NVIDIA Story
How NVIDIA Uses AI to Scale Global Translations
Objective
Content moves fast. Translation doesn’t. Every global marketing team knows the bottleneck—campaigns held up, regional launches delayed, and a tangle of vendor plug-ins adding cost every time a platform changes or a vendor turns over.
NVIDIA's Digital Marketing team faced the same problem—and solved it with NVIDIA AI. Using NVIDIA Nemotron™ speech, the team built an internal translation platform trained on more than seven years of NVIDIA marketing data, and the results redefined what localization at scale looks like.
Topic
Generative AI / LLMs
Products
Key Takeaways
- ∼70% reduction in translation turnaround time per request.
- ∼25% reduction in annual cost savings by eliminating legacy integration plug-ins and modernizing vendor usage.
- 3,000+ translation requests, 20,000 unique files, and 11 million words processed through the platform, as of spring 2026.
Modernizing Translation for Global Marketing at Scale
NVIDIA’s translation workflow depended on a vendor ecosystem stitched together with integration plug-ins—one per platform, each requiring custom development work. Every vendor change meant rebuilding integrations, adding risk to live marketing operations, and absorbing thousands of dollars in plug-in costs per year. As global campaign volume grew, the model couldn’t scale. Regional teams waited in queues. Video, web, and design asset localization lagged behind. And with AI already reshaping the translation industry, continuing to rely on a manual, plug-in-dependent architecture meant falling behind a curve that NVIDIA helped create.
“Translation vendors were all moving to AI—it was only a matter of time,” said Faylene Bell, Senior Director of Web Operations, Digital Marketing at NVIDIA. “Our thinking was straightforward: If that’s where the puck is going, we should be using our own technology to get there first.”
Building an AI-Powered Translation Backbone With NVIDIA Nemotron Speech
NVIDIA’s Digital Marketing team built an internal translation service on NVIDIA Nemotron Speech and the NeMo framework. At its core is a Nemotron speech translation model, pretrained on more than seven years of NVIDIA marketing translation data—giving it deep familiarity with NVIDIA products, terminology, and voice from day one. For video content, the pipeline adds Parakeet for transcription, running on NVIDIA Cloud Functions (NVCF) to provide end-to-end workflows from transcription through translation.
“Working with the Digital Marketing team gave us a real-world proving ground for Nemotron speech and NeMo framework capabilities,” said Oluwatobi Olabiyi, Director of Engineering at NVIDIA. “ Seeing models trained on seven years of NVIDIA marketing data get put to work at this scale—across platforms, languages, and formats—is exactly the kind of internal deployment that sharpens the technology for everyone.”
The first major integration was with Adobe Experience Manager (AEM), NVIDIA’s web content management system. Teams could view an English page and see it automatically translated with layout and assets fully preserved—a capability that was genuinely novel at launch. Since then, the platform has expanded to email and campaign tools, enabling teams to request and receive translations directly inside existing workflows. Because its API-driven quality standards are configurable, content can be automatically routed for human linguist review where it matters most, and new platforms can be onboarded without rebuilding the integration layer.
Faster Localized Content, Lower Costs
The pages and campaigns running on the new platform weren’t starting from zero. Translation workflows were already in place, managed by professional linguists. The gains came on top of that baseline.
Turnaround time dropped by approximately 70% per request. Eliminating legacy plug-ins and adopting a modern vendor model also drove a 25% reduction in annual costs while still routing expert linguists to content where human judgment matters most. Adoption has been strong: more than 3,000 requests processed, representing tens of thousands of unique files and millions of words localized through the platform.
The result was a fundamental shift—from fragmented, ticket-based localization to a shared platform any team can build on. Regional marketers now have self-service tools and faster turnaround. New platforms onboard through a single API, without custom plug-in work. And every integration informs best practices for how other NVIDIA teams—and enterprise customers—can deploy AI translation at scale.
Looking ahead, the team is expanding support to design assets, additional video formats, and more languages, further unifying how NVIDIA creates and localizes content worldwide. What the Digital Marketing team built for its own global campaigns, it built on the same stack available to any enterprise ready to close the gap between content velocity and global reach.
“Every time we use NVIDIA’s own AI stack to solve a real operational problem — and deliver results like these — we’re not just improving our workflows. We’re proving what’s possible for our customers. The translation platform is another example of NVIDIA being its own first customer for enterprise AI.”
LaSandra Brill
VP of Digital Marketing, NVIDIA
Learn more about NVIDIA Nemotron speech—and translation supporting 36 languages.