INNODATA INC (INOD) stock intrinsic value fair value estimate & owner earnings analysis

Innodata prepares, labels and evaluates the data that artificial intelligence companies use to train and refine their models. Its customers include several leading AI research labs and five of the large technology companies often called the Magnificent Seven. The business changed completely in 2024: revenue was $87 million in 2023, then $170 million in 2024 and $252 million in 2025, and $182 million in just the first half of 2026 (up 56% on the same half a year earlier). Before that shift, owner earnings - the cash left for shareholders after the capital spending needed to keep the business running, after stock compensation, and after stripping out swings in receivables and payables - were close to zero for most of 2016 through 2023. They were about $0.87 per share in 2024 and $1.10 in 2025. We value the business as it is today, from the 2024-2025 period, and do not blend in the older, smaller business. Two accounting judgments matter. Stock compensation is treated as a real cost: it was $11.1 million in 2025 and $13.1 million in the first half of 2026 alone. And only the part of capital spending needed to maintain the business is charged against earnings; spending that tracked revenue growth is not. The company carries no borrowings and its $30 million Wells Fargo credit line was undrawn. The cash balance needs care. Cash and short-term investments were $250 million at June 30, 2026, but most of the jump since December came from customers prepaying for pass-through costs - customer advances rose from $1.8 million to $67 million and payables and accrued expenses rose by about $72 million. That cash belongs to suppliers, not shareholders. The company itself puts cash net of those prepayments at about $134 million. We hold back a further $25 million as working cash and $5.3 million for an old Philippine labor judgment against a former subsidiary that the company disputes, leaving about $104 million, or about $2.98 per share, of surplus cash. The very large first-half operating cash flow ($164 million) is almost entirely this prepayment and is not treated as earning power. The central fact of this business is how little of its revenue is contractually secure. One customer was about 58% of 2025 revenue and 48% of 2024 revenue. The company's own filing says its agreements generally do not oblige customers to buy anything in future periods and can usually be ended on 30 to 90 days' notice. There is encouraging news: in the second quarter of 2026 the largest customer fell to 37% of revenue while a Big Tech customer rose to 34%, and the company added a new fast-growing frontier lab. But this is diversification across two or three giant accounts, all of which have the resources to do this work themselves or move it elsewhere. Because none of this volume is contracted, we credit no growth beyond what has already been earned: the market for AI data expanding is not something we can count on. Base case: earnings start at about $1.38 per share of cash before any interest, an increase of about 25% on 2025, matching the 26% growth actually achieved in 2025 and well below the first half of 2026's pace, which already earned roughly half of that figure. That level holds for five years, then fades over the next five years to about $0.45 per share as AI labs bring work in-house, prices compete down and individual programs end, which is how this kind of project work has behaved. The tail is supported by the Big Tech relationship and smaller accounts, and it ends after year fifteen because nothing contracted reaches that far. The business is worth about $331 million, the company owes nothing, and with about $104 million of surplus cash the shares are worth about $12.51. Bear case: the largest customer cuts most of its work during 2026's second half, as Google did to the AI data supplier Appen when it ended an $82.8 million contract effective March 2024 with about two months' notice. Innodata keeps the year's first-half earnings, about $1.10 per share for 2026, and then the business settles near $0.20 per share for nine years. That is a little better than the near-zero earnings of 2016-2023, because the remaining Big Tech and enterprise work is larger than the old business. The business is worth about $81 million, with no debt, and adding the surplus cash gives about $5.30 per share. The cash is a real cushion here, providing more than half of the bear value. Bull case: the diversification seen in the second quarter holds and the current run rate continues - about $1.85 per share for ten years, roughly the pace of the first half of 2026 - then about $1.00 per share for ten more years as a more diversified business keeps its accounts. The business is worth about $616 million, with no debt, which gives about $20.71 per share. Our previous estimates were built on a different basis that credited continued revenue growth from the expanding market, so they are not comparable with these figures. In the business itself, since then, customer concentration has eased noticeably and earnings have kept rising, both of which help, while the filing's description of customer terms confirms how little of this revenue is locked in. Margin of safety. We are fairly confident in the near-term earnings level, because half of 2026 is already reported and the earnings show up in net income ($29.3 million for the half) as well as cash. We are much less sure how long it lasts. The worst plausible outcome is the loss of the largest one or two accounts within a year, which the 30-to-90-day termination terms allow. The structure would survive that: there is no debt and the surplus cash alone is worth about $3 per share, but earnings could return to the near-zero levels of 2016-2023. The things we are most likely to be wrong about are how long the AI labs keep outsourcing this work and how fast pricing falls. If the fade starts immediately rather than after five years, the shares are worth roughly the bear figure of about $5. The price assumes about a three-year holding period. The company pays no dividend we have assumed, so the return comes from retained cash and the value of the business. From $6.50, the base case implies about 29% a year over three years, counting the cash the business earns along the way. The bear case implies a loss of about 5% a year, cushioned by the cash on hand. The margin-of-safety price is $6.50 because at that price the cash and roughly two years of current earnings cover nearly the whole price, so in the bear case a sudden customer loss implies a modest loss rather than a ruinous one, while the base case clears the target comfortably.