Background

NVIDIA is a global leader in accelerated computing and artificial intelligence, with roughly 42,000 employees across the U.S. and 37 other countries. Its business spans everything from consumer gaming to highly technical enterprise products, each requiring content that resonates with very different audiences, in 100+ markets—creating demand for accurate, on-brand content at global scale..

NVIDIA built an AI-powered translation platform on NVIDIA Nemotron Speech. To extend its capabilities, NVIDIA and LILT designed a flexible, AI-first workflow. NVIDIA’s model generates the initial translation. LILT then refines the output with adaptive multilingual AI models, routes critical content to expert human reviewers, and supports additional languages. Feedback from LILT flows back to NVIDIA, helping improve translation quality over time.

The Challenge

The legacy setup was holding the team back in a number of critical areas:

• Manual, fragmented workflows: Enterprise, consumer, corporate, and regional teams each handled translation differently, resulting in a patchwork of handoffs and inconsistent processes.
• Tool sprawl: The team was running through four different translation platforms to get content out the door.
• Redundant spend: Disconnected processes meant company-specific terminology (ie., translation memory) couldn’t be leveraged across vendors, so NVIDIA was paying repeatedly for content it had already translated.
• Limited alignment with NVIDIA’s AI roadmap: The existing translation approach could not integrate NVIDIA’s AI models or evolve with the company’s technical roadmap.

“When you’ve had a partner for that long, this team does it one way and that team does it another. We had to find a way to drive consistency and that’s where the change management came in.”

Jim Roots, Director, Digital Marketing at NVIDIA

Why NVIDIA Selected LILT

To pull off the transformation they had in mind, NVIDIA needed a partner who could meet three high bars:

• Deliver high-quality translations that are consistently publish-ready to regional teams
• Align and consult with NVIDIA's technical roadmap
• Integrate NVIDIA's AI translation models

LILT stood apart because it could do more than deliver translations. It could integrate NVIDIA Nemotron Speech into the workflow, combine adaptive AI with expert human review, and return structured feedback that continually improved NVIDIA’s translation models.

The modernized workflow

Working together, the two teams designed a flexible, AI-first workflow built on NVIDIA’s own models. Content is translated through NVIDIA Nemotron speech; a collection of open models for speech recognition, text-to-speech output, and translation. NVIDIA is putting its technology to work on its own content with LILT then handling the rest, refining output with its adaptive multilingual AI models, providing expert human verification on critical content, and covering the long tail of languages in addition to the 37 that Nemotron speech supports in Neural Machine Translation (NMT).

Additionally, LILT engaged directly with NVIDIA’s Digital Marketing engineering team early in the partnership, helping fine-tune the Nemotron speech models against NVIDIA’s brand, tone, and terminology and providing structured improvements over time. LILT built in regular cadences from the outset, ensuring the latest models were used and continually trained to improve quality.

“We needed a partner who could help improve our translation model. That’s a different level of partnership. The ability to feed updates and improvements back into the model has been incredibly impactful.”

Faylene Bell, Senior Director of Web Operations, Digital Marketing at NVIDIA.

LILT’s adaptive AI also keeps the system improving with use: real-time learning from human expert edits flows back as training signals, so the model keeps pace with NVIDIA’s evolving voice across product launches, social, and other fast-moving content.

Results

Consolidated stack, automated workflows drive efficiency

With LILT in place, NVIDIA’s engineering team consolidated and automated what used to be a manual workflow. The number of translation platforms in the stack dropped from four to two — NVIDIA’s homegrown solution and LILT. This consolidation cut recurring vendor fees and the operational weight of managing multiple tools. The move to automated, connected workflows delivered significant time savings: work that used to require manual handoffs across teams now flows through connected pipelines, freeing the team to ship more content, faster.

Translation Model Quality Improves 30%

With LILT’s feedback loop into NVIDIA’s engineering team, the quality of NVIDIA’s translation model lifted substantially by 30% within a year. That quality jump also let NVIDIA hand off quality assurance work that the team used to own, freeing capacity for higher-value work.

“The quality has improved significantly using the models we’ve worked hard on internally. That’s the biggest thing for us.”

Faylene Bell, Senior Director of Web Operations, Digital Marketing at NVIDIA.

12M+ Words Translated at Scale

LILT’s infrastructure has supported NVIDIA through 12M+ words of translation in a single year, and absorbed the content spikes that come with major company events. For NVIDIA’s flagship annual event, GTC, content is consistently turned around within a 6–12 hour window, fast enough to keep up with the pace the event demands.

Costs Fall 32% as Volume Doubles

As model quality climbed, human revision dropped and translation costs followed. The combined effect of better models, leveraged translation memory, and an AI-forward workflow drove translation costs down 32%, while the team doubled the volume of localized content shipped. The cost savings leadership had been striving to achieve through modernization were realized.

Toward a Fully AI-Powered Workflow

With quality established, NVIDIA is expanding automation to content types that meet defined quality thresholds. LILT’s agentic workflows automate intake, routing, review, and delivery, while LILT’s MCP integrations bring translation directly into NVIDIA’s generative AI assistant.

A Blueprint for AI-Powered Global Content

NVIDIA and LILT turned a fragmented translation operation into an AI-first system that improves with every project. The partnership gives NVIDIA a scalable path to faster global publishing, lower costs, and increasingly automated workflows—while keeping quality at the center.

Go deeper: See how NVIDIA built its AI-powered global translation platform.