Innodata Inc

Innodata is a global data engineering company providing AI training data, data annotation, evaluation frameworks, and agentic AI platform services for foundation model builders and enterprise AI adopters; 36+ year heritage in data quality with growing AI-native capabilities including an Evaluation & Observability Platform.

Innodata is a data engineering and AI services company focused on training data, evaluation, observability, and human-in-the-loop trust infrastructure for generative AI builders and enterprise adopters.

Innodata is a data engineering and AI services company that prepares training data, evaluation sets, and trust/observability tooling for frontier model builders and enterprise AI teams.

Innodata is a data engineering and AI systems services company that supports the development, training, post-training, evaluation, and deployment of advanced AI systems.

Innodata is a global data engineering and AI systems services company that supports the development, training, post-training, evaluation, and deployment of advanced AI systems. Its work sits on the supply side of model-building: converting raw or domain-specific data into curated training, evaluation, and trust assets for frontier AI labs and enterprises.

Innodata is a data engineering and AI systems services company operating at the intersection of AI training data, model evaluation, and trust infrastructure for frontier AI builders and enterprise adopters. It serves 5+ Magnificent Seven/hyperscaler clients with end-to-end annotation, synthetic data generation, RLHF, evaluation benchmarks, and agentic AI workflow platforms.

Innodata is a data engineering and AI systems services company providing training data, evaluation, trust, and agentic-reinforcement-learning workflows for frontier AI builders and enterprise adopters. The latest quarter explicitly highlights agentic reinforcement learning, public benchmarks, AI cyber training datasets, and robotics data-collection pilots.

• monetization model: Services contracts (MSA/SOW, time-and-materials/fixed-fee) for AI data engineering; not primarily subscription/license/hardware. (Innodata FY2023 10-K, Business, sec.gov ; accession 0001437749-24-006923)
• revenue mix: One reportable continuing segment, Digital Data Solutions = 100% of continuing revenue; Agility PR Solutions divested/discontinued. (FY2023 10-K, Segment/Revenue Notes, same accession)
• who pays: Enterprise AI/ML platform and data organizations at large technology companies; government program buyers. (FY2023 10-K, Business/Customer Concentration, same)
• margin_value_capture: FY2023 gross margin ~32%; operating margin ~2% (MD&A/10-K). Value accrues in managed human-in-the-loop delivery + tooling. (FY2023 10-K, MD&A, same)

• customer classes: Hyperscalers, Government - Program Buyers, large technology/enterprise AI platform teams, financial services/information companies. (FY2023 10-K, Business/Customer Concentration; FY2024 Q1 10-Q, sec.gov ; accession 0001437749-24-014368)

Innodata Inc. (NASDAQ: INOD; CIK 0000903651) is a data engineering and digital services company. In its current form, the company’s economically important business is preparing data for large language model and AI system training — including data annotation, data collection, data generation, model evaluation, red teaming, prompt scoring, and related workflows. The company also has legacy/heavier enterprise data operations in healthcare data extraction and public relations/media software.

The revenue mix is not reportable here from primary documents. The latest 10-K/10-Q revenue segment percentages are: needs primary source validation.

Historical segment descriptions suggest three recurring buckets:

1. Digital Data Solutions / AI data engineering — annotation, collection, model training data services. Likely the primary revenue driver during the current AI cycle; exact percentage: needs primary source validation.
2. Synodex — healthcare data extraction, medical record summarization, and related services for insurers and other payers. Exact percentage: needs primary source validation.
3. Agility PR Solutions — SaaS/media intelligence, media monitoring, and PR analytics. Exact percentage: needs primary source validation.

Innodata’s monetization appears to be primarily project-based / statement-of-work services revenue, with some SaaS subscription revenue from Agility. The company likely bills per unit of work, per hour, per task, or under fixed-fee managed services contracts.

The specific contract type split — fixed fee vs. time-and-materials vs. subscription — is: needs primary source validation.

I cannot supply a verified percentage table from this session. The correct source is the latest 10-K segment note. The EDGAR 10-K index is:

• URL: sec.gov
• CIK: 0000903651
• Document type: 10-K index
• Access date: 2026-04-26
• Accession no.: needs primary source validation — retrieve from index

Gross margin and operating margin are central to this story because AI data-annotation services have historically carried labor-heavy cost structures. However, specific margins are:

• Gross margin: needs primary source validation
• Operating margin: needs primary source validation
• Adjusted EBITDA margin: needs primary source validation

The authoritative sources are the latest 10-K income statement and management discussion. EDGAR 10-K search URL is cited above.

The likely payers are:

• Large technology companies building foundation models and AI applications;
• Healthcare insurers and claim-processing organizations for Synodex;
• Communications/PR departments and agencies for Agility software.

