Innodata Inc. (INOD) Earnings Call Transcript & Summary

November 5, 2026

NASDAQ US Industrials Professional Services earnings 42 min

Earnings Call Speaker Segments

Amy Agress

executive
#1

Thanks, operator. Good afternoon, everyone, and thank you for being with us today. On the call we have Rahul Singhal, Innodata's President and Chief Executive Officer, along with Jayant Chauhan, our Chief Financial Officer. Joining us as well is Aneesh Pendharkar, Senior Vice President, Finance and Corporate Development. Rahul will speak first and give some perspective on the business, after which Jayant will walk through our third quarter results. Following that, we'll open things up for analyst questions. Before we begin, a reminder that this call will include forward-looking statements — predictions, projections, and other statements concerning future events. Such statements represent management's views as of today and rest on current expectations, assumptions, and estimates as of today, so they carry risks and uncertainties. Results in reality may differ materially from what those forward-looking statements contemplate; they include, without limitation, statements about our future financial and operational performance, business strategies, growth initiatives, expectations, projections, and guidance. A more detailed account of the factors that could cause actual results to differ materially appears in today's earnings press release and in our latest annual report on Form 10-K — including the sections titled "Cautionary Note Regarding Forward-Looking Statements" and "Risk Factors" — along with our later quarterly reports on Form 10-Q and other reports and filings with the Securities and Exchange Commission. Those documents can be accessed via the Commission's website and on our own website; we assume no obligation to publicly update or revise any forward-looking information. Also, on this call we may discuss certain non-GAAP financial measures. In our earnings release filed with the SEC today, and in our other SEC filings, which are posted on our website, you'll find additional disclosures about these non-GAAP financial measures, including reconciliations of them with comparable GAAP measures. Thank you. I'll now hand the call over to Rahul.

