As Meta’s new AI assistant “Muse” surges to the top of the App Store, a quiet, New Jersey-based company is emerging as a critical—if unheralded—architect of its success. While Meta’s $14.3 billion investment in industry giant Scale AI grabbed headlines, reporting indicates that Innodata ($INOD), a decades-old firm once known for simple data digitization, remains deeply embedded in Meta’s most ambitious AI projects.
In August, amidst concerns that its Meta business was plateauing, Innodata disclosed a new contract focused on the “personalization of long-horizon agents.” While neither company has officially confirmed the link, industry analysts and former employees suggest this project is the engine behind Muse. The assistant, which can navigate apps, recall user preferences, and perform multi-step digital tasks, requires precisely the kind of human-in-the-loop training and “simulated environments” that Innodata has spent the last year refining.
The stakes for the partnership are immense. In the current “frontier AI” race, where computing power is constrained, the quality of a model is increasingly defined by its data. To teach a model to be a truly useful agent—such as booking a complex trip or managing a calendar—AI developers need more than just scraped internet text; they need human judgment to bridge the gap between “valid” outputs and “meaningful” results.
For Innodata, the relationship with Meta has been transformative. In 2025, that mystery customer accounted for roughly $146 million, or 58% of the company’s total revenue. This reliance fueled a staggering stock rally, propelling Innodata from a $4 valuation in early 2023 to over $120 per share by June 2026. However, the company has faced significant turbulence since, with shares sliding back toward $57 amid investor fears that Meta might consolidate its data operations in-house following its massive stake in Scale AI.
Despite these fears, internal evidence suggests the relationship is far from over. Former employees who worked on the Meta account describe a deepening entanglement, with Innodata personnel working closely across multiple AI workflows. One former director noted that during 2024, roughly 400 full-time equivalent employees were dedicated solely to Meta, and recent commentary from Innodata management suggests they are growing alongside Meta’s flagship programs rather than being replaced.
The company is now pivoting heavily toward “agentic reinforcement learning,” creating artificial, messy digital scenarios where AI agents learn to recover from errors and handle conflicting instructions. This expertise in “objectifying subjectivity” is what allows Muse to understand not just how to book a flight, but how to plan a trip that aligns with a user’s specific tastes.
Innodata, however, remains a controversial player. Short sellers have previously targeted the firm, questioning the sustainability of its AI revenue, and the company has weathered both SEC and DOJ investigations, which were closed in 2025 without enforcement action. With the company’s executive team having sold significant portions of their holdings throughout the year, the path forward remains volatile.
As Meta’s Muse continues its rapid adoption, Innodata finds itself in a precarious yet powerful position. Whether this “agentic” work becomes a durable, long-term profit engine or a fleeting assignment will be the ultimate test for the firm’s incoming CEO, Rahul Singhal, who takes the helm on September 30. For now, in the shadow of Silicon Valley’s giants, a New Jersey data firm continues to train the mind behind the world’s most popular new AI.
Disclaimer: This content is auto-generated for informational purposes only.
Source: Read Original News