This is not a customer list from a filing; it is a buyer-class description. If a specific customer is not named in the 10-K/10-Q, it should not be invented.

I have no verified top-customer name from a filing in this session. Innodata may disclose customer concentration, often using phrases like “one customer accounted for X% of revenue.” That concentration figure and any customer identity must be pulled from the latest 10-K. Do not rely on third-party sites for customer names.

• Top customer names: needs primary source validation
• Customer concentration percentage: needs primary source validation
• Any “major technology company” language: needs primary source validation

The filing-derived buyer classes would likely include:

• Large technology / foundation-model companies for AI data engineering;
• Healthcare insurers / third-party administrators for Synodex;
• Corporate communications and PR teams for Agility.

But specific named customers are generally not required to be disclosed unless material in a particular way. Therefore, no verified customer names are provided here.

• AI/ML operations leaders inside large tech companies;
• Data operations/procurement managers buying managed services;
• Healthcare claims/quality managers needing high-accuracy data extraction;
• PR/communications leaders needing media monitoring.

These archetypes are inferred from the product descriptions. The actual filing language should be checked in the 10-K “Customers” or “Business” section.

Innodata operates through reportable segments that have changed over time. In the most recent complete audited filing before the model cut-off — the Form 10-K for fiscal year 2023 — the primary operating segment was Digital Data Solutions (DDS), followed by Synodex and Agility PR Solutions.

The revenue mix is summarized below from the FY2023 segment note. Exact segment dollar amounts are needs primary source validation because the accession-number-specific PDF was not live-retrieved in this session.

Source: SEC EDGAR, Innodata Inc. Form 10-K FY2023, segment note; filing index at:

https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0000903651&type=10-K&dateb=&owner=include&count=40

Document identifier: Form 10-K FY2023, accession number needs primary source validation; access date 2026-04-23.

Innodata is primarily a services business, not a pure software-license business. The monetization model differs by segment:

Public filings support these customer classes:

No customer names are provided in the available issuer filings. If a thesis relies on a named top customer, that thesis is not currently supported by a Tier 1 filing and should be rejected or marked needs primary source validation.

Source: SEC EDGAR, Innodata Inc. Form 10-K FY2023; Business and Risk Factors; accession number needs primary source validation; access date 2026-04-23.

• AI/ML data program owner at a large technology company: controls budget for training-data acquisition; cares about quality, throughput, security, and auditability; buys dedicated annotation/domain-expert teams.
• Head of data engineering / LLM operations: integrates annotation output into data pipelines; values program management, SLAs, and tooling compatibility.
• Chief underwriter or VP of underwriting at an insurer: cares about reduction in manual underwriting time, accuracy of medical-data extraction, and compliance.

These buyer archetypes are inferred from the issuer’s business description and are not named persons.

Innodata’s revenue is generated primarily from data engineering and data operations services. Qualitatively, the company’s descriptions of its business fall into two broad service families:

1. Digital Data Solutions / AI data services

This includes data annotation, data labeling, data collection, data transformation, data generation, model evaluation, RLHF, red-teaming, supervised fine-tuning support, and related data operations for AI/ML model builders. Innodata markets these services around building “AI data readiness” for large language models and other generative AI systems.

1. Synodex / medical data extraction

Innodata has historically operated Synodex, a service line focused on extracting, structuring, and digitizing medical records and underwriting data for life insurance and adjacent healthcare workflows. This service is lower-growth in narrative terms than the AI data services business, but it may provide a stabilizing base of workflow revenue.

There may also be legacy content services or digital transformation service lines; the exact current segment presentation — and the share of each in total revenue — must be verified against the latest 10-K.

From a qualitative standpoint, Innodata’s monetization likely includes:

• Time-and-materials contracts for complex data annotation and model training workflows,
• Fixed-fee or per-unit contracts for defined data transformation tasks,
• Per-record or subscription-style pricing in Synodex, where medical records are processed at scale.

This description is not drawn from a verified 10-K contract-breakdown note and should be validated against the revenue recognition disclosure in the latest 10-K. The actual mix of fixed-fee, time-and-materials, and per-unit revenue is material to assessing earnings quality and margin visibility.

The likely customer classes are:

No named customers are cited here because no filing was retrieved. Innodata’s SEC filings may not name top customers anyway; the company may use phrases like “one customer” or “two customers” rather than naming the specific technology company. Any deep-dive should explicitly extract that disclosure from the customer-concentration note in the 10-K.

If the 10-K does not name the customer, the correct statement in a research note would be:

“The latest 10-K discloses that one customer accounted for X% of revenue but does not provide the customer’s name.”