Rahul Singhal

executive
#2

Thank you, Amy, and good afternoon everyone. Before diving into the quarter, I want to take a moment to acknowledge what this represents. This is my first earnings call as Innodata's CEO. I'm honored by the trust that Jack and the board have placed in me, and I'm keenly aware of what I've stepped into: a company with extraordinary momentum, an extraordinary team, and an extraordinary opportunity in front of us. I'm eager to walk you through our results and our strategy. I also want to acknowledge Jack. As Executive Chairman, he remains deeply engaged, collaborating with me and our research team to build differentiated capabilities and bring them to the federal government and to enterprise clients. Our strategy is a shared one, and the research-led engine behind our results is precisely what I intend to continue building on. Last quarter I committed that I would repay the confidence placed in me with results, and I'm pleased to start with a quarter that lets me do exactly that. Q3 was another record revenue quarter for Innodata. Revenue reached $95.6 million, up 53% year-over-year and ahead of the $83.3 million analyst consensus by approximately $12.3 million, or 15%. That marked our 13th straight quarter of year-over-year growth. We also exceeded analyst consensus on other key metrics. Adjusted gross margin was 40%, in line with the 40% target we have publicly stated. Adjusted EBITDA came to $17.7 million, or 19% of revenue, beating consensus by roughly $2.3 million. Net income was $10.3 million, up 24% year-over-year, and fully diluted earnings per share were 29 cents, compared with the 17 cents consensus. As we had hoped, Q3 and Q4 so far have delivered strong new business wins. The customer we described last quarter as our second-largest became our largest customer in Q3, and it continues to expand its work with us. During the quarter, we signed a new program with this customer for pre-training data, with a contract value of $29 million. Well into Q4, we have signed another program with this customer valued at approximately $18 million, and we have a robust pipeline with them heading into 2027. We have also secured five new programs at another customer, a large hyperscaler. To give you a sense of the momentum and success we've already built with this customer, over the trailing 12 months we generated approximately $3.1 million of revenue with them. On the strength of these new programs, we now forecast that this customer may contribute more than $20 million in revenues in 2027, and beyond that there is significant additional pipeline under discussion; we also added five new logos in Q3, including model builders and an enterprise data platform. In recent weeks, we signed an enterprise-focused foundation model company. Beyond that, we are holding promising conversations with a leading frontier model developer about a major expansion, and we have just begun discussions with a leading social media company. What we’re winning demonstrates how tightly we continue to align with what builders of frontier models require, even as those needs grow more sophisticated in virtually every dimension. In just the last 18 months, the emphasis in model training has shifted from evaluations built on humans to agentic traces within sandbox replicas of real-world settings—frequently called RL gyms—and from synthetic data to renewed appetite for real-world data. Each of those shifts has broadened what is possible and what is required, and each has created new opportunities for us. Our significant new wins are also enabling meaningful customer diversification. One trend is worth noting: several times, researchers we’ve collaborated with at the leading labs have departed for other labs or for well-funded startups, and after getting settled, they reach out to us; they know our capabilities and our team, they took the relationship with them; that has worked in our favor, and we believe it will continue to help us grow our pipeline faster than SG&A does. Today we are raising our 2026 guidance from 40% to 42%. Given how much business we are winning, we could have increased it further, but in mid-October our historically largest customer decided to end one program with us. Luckily, we expect new programs at that customer to replace that revenue by Q1, although we anticipate it will cut Q4 revenue by about $12 million. Again, as we’ve noted previously, I want to emphasize that among our large customers, at any moment some programs are concluding, others are starting, and still others are being discussed. That reflects the dynamic nature of our business and shows it operating as intended. I’d like to take a few minutes now to go deeper on six specific growth drivers that we expect will fuel our growth over the next 12 months. I’ll start with the first driver: physical AI. Capital is moving into physical AI faster than into any category since the LLM boom began. Physical AI is sometimes called embodied intelligence or simply AI robotics. In the first half of 2026, worldwide venture funding in this space totaled $47.4 billion across 521 deals. That exceeds the combined total invested in physical AI from 2022 through 2024. The bets center on robot foundation models and on companies seeking to build consumer and industrial robotics. New research confirms that robotics foundation models improve with data the same way LLMs do. Because no internet-scale corpus of physical interaction exists, the data must be generated via teleoperation, egocentric human video, and simulation—all work we’ve begun. As models move from lab demos to deployment in factories, warehouses, and homes, we expect demand to grow from one-off pre-training corpora to continuous task-specific data for fine-tuning and evaluation. As a data provider, we become recurring infrastructure, with per-share figures of $0.17 and $0.29 underscoring the momentum. We’ve put a lot of effort into capturing this opportunity, and we’ve made meaningful headway in robotic data collection — spanning egocentric data, UMI gripper data, and motion capture data — as well as in constructing large-scale AI-assisted action annotations for VLM training and in carrying out robot safety testing and evaluation. Our physical AI data business is expanding considerably. We operate extensive egocentric data collection programs, stereo camera headset capture inside homes across the United States and Latin America, and six-camera headset capture of intricate industrial tasks at hundreds of businesses. Our customer is a top humanoid and frontier robotics lab with substantial ambitions. One humanoid customer we brought on recently stands as a potentially multi-million dollar opportunity for us in the coming year, and we see $0.17 to $0.29 in value per unit flowing from engagements of this kind. In September, we introduced our cutting-edge Vicon motion capture lab to record high-fidelity 3D human motion for world model training and humanoid robot training, and we’ve shown an end-to-end real-to-sim-to-real pipeline — taking captured human motion into Isaac Sim, retargeting it onto humanoid kinematics, and operating our own Unitree G1. Past collection, we’re quickly turning into a trusted annotation partner for physical AI labs, having delivered more than 620,000 robot-executable labeling jobs so far this year; this is a fresh focus area for us that I anticipate will turn into a major growth driver for 2027 as robots shift from labs into production.

