NVIDIA ACE

NVIDIA ACE (Avatar Cloud Engine) is a suite of generative AI microservices designed to bring lifelike, interactive digital humans and avatars to games, customer service applications, and enterprise software. NVIDIA first announced the technology as the Omniverse Avatar Cloud Engine on August 9, 2022, positioning it as a set of cloud-native AI models that would let businesses of any size access the computing power needed to build assistants and avatars capable of understanding multiple languages, responding to speech, and making intelligent recommendations. The platform reached general availability for cloud deployment in mid-2024, alongside early access for local RTX AI PCs, with adoption from companies spanning customer service, gaming, and healthcare, including Dell Technologies, ServiceNow, Inventec, and Perfect World Games.

At its technical core, ACE combines several specialized NVIDIA AI microservices that developers can mix and match: NVIDIA Riva handles automatic speech recognition, text-to-speech synthesis, and translation across multiple languages; NVIDIA Nemotron large language models provide contextual language understanding and response generation, including Nemotron-3 4.5B, a small language model purpose-built to run on-device with accuracy comparable to cloud-based LLMs; and NVIDIA Audio2Face generates realistic facial animation directly from an audio track, with a companion Audio2Emotion model that infers a character's emotional state from speech to drive more expressive performances. Because these are packaged as modular NVIDIA NIM (NVIDIA Inference Microservice) containers, developers can run them flexibly across the cloud or locally on RTX-powered PCs depending on available hardware.

ACE's clearest showcase to date is Covert Protocol, a technology demo built by Inworld AI in partnership with NVIDIA and unveiled at GDC and GTC 2024, where players take on the role of a detective having real, unscripted conversations with AI-driven digital human characters built on Unreal Engine 5 and MetaHuman. Inworld's engine layers cognition, perception, and behavior systems on top of ACE's speech and animation microservices, letting each conversation and playthrough unfold differently based on what the player actually says, a meaningfully different approach to non-player characters than pre-scripted dialogue trees. Beyond gaming, ACE technologies are being adopted for telehealth digital humans, enterprise customer service avatars, and virtual factory applications involving humanoid robots.

Pricing follows NVIDIA's broader NIM and AI Enterprise licensing model rather than a simple per-seat or per-token structure: hosted NIM microservices are free to call for prototyping and evaluation through build.nvidia.com and the NVIDIA Developer Program, subject to model-specific rate limits, while production deployment requires an NVIDIA AI Enterprise license at $4,500 per GPU per year (or approximately $1 per GPU-hour in the cloud), with a free 90-day production-grade evaluation license available to test before committing. This makes NVIDIA ACE best suited for game studios, enterprise software teams, and healthcare or customer service platforms with real engineering resources and NVIDIA GPU infrastructure who want to build genuinely interactive, unscripted digital humans, rather than teams looking for a simple, low-cost off-the-shelf avatar widget.

NVIDIA ACE microservices follow NVIDIA's NIM (NVIDIA Inference Microservice) pricing model rather than a standalone product price. Hosted NIM endpoints are free to call for prototyping and evaluation through build.nvidia.com under the NVIDIA Developer Program, subject to model- and account-specific rate limits that NVIDIA does not publish as one universal quota. NIM containers can also be downloaded for free from NGC for development, testing, and experimentation on up to 16 GPUs. Production deployment requires an NVIDIA AI Enterprise license, priced at $4,500 per GPU per year for a self-managed subscription (or roughly $1 per GPU-hour when billed through cloud marketplaces on AWS, Azure, or GCP), with a $22,500 perpetual license option also referenced in third-party analysis. A free 90-day production-grade evaluation license is available before committing to a paid subscription. Public list pricing for the full AI Enterprise suite is not published; exact costs depend on GPU type, deployment scale (cloud vs. on-premises DGX), and support tier (Business Standard vs. Business Critical 24x7 coverage), and require contacting NVIDIA sales or an authorized partner for a formal quote.