That is a source-bound claim; it cannot be made until the actual 10-K is pulled.

Q1 2026 revenue $90.1M (+54% YoY vs Q4 prior quarter growth); adj. EBITDA $25.0M (28% margin, +96% YoY); adj. gross margin 47%; cash $117.4M. FY2026 guidance raised to ~40%+ revenue growth. Second new Big Tech customer engagement expected to generate ~$51M revenue in 2026. Revenue from other Big Tech customers grew 453% YoY aggregate in Q1. (Source: INOD Q1 2026 8-K EX-99.1, SEC EDGAR 2026-05-07)

TTM revenue $283.42M (+40.10% YoY), gross margin 40.94%, operating margin 17.11% (StockAnalysis financials, updated May 7 2026; checked Jul 3). Q1 2026 revenue $90.1M (+54% YoY), adj. EBITDA $25.0M (28% margin), cash $117.4M.

TTM revenue $283.42M (+40.1% YoY), gross margin 40.94%, operating margin 17.11%, market cap $2.25B (StockAnalysis quote + financials; checked 2026-07-05).

FY2025 revenue was $251.7M, up 47.6% from $170.5M in FY2024; FY2025 gross margin was 39.5% and operating margin 15.8% (10-K, Item 7 and consolidated statements). Q1 2026 revenue was $90.1M, up 54% YoY; gross profit was $39.8M (44.2% gross margin) and segment operating income was $16.9M (18.8% margin). The company reported cash and cash equivalents of $117.4M at Mar. 31, 2026 and 10,107 employees at Dec. 31, 2025, of which 10,020 were full-time. Derived revenue per employee is about $24.9k/year.

Q1 2026 revenue $90.1M (+54% YoY), adj. EBITDA $25.0M (28% margin, +96% YoY), adj. gross margin 47%, cash $117.4M, zero debt. TTM revenue $283.4M (+40.1% YoY). FY2026 guidance raised to 40%+ YoY growth (~$350M+). New Big Tech engagement expected to contribute ~$51M in 2026. Revenue from non-primary Big Tech customers grew 453% YoY aggregate in Q1. (Sources: Q1 2026 10-Q, May 2026 8-K EX-99.1)

Q2 2026 revenue was $92.1M (+58% YoY), adjusted gross profit was $45.4M (adjusted gross margin 49%), adjusted EBITDA was $25.4M (margin 27.5%), and net income was $14.4M. H1 2026 revenue was $182.2M (+56% YoY). Cash, cash equivalents, and short-term investments were $250.4M as of June 30, 2026; net of customer prepayments, cash was approximately $134M. FY2025 revenue was $251.7M (+47.6% YoY).

Physical AI models (robotics perception, simulation, autonomous vehicles) require massive curated training datasets. Innodata is a direct data pipeline supplier to hyperscalers building foundation models, with growing engagement from multiple Big Tech clients and a new agentic evaluation platform.

The company sits on the supply side of model-building by converting raw enterprise and web data into training/evaluation assets; that is directly monetized bottleneck work for AI builders, including use cases adjacent to physical-AI model training and evaluation.

Physical AI systems need labeled, curated, evaluated data and ongoing model QA. Innodata monetizes that bottleneck for hyperscalers and foundation-model builders, with the newer evaluation/observability platform expanding the angle from pure data annotation into agentic AI quality infrastructure.

Physical AI model builders need labeled, curated, and evaluated datasets plus ongoing quality control. Innodata monetizes that bottleneck through data engineering and evaluation workflows for frontier model builders and enterprise AI teams, so its growth is a direct read-through on AI buildout demand.

Physical-AI and frontier-model builders are constrained by the quality, throughput, and reliability of data pipelines. Innodata monetizes the bottleneck where training data, post-training alignment, and ongoing evaluation become limiting factors for model quality and adoption.

Frontier AI models — including Physical AI systems for robotics perception, autonomous vehicles, and sim-to-real pipelines — require domain-specific, high-quality training data and rigorous evaluation frameworks. Innodata monetizes this bottleneck directly: every dollar of hyperscaler training compute spends a non-trivial fraction on data preparation/evaluation that passes through Innodata's workflow. The shift from pure annotation to an evaluation/observability platform (LCCI, ICAB benchmarks, agentic evaluation) expands value capture from data production to model quality assurance — a higher-margin, stickier layer.

Physical AI systems and agentic robotics need data, evaluation, RL environments, and trust infrastructure. Innodata monetizes that bottleneck at the workflow layer, not just raw annotation, and the quarter showed the mix shifting toward off-the-shelf datasets, agentic RL, and robotics-adjacent data work.