Rahul Singhal

executive
#3

We've kicked off an annotation pilot alongside a leading robotics lab, and should it expand, that might turn into a meaningful opportunity; we also landed a pilot with a leading frontier lab. If that pilot turns out well, we anticipate it could bring in up to $7 million in run-rate revenue during 2027. Beyond that, we're working with a large technology firm on safety red-teaming data for robot foundation models. Now I'll move on to driver number two: RL environments for personal and knowledge work agents. As I mentioned earlier, there's a shift toward training within RL gyms. The reasoning here is simple: an agent gains knowledge by taking actions and watching what results, and labs are unable to let agents learn on live systems for safety and privacy reasons, plus many actions can't be undone. Training happens inside a sandbox replica, giving the agent room to explore, err, and reset thousands of times. Innodata is now producing at scale the kind of high-quality RL environments that we expect will lead model training for the foreseeable future. We saw this shift coming, and we've built a robust research group with substantial know-how in environment design and reinforcement learning. Our particular emphasis is on a pair of use cases—personal agents and knowledge work agents—and we aim to be the leader in both. Regarding personal agents, we've spent a considerable stretch collaborating with one of our biggest customers on personalization for its consumer agents. Anyone who tracked the consumer agent releases this year witnessed how that degree of personalization performs when hundreds of millions of users have it. We contributed to the effort that made it happen. The way those kinds of agents figure out what someone genuinely wants is grounded in our data, our environments, and our methodologies. In our view, personal agents represent the coming competitive battleground, and the companies that come out ahead will be those whose agents genuinely understand you and can customize experiences around you. The advanced RL gyms that capture that level of depth serve as a major differentiator in that race, and according to our customers, the methodologies and tooling we've built for pulling out the data behind personal agents are unlike anything they've encountered. Over the next several quarters, we anticipate constructing additional personalization agent gyms of this sort. For enterprise customers, the competition among model developers has moved away from coding agents and toward agents handling routine knowledge tasks. These agents operate across desktop and browser workflows involving multiple systems, begin with unstructured inputs, and call for judgment at each stage. Today's models are expected to take on the expert-level work that rules-based automation could never handle. To train these models, we're constructing advanced RL gyms, and we're also developing the components that go inside them. This effort is still in its early stages, and we think our team and our technology leave us well positioned, with every intention of leading in the training of this new generation of agents. Now I'll turn to driver number three: trust and safety. Let me begin with alignment—what we mean by that is the effort to ensure AI systems act safely and consistently with human intent and policy. There are a couple of quite simple reasons why we think this work is only becoming more necessary. First, as AI shifts from single-shot responses to reasoning across multiple steps, and from chatbot assistants toward autonomous agents, the price of an alignment failure goes up and assessing safety gets harder. Second, models are now being put to use in sensitive fields like health and finance, where a failure has genuine consequences. As everyone has witnessed lately, the more capable models become, the larger the risks that come with them. Labs are reacting by directing more of their efforts toward alignment post-training, and in particular toward human-in-the-loop data. For Innodata, that opens up a meaningful opportunity. Our alignment work at present covers a number of high-stakes areas. For instance, we are in the process of finalizing a new statement of work worth roughly $6 million in potential revenue, covering chemical, biological, radiological, nuclear, and explosive—CBRNE—risk evaluation within agent safety work. This involves a major hyperscaler. As we mentioned during our Q1 call, that company chose us as a worldwide trust and safety partner. Beyond that, the customer has verbally committed to a program where we'll evaluate the agentic safety of certain platforms they operate. In our view, that program might bring in roughly $2.5 million in revenue next year. We also have pilots running across multiple mission-critical safety initiatives spanning health, cybersecurity, and robotics. We believe these pilots will pan out and that the programs will expand substantially. Several of these pilot programs rely on what we call temptation gyms—settings built to reveal if a model will pursue a valid objective by improper methods, or take shortcuts in the process. I'll discuss those further in a moment. When it comes to trust and safety, much like with our agentic and RL efforts, we're gaining ground thanks to how customers see us: as a trusted advisor, a rigorous applied research outfit, and a reliable delivery partner; that combination keeps us entrenched within these accounts, and it lets us help shape demand instead of just reacting to it. I'll move on now to driver number four: end-to-end training pipelines running from pre-training all the way to agentic AI. Here's the context: the open web is being depleted. Epoch AI projects that labs will use up the available supply of public human-written text at some point between 2026 and 2032. These days every stage of model development relies on data that has to be deliberately created, including specialized pre-training corpora along with RL environments and evaluations. I think Innodata is one of only a handful of firms worldwide that handles this kind of work from start to finish, and we bring serious domain knowledge to each stage of it. When it comes to pre-training, we don't just rely on the usual web-based sourcing — we obtain distinctive, hard-to-find datasets instead. We're seeing interest in that from multiple quarters. The largest opportunity in front of us is a possible program we're currently talking about with a top-tier frontier lab. The goal is for it to span many different data types, with a monthly run rate reaching the millions of dollars. Take video: it's a good illustration of data that can't just be scraped at the quality and with the rights that labs require; we're currently in talks with one customer about a high-end cinematic video opportunity worth roughly $2 million. Charts are another focus area for us. Complex charts and graph datasets are being built by our teams to bolster visual reasoning in models, and we've brought on several new customers for these datasets. We're also growing our real-world business data licensing practice. Frontier labs want raw, real-world data, along with agentic trajectories with verifiable outcomes that are needed to train their models. Because our work with customer engineering groups is so hands-on, we know well ahead of time which data sets will produce the performance gains they want—and that expertise also makes us the obvious choice for data owners, who rely on us to match their information with the problems it can address. Human evaluation continues to move from broad generalist crowds toward specialized expert pools; we just secured a new agreement valued at roughly $1.2 million with a top worldwide supplier of legal and regulatory solutions.

