one blind ranking in October, Cartesia, Google and OpenAI held first, second and sixth place, as we reported.
Four Years of Nvidia Ties
Nvidia and ElevenLabs have worked together since 2022, the release says. ElevenLabs’ first text-to-speech model was trained on Nvidia GPUs, Nvidia has used ElevenLabs voices for more than 50 videos since 2024, including translations of Huang’s keynotes, and in 2025 it became a strategic investor.
ElevenLabs said in September that a $300 million employee tender valued it at $22 billion, double its Series D valuation in February, after a first round in 2022 at $9 million. That is about 2,444 times in four years, by our arithmetic. Enterprise accounts for 55% of its revenue, it said. The 15 MW is a quarter of the 60 MW IT load in Daikin’s Vera Rubin cooling design, as we reported.
The Structural Read
As we read the release, Nvidia supplies two layers to ElevenLabs, the open language model and the chips, while ElevenLabs supplies the voice layer, the part it built first.
As we read the Poland deployment, the business case rests on one rate: each percentage point of no-shows on 40 million bookings is 400,000 appointments a year, by our arithmetic, and the release gives no result yet.
ElevenLabs release, 11 Nov 2026
“The deployment runs on-premises using NVIDIA Confidential Computing, keeping citizen data and audio within government infrastructure and protected during processing”
Three Implications
FOR ELEVENLABS It adds 15 MW of Vera Rubin capacity on top of its Blackwell Ultra systems, for its largest customers and its research teams, the release says.
FOR NVIDIA Its open Nemotron 3.5 Super VL is now one of the models in ElevenAgents, which ElevenLabs says ran its support benchmark at 70% lower cost than other leading open models.
FOR PUBLIC SERVICES Poland’s appointment system handles 40 million bookings a year; 4 million to 8 million are missed at the stated no-show rates, by our arithmetic.
The framework reads: “Close the SCARCE layer (keep proprietary) + Open the ABUNDANT layer (commoditize)”
As we read it, the release splits the stack that way: Nvidia offers Nemotron as an open model, the layer it makes abundant, and sells the scarcer layer, the 15 MW of Vera Rubin compute; ElevenLabs supplies its own voice models. The framework predates the release and does not discuss it.
ElevenLabs’ valuation at its first round in 2022, its Series D in February 2026 (implied as half of $22 billion) and its September 2026 tender offer, from its 30 September post. The multiple is our arithmetic.
What Is Not Established
ElevenLabs gives no cost, site or date for the 15 MW. Its summary bullet describes the capacity as compute based on Blackwell Ultra systems and the Vera Rubin platform, while the body describes Vera Rubin capacity added on top of existing Blackwell Ultra systems. The 70% figure is self-reported on ElevenLabs’ own benchmark. The release gives no result yet for no-shows in Poland, and the physical-AI work is described as exploratory. The deal terms between the two companies are not disclosed. We did not contact ElevenLabs or Nvidia for this piece.
This Week In The AI Supercycle
The Great Convergence of Capital, Power, and Geopolitics
This week’s issue: for two years the AI build-out was priced as a supply story, and last week as a credit story. This week it was priced as a permission story, and the answer arrived as a cost, not a ban.
ElevenLabs is adding 15 MW of Nvidia Vera Rubin capacity and offering Nvidia’s open Nemotron model inside ElevenAgents, which it says ran its support benchmark at 70% lower cost than other leading open models. Its first named public deployment targets a system where 4 million to 8 million appointments a year are missed, by our arithmetic.
A note on sourcing. We read ElevenLabs’ release on the expanded Nvidia partnership and its recap of the Summit NYC announcements, both published on 11 November 2026, and its 30 September 2026 post on the $22 billion tender offer, in full. The 70% cost and accuracy claims are ElevenLabs’ own benchmark results. Missed-appointment counts, the valuation multiple and the 15 MW share are our arithmetic.
The Scribe v3, voice-ranking and Daikin figures are from our earlier, linked pieces. We did not contact ElevenLabs or Nvidia for this piece.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
