Revenue +58%, margin 49% and more contracts ahead. Is this AI stock overlooked?
This company doesn't make chips or run data centers. It sells data that language models learn from, and tests that the biggest AI labs use to verify what their models still can't do. It recently reported its best quarter in history: revenue of $92.1 million, up 58% year over year, adjusted gross margin of 49%, nine percentage points above the target it had publicly set for years. Adjusted EBITDA rose from $13.2 million to $25.4 million, up 92%, because costs grew more slowly than revenue. Net profit doubled, and the company beat analyst consensus on revenue, margin and earnings per share.
Key points
After the best quarter in its history, the stock jumped 16%. The next day it gave back the entire jump and closed lower than before the results. The market saw three things in that release that didn't fit into the headline.
The largest customer cut revenue by about $16 million quarter-over-quarter, while the second largest added exactly that much. That's why total revenue rose only 2.2%, even though the largest client's share fell from 56% to 37%.
After the record quarter, the company confirmed but did not raise its outlook. Its own numbers imply a pace of around $85 million per quarter for the rest of the year, compared to the $92.1 million it just achieved.
Operating cash flow for the half-year was $164.4 million on revenue of $182.2 million. But most of that doesn't belong to the company. After adjustments, $134 million remains of the reported quarter-billion cash position.
The biggest player in the industry lost its main customer, which had been spending around $200 million a year, within a few weeks. One change in ownership structure was enough. The same relationship structure underpins 71% of this company's revenue.
The stock reacted with a 16% jump in after-hours trading. The next day it gave back the entire gain in a single session and closed lower than where it was the day before results. From the peak it reached that day, it is now down nearly a third, and from its 52-week high it is down more than half. The market didn't overlook the record quarter. It rejected it.
That reversal wasn't driven by the past quarter's numbers. It was driven by three things announced that same evening: a confirmed, not raised, full-year outlook, a CEO change, and a program to sell new shares worth $300 million, at a company with no debt. And then a fourth thing that management didn't announce at all, because it was hidden in the percentages inside the quarterly report. That is what most changes how those 58% should be viewed.
We're talking about the American company Innodata $INOD. Based in New Jersey, it has been operating for over thirty years, and today's business looks completely different than four years ago: the latest quarter alone brought in more than all of 2022. It counts five Magnificent Seven companies and several leading AI labs among its customers, and the second quarter was its twelfth consecutive quarter of year-over-year revenue growth.
Yet during trading on September 11, 2026, it is around $55.40 with a market value of about $1.9 billion. Its 52-week high is $125.14 and its 52-week low is $34.23, which itself indicates how volatile this stock can be. Only a handful of analysts cover it, and their average price target is more than double the current price. That is usually either a sign of undervaluation, or a sign that the risks the market sees are hard to reflect in such a narrow set of estimates.
A record quarter that the stock gave back within two days
The surprise was bigger on earnings than on revenue. Innodata reported net income of $14.4 million versus $7.2 million in the same quarter last year, and diluted EPS of $0.41 versus $0.20. Revenue beat analyst estimates by about 7%, adjusted EBITDA by about half. However, part of that difference did not come from operations: the effective tax rate fell to 18% in the quarter, against a long-term expected range of 23% to 25%.
Improvement is also visible in GAAP numbers, where nothing can be normalized. Gross margin rose from 39% to 46%, operating income from $8.9 million to $15.9 million. Over the year, the company grew its volume of work by more than half, while keeping a larger share of every dollar earned.
Yet the trading around the results was dramatic. On the release day, the stock fell during the session and closed at $65.48, then jumped to $75.88 in after-hours trading, and the next day reached $78.05. From there it fell back and ended the session lower than the day before. The record numbers didn't help the stock in the following weeks either. It lost about a tenth of its value in August and kept falling in September.
On the day of the results, after market close, the company filed a prospectus for an at-the-market program to sell up to $300 million of shares with Goldman Sachs as lead agent. At the then-current price, that represented roughly 13% of existing shares; at today's price, roughly 16%. Management described it as a flexible tool to fund growth and strategic opportunities, not as an announced offering. But for a company with no debt and a strong cash position, the market asked the usual question: for what.