Rahul Singhal

executive
#4

With this agreement, our legal specialists are going to deliver human-in-the-loop validation as well as the expert feedback that gets used to fine-tune the customer's specialized legal AI models. This win shows how strong we are in high-stakes workflows and sets up a scalable blueprint for high-margin, multi-year growth in enterprise AI; we're bringing that same expert model to agentic AI, with knowledge workers building agentic gyms — environments where agents do real computer use and knowledge work tasks against realistic data. As agents handle more autonomous work, labs need to know not just whether an agent can finish a task, but whether it finishes it honestly. Our temptation gyms are built precisely for that. The setups here are built on purpose to present opportunities for taking shortcuts, manipulating a reward, making up an outcome, or misstating the work actually performed, and they come with instrumentation that lets labs observe whether an agent seizes those opportunities. That way, labs can detect conduct which, in real-world deployment, would add up to misrepresentation or fraud. Integrity testing of this sort is turning into a necessity, and it calls for both domain knowledge and the ability to build environments. Temptation gyms represent a new category, and we think that moving early gives us a capability that will set us apart. Now I'll turn to driver number five: enterprise AI. As we've noted previously, deploying generative AI models across the enterprise is a very sizable opportunity for us. Over the past three quarters we've been developing this practice, and the momentum keeps building. In the most recent quarter alone, we brought in roughly $5 million in revenue run rate from agentic implementation work for one large customer. We also kicked off two fresh enterprise relationships—one with a global technology hardware company and another with a major financial services firm. With a social media customer we've worked with since last year, we broadened the engagement to around $2.5 million annually. These wins, in our view, provide a solid base from which to accelerate in 2027. Our partnership strategy has three tiers, designed so we can support enterprise customers at the foundation model layer, the data layer, and the infrastructure layer. Working alongside Anthropic, which is a top foundation model provider, we're able to push enterprise agentic solutions all the way into production. Teaming up with Databricks and Snowflake, both leading data platforms, lets us create the AI-ready data foundations that any successful deployment relies on. Through our Google Cloud partnership, we can deliver and scale solutions atop enterprise-grade cloud infrastructure. On top of that, these channel partnerships open up extra growth avenues and joint opportunities; we feel our background in training frontier models gives us a distinct edge in grasping how to craft golden datasets, benchmark models, develop custom rubrics, and carry out end-to-end data engineering at scale—yielding robust, production-ready solutions for enterprises, and handing us an advantage traditional systems integrators don't have. Finally, let me turn to driver number six, our federal practice, which we started roughly twelve months back. In that time, it has established a foothold in government AI initiatives that build on one another with every award. Over the first year, this produced two programs, along with a third secured through our prime partners. The opportunity we described in May as a finalist — an AI development environment for the Department of War — has now entered phase-one delivery, with a production decision expected in the coming months. We also won an award for advanced data labeling on a U.S. Navy autonomy program, which is the initial phase of a competitive down-select, chosen because of the computer vision capability we showed the government this year; additionally, we have been identified for a subcontractor role on a second multi-year Department of War program. All of these are staged, with the bigger award determined once the present phase performs, and that is precisely where we have opted to compete. We hold a verbal win for a single contract worth $5 million annually across five years, with signing anticipated before the fourth quarter ends. Our hope is to sustain momentum throughout all programs. We close out the year with genuine momentum, a well-defined strategy, and tailwinds coming from each of the six growth drivers I have just taken you through. Never have I felt more confident regarding where Innodata stands. The opportunities ahead of us are large, and they sit close at hand; we are putting investment behind capturing them, and we expect those investments to pay back fast and to establish durable advantage. The coming 12 to 18 months will center on execution, and that is where Jayant and I are concentrating. I will now hand the call to Jayant, our CFO, so he can share his perspectives on the business and our Q3 financial results.

