$INOD: The Company That May Be Training Meta's Muse Just Got Outed. The Stock Is Up 22% in Two Days.
Innodata rallied 22% in two sessions after Hunterbrook tied its biggest unnamed customer to Meta's hit AI agent Muse. The filings show a $67 million prepayment tell, a diversifying customer base, and a valuation hanging on one unconfirmed sentence.
Meta finally has its consumer AI hit. Muse, the personal agent it launched this month, reportedly went to No. 1 on Apple's U.S. free iPhone app chart within ten days, and on Monday, September 21, Meta stock jumped 11.3%, adding well over $100 billion in market value in a single session.
The more interesting trade, though, is happening three tiers down the supply chain, at a 38-year-old data company in Ridgefield Park, New Jersey that most investors had never heard of until 2024.
Innodata (Nasdaq: INOD) closed Tuesday at $69.97, up 14.6% on the day and up 22.4% in two sessions. The trigger: investigative outlet Hunterbrook Media published a report on Tuesday arguing that Innodata's largest customer, which the company has never named in its filings, is Meta, and that Innodata's newest disclosed assignment, a program covering "personalization of long-horizon agents" for that customer, sounds a lot like the training work behind Muse. Hunterbrook's affiliated fund disclosed it is long the stock, so read the enthusiasm accordingly. Neither Innodata nor Meta confirmed the connection.
Here is what makes this a genuinely interesting setup rather than a sympathy pump: the Meta relationship is simultaneously Innodata's entire bull case and its most dangerous liability. One customer generated 58% of Innodata's $251.7 million in 2025 revenue, roughly $146 million, according to the company's 10-K. That customer's spending then dropped by about a third in a single quarter this year, from roughly $50.5 million in Q1 to $34.1 million in Q2, and the stock spent the summer sliding on exactly the fear you would expect: that Meta, having paid $14.3 billion for 49% of Scale AI, would bring this work to its new in-house supplier and leave Innodata behind.
The Muse launch flips that narrative. If the hottest consumer AI product in the world needs continuous rounds of human-guided training, evaluation, and reinforcement-learning environments, and if Innodata is the vendor doing that work, then the customer-concentration bug becomes a feature again.
Why this is trending now
Three things landed on top of each other. First, Muse's launch reception: Meta describes an agent that remembers preferences, works across apps, browses, and keeps executing after you close the app, and the app-store numbers suggest consumers are actually using it. Second, Hunterbrook's report, which used interviews with former employees, public hiring records, and Innodata's own disclosures to tie the two companies together. One former director quoted in the report said roughly 400 Innodata full-time equivalents were working on the Meta account back in 2024, and a former employee pushed back directly on the Scale-displacement fear: asked whether Meta cut outside help after the Scale investment, they answered, "If anything, I would say it's more."
Third, and least noticed: Innodata has been quietly assembling a story beyond Meta. A second unnamed Big Tech customer doubled from 17% of revenue in Q1 to 34% in Q2, almost matching the largest account. The company signed "one of the fastest-scaling frontier labs" as a new customer in Q2. On September 15 it put Admiral Michael Rogers, the former NSA Director and Commander of U.S. Cyber Command, on its board to push a federal AI practice. And on September 30, a week from now, a planned CEO transition takes effect.
So the question the market is now pricing at $70 a share, up from $34 in March and down from $121.50 in June: is Innodata a durable arms dealer to the agentic AI buildout, or a single-customer body shop enjoying one more momentum wave? The filings contain real answers, including a $67 million tell on the balance sheet that almost nobody is talking about.
Get the next briefing free in your inbox: Join AlphaBriefing
The rest of this briefing is for paid members: the customer-concentration math quarter by quarter, the $67 million prepayment signal buried in the 10-Q, what the stock is actually pricing at $70 versus the company's own 40% guidance, the catalyst dates through November, and scenario-by-scenario price zones.
AlphaBriefing Paid gets you every investment thesis, scenario framework, and catalyst brief we publish, the analysis private intel clients pay four figures for, at a fraction of that.