Jayant Chauhan

executive
#5

Thank you, Rahul, and good afternoon to all of you. I'll walk you through what we delivered in the third quarter, and I'll also cover our thinking on investments and how we allocate capital. Revenue for Q3 2026 reached $95.6 million, representing 53% growth compared with the same period last year and a 4% increase from the prior quarter — that makes 13 quarters in a row of year-over-year expansion. This exceeded the analyst consensus figure by $12.3 million, a 15% beat. Adjusted gross profit reached $38.3 million, translating into an adjusted gross margin of 40%, which lands exactly at our 40% target. Program mix is what determines gross margin in any particular quarter. Adjusted EBITDA was $17.7 million, equal to 19% of revenue, and that's up 9% year-over-year. This surpassed the $15.4 million consensus by roughly 15%. We posted net income of $10.3 million, which is a 24% increase over the $8.3 million reported in Q3 of last year. On a fully diluted basis, earnings per share came in at 29 cents, while consensus stood at 17 cents — roughly a 70% beat. Our effective tax rate for the year to date was about 12%, versus our long-term target range of 23% to 25%. The favorable tax rate resulted from tax benefits we recognized during this quarter. Moving now to the balance sheet: we ended the quarter with cash and cash equivalents of roughly $164 million. Excluding about $10.3 million in customer prepayments, which function as a pass-through, cash totaled $153.7 million. That represents an increase of $19.3 million compared with $134.4 million on the same basis at the close of Q2. There's zero debt on our books. The $50 million credit facility with Wells Fargo hasn't been tapped at all. Our ATM program also went unused this quarter. Now that I've had a full quarter in the role, I'd like to share a few thoughts about investments and how we allocate capital. In light of the opportunities Rahul laid out, we'll continue putting money into areas we think will turn into significant growth drivers by 2027 and afterward. The plan is to fund those investments out of operating cash flow, and we anticipate they'll pay back fast. Turning to capital allocation: sitting on roughly $154 million in cash, net of customer prepayments, and carrying no debt, we enjoy genuine flexibility. Organic investment in those growth drivers comes first. Past that, any M&A we pursue will be chosen carefully, limited to deals we think could speed up our strategy, and we plan to fund those investments while keeping the balance sheet strong. In summary, Q3 brought record revenue, adjusted EBITDA that beat expectations at our 40% gross margin goal, and $19.3 million in underlying cash generation, even as we kept putting money into the areas we expect to power our next stage of growth. That said, I'll hand things back to the operator. Operator, we're ready to take questions.

George Sutton

analyst
#6

Thanks, guys, and I appreciate all the details. Could you walk us through how your labor pool process operates? You've got a program that comes to an end, and this is a constant occurrence — normally, I'd imagine you'd want to move those people into other roles right away. You said Q1 might involve a similar type of deployment. Do those roles call for different skill sets? Do you go back and hire from that pool again? Could you paint a picture for us there?

Rahul Singhal

executive
#7

Hi, George. Thanks for asking. The way these programs begin and conclude, I'd say, depends on what kinds of tasks the frontier model builders are focused on. At times the skills being worked on are redeployable, and at other times those skills aren't redeployable. Another point worth noting is that when these various frontier model builders are at different stages of their training and are training the various tasks, it comes down to what is available and what can be reused at that moment from a labor pool standpoint. Regarding Q1, I'd say those are different programs that we expect to start.

George Sutton

analyst
#8

Got it. You said something interesting while discussing physical AI — the robotics work in particular — about aiming to become recurring infrastructure, which is a pretty big departure from the usual programs you've historically handled. Could you expand on that a bit more?

Rahul Singhal

executive
#9

Sure. Physical AI is something we're really enthusiastic about. When you look at the money flowing into physical AI, my belief is that it will surpass the investment that went into LLM. One of the most fascinating aspects of physical AI, or embodied intelligence, is that you cannot scrape data off the web the way the LLMs did at the start. There is no pre-training data or pre-training corpora available. These models get trained by first using what's called egocentric data, where cameras are used to gather data across various tasks. A task might be gathering data for loading a dishwasher, or gathering data for folding your laundry, right? That's egocentric data; then there are what are called UMI datasets, and then there are what are called motion capture datasets; these are distinct modalities, distinct approaches, distinct technologies and techniques in use, and each one becomes progressively much more interesting, much more complex, and the models begin to improve. That's where world models come into the picture. Since there's no single layer of physical corpora that anyone can draw on, you must constantly collect, create, and annotate the data so these robots can actually execute around it. To me, that's a highly interesting and exciting opportunity for us.

George Sutton

analyst
#10

Great. So just one question for Jayant. Regarding the Q4 guidance, historically your practice has been that the guidance only reflects business you already have booked. Should more business come in during the quarter, that generates some potential upside. Could you walk through the thinking behind the Q4 guide, and also touch on the pipeline of opportunities and roughly the pace of deal cycles that might deliver additional revenues in Q4?

Jayant Chauhan

executive
#11

Hi George. Thanks for asking. As for how we approach guidance, that method remains in line with what we've done previously. We stay prudent about it, and you're correct that generally we include only revenue from active programs, new awards, or smaller pipeline opportunities where we truly have a clear line of sight. We don't include large pipeline opportunities until they've been awarded, and until we know when that revenue will be recognized. We remain consistent with that approach right now and going forward as well. On your second question about what we're seeing in the pipeline, I think Rahul discussed many of those opportunities that we're seeing today. As for guidance for 2027, as you know, we plan to announce it together with our Q4 results, and you'll hear more on that over the next quarter.

George Sutton

analyst
#12

Thanks.

Allen Klee

analyst
#13

Yeah, hello. You mentioned that your adjusted gross margin reached the 40% plus target for the quarter. During the two quarters before that, it somewhat over-earned — it came in above — so could you give us a bit more detail on what happened with mix, and is there any reason you believe it may move in a particular direction ahead? Thanks.

Rahul Singhal

executive
#14

Hi, Allen. I think, as we've said all along, our gross margin target is 40%, and the margin mix across those programs shifts from quarter to quarter; we also often run a lot of programs that we believe could be significant growth drivers for us in the quarters that follow, or that build capabilities, which we typically take on at a lower margin. We have consistently guided to a 40% gross margin, and we'll keep guiding to a 40% gross margin, because I think this business requires us to make sure we're investing in various programs, building our capabilities, and investing ahead in a way that lets us keep leading the market.

Allen Klee

analyst
#15

Thank you. Now regarding the project that was discontinued in October — and I believe you mentioned it would have a $12 million impact in the fourth quarter, but that same former largest customer is going to add that amount additionally in the first quarter — as for the margins of the business that was shut down, is there a way to think about whether that is materially different from the margins of the new business?

Jayant Chauhan

executive
#16

Oh, hi. Let me handle that one, Rahul. Look, since we don't really provide guidance or discuss margin at the project level, I would simply point to the fact that we'll address 2027 and what that looks like when we report our Q4 results.

Allen Klee

analyst
#17

Alright, so the final point on capital allocation — when you mentioned organic growth, was your intent to suggest that the sum we should expect you to put toward organic growth, in order to fund it, is going to rise compared with prior levels, or what message were you hoping to convey to us? Thanks.

Jayant Chauhan

executive
#18

I'll handle that one, Rahul. Allen, yes — as we've previously discussed with you folks and with our investors too. There have been plenty of questions regarding our thinking on capital allocation, particularly in light of the strength of the balance sheet. Essentially, I just want to make clear that organic investments remain our top priority, and that is where our focus will be as we put money behind the different growth drivers and opportunities that Rahul described. That stays the priority. Should we, opportunistically, come across M&A opportunities capable of speeding up our growth strategy and generating value for shareholders, we'll keep examining those as well, though organic investments come first.

Allen Klee

analyst
#19

Thanks.

Hamed Khorsand

analyst
#20

Hello. I have just a single question regarding this big customer shifting one project over to a different one. Was that other project awarded to you, or do you have confirmation that it's taking place in Q1?

Rahul Singhal

executive
#21

Hamed, we don't discuss precise timing. We expect to win additional work, and we're at late stages on several programs that we expect will assist us in closing that gap. I think the way to look at it is that programs begin and end. New programs are beginning. We're engaged with multiple stakeholders and multiple business units inside our biggest customer, and this business moves extremely, extremely quickly. We expect to close the gap.

Hamed Khorsand

analyst
#22

So it seemed like there's a new big customer on your end. How much of revenue does that account for?

Jayant Chauhan

executive
#23

Are you referring to the customer Rahul mentioned in his remarks earlier — the third-largest one? Yes. Right now, we expect roughly $20 million in revenue from that customer over the coming year.

Hamed Khorsand

analyst
#24

All right. And regarding what you're observing with your biggest customer, has anything shifted there, or has it stayed fairly steady where revenue goes?

Rahul Singhal

executive
#25

From what we can tell with our biggest customer, this market is shifting a lot quicker, and the investments head in many different directions. Customers lean toward putting money into areas such as VLMs — that's vision learning models — they're putting money into robotics, putting money into LLMs, putting money into agents, correct? The business moves quickly, and we engage with many different business units throughout the full range inside the biggest customers, attempting to serve their requirements and collaborate with them. We devote time alongside them on reinforcement learning gyms, where we're building state-of-the-art gyms that let agents be trained, offering our know-how and our thought leadership to help them try to dominate that market. I believe various areas are where we keep collaborating with them across many different programs.

Hamed Khorsand

analyst
#26

Thank you.

Rahul Singhal

executive
#27

Thanks, operator. In closing, Q3 2026 marked yet another quarter of record revenue for Innodata and another beat on every measure; we posted 53% revenue growth—our 13th straight quarter of year-over-year gains—adjusted gross margin of 40%, adjusted EBITDA above consensus, and solid net income, all while we kept investing in the research and innovation fueling this business. Diversification also continued. A client accounting for under 5% of revenue a year back now ranks as our biggest customer, so we have two customers of comparable size. Another major hyperscaler, one that generated just $3.1 million of revenue with us over the trailing 12 months, is on course to exceed $20 million as a customer next year. We also landed notable new customers spanning frontier labs, enterprise data platforms, and robotics. We are lifting our guidance to 42% or more growth for 2026. Considering our wins, that increase would have been far larger were it not for an air pocket in Q4 revenue tied to our historically largest customer, which we expect to be filled by Q1. Most importantly, we see ourselves sitting at the epicenter of AI's most significant transition today: the shift toward the next S-curve of agents, reinforcement learning, and embodied intelligence—and at the heart of the trust, safety, and alignment efforts that will decide whether that shift succeeds. Those are chiefly data problems, and they fall squarely within our wheelhouse. My belief is that 2026 will end as a strong year for Innodata, and 2027 will be just as—or even more—exciting; leading this company is a privilege. I thank our team and customers worldwide, and all of you, for staying on this journey with us. We look forward to speaking with you next quarter. Thank you.

Operator

operator
#28

This concludes today's call. Thank you for attending. You may now disconnect.

Read the full transcript via the API

You're viewing the first half of this call. Get the complete Innodata Inc. transcript — plus 132,000+ transcripts from 7,000+ companies, speaker segments, AI summaries and full-text search — through the EarningsCalls.dev API.

Get the API View API docs →

This call discussed

For developers and AI pipelines

Programmatic access to Innodata Inc. earnings transcripts and 132,000+ others is available through the EarningsCalls.dev REST API. Plans from $39.99/month — full transcripts, speaker segments, full-text search, and the recently-added /api/v1/transcripts/recent polling endpoint for ETL pipelines.

More tools for stock research

OptionsBell: an unusual options activity scanner that tracks options flow across all US stocks and emails you when large trades appear in Innodata Inc. or any ticker on your watchlist.

StockMarketScan: a stock screener with alerts: 24 pre-built screens refreshed every 5 minutes, watchlists and email notifications. Both offer API and MCP access and fair monthly pricing.