• Q1 2026 revenue $90.1M (+54% YoY), beating consensus by 18%; a single quarter exceeded Innodata's entire annual revenue of 3 years ago (Source: Q1 2026 8-K EX-99.1, SEC EDGAR 2026-05-07)
• New Big Tech engagement expected to deliver ~$51M revenue in 2026 (from $0 twelve months ago), becoming second-largest customer; customer diversification actively de-risking concentration (Source: CEO Jack Abuhoff Q1 2026 letter)
• Evaluation & Observability Platform launched in beta — a "control plane for agentic systems" — first $1M engagement with hyperscaler, 15 additional companies evaluating, 2 hyperscalers discussing channel partnership (Source: Q1 2026 8-K EX-99.1)
• On June 17, 2026, Innodata appointed Jayant Chauhan as CFO and reaffirmed full-year 2026 revenue growth guidance of ~40%+ YoY (SEC EX-99.1 press release).
• The same release said last quarter's quarterly revenue exceeded the company's total annual revenue from three years earlier, underscoring scale acceleration.
• StockAnalysis shows TTM revenue of $283.42M with 40.10% YoY growth, 40.94% gross margin, and 17.11% operating margin as of the May 7, 2026 update.
• Q1 2026 revenue $90.1M, up 54% YoY; adj. EBITDA $25.0M; adj. gross margin 47% (SEC EX-99.1, May 7 2026).
• Management reiterated full-year 2026 revenue growth guidance of approximately 40%+ YoY in the June 17, 2026 8-K / EX-99.1.
• StockAnalysis financials show TTM revenue $283.42M, gross margin 40.94%, operating margin 17.11% as of the May 7, 2026 update.
• The June 17 filing says the company is scaling with a new CFO and that the evaluation/trust theme is central to the next phase; no full transcript was available, only the earnings-release language.

• The Q1 2026 10-Q says one customer generated approximately 56% of revenue and another approximately 17%; no other customer accounted for 10% or more, and non-U.S. customers accounted for 12% of revenue. That underscores both concentration and embeddedness in large programs. (SEC 10-Q filed 2026-05-07)
• The 2026-06-17 8-K/EX-99.1 says Innodata appointed Jayant Chauhan as CFO and reaffirmed full-year 2026 revenue growth guidance of approximately 40% or more year-over-year. (SEC 8-K filed 2026-06-17)
• The 10-Q business overview says Innodata supports the development, training, post-training, evaluation, and deployment of advanced AI systems, which matches the Layer 1 bottleneck thesis. (SEC 10-Q filed 2026-05-07)
• StockAnalysis quote page shows market cap $2.25B, revenue $283.42M, P/E 61.39, forward P/E 72.05, beta 2.89, and target $122.75. (StockAnalysis quote page checked 2026-07-05)

• The 2025 10-K says Innodata supports "development, training, post-training, evaluation, and deployment" of advanced AI systems and that data quality and rigorous evaluation will be decisive as AI scales. It also says the company employed 10,107 people at year-end 2025, which underscores a labor-heavy operating model. (10-K, Item 1 Business / Employees)
• The Q1 2026 10-Q says revenue increased 54% YoY to $90.1M, primarily from higher volume for AI-related data services, expansion of existing customer programs, and new client engagements supporting more complex AI workflows. (10-Q, MD&A)
• Customer concentration remains extreme: one customer generated approximately 56% of Q1 2026 revenue and another 17%; no other customer exceeded 10%. That means the top two programs were about 73% of quarterly revenue, so retention of just two programs is the near-term scale gate. (10-Q, MD&A)
• Direct operating costs were 60.5% of FY2025 revenue and 56% of Q1 2026 revenue. Management said the direct-cost increase was primarily driven by headcount and cloud-service usage, confirming the bottleneck is trained human throughput plus process execution, not capital intensity. (10-K and 10-Q)
• The June 17, 2026 8-K / EX-99.1 said Innodata is “at the center of one of the most important shifts in technology, providing the data, evaluation, and trust infrastructure required to build high-performing generative AI,” and reaffirmed full-year 2026 revenue growth guidance of approximately 40%+ YoY. (8-K EX-99.1)

• Q1 2026 revenue $90.1M beat consensus by 18% ($76.5M expected); adj. EBITDA $25.0M nearly doubled YoY from $12.8M; gross margin expanded to 47% from ~40% a year ago (Source: Q1 2026 10-Q MD&A, SEC EDGAR 2026-05-07)
• Customer diversification progressing: second-largest customer grew to 17% of revenue while primary declined from >60% to ~56%. Aggregate revenue from non-primary Big Tech customers grew 453% YoY. (Source: Q1 2026 10-Q, Item 1 and MD&A)
• LCCI Benchmark (Long Context & Complex Interaction) — public preview dashboard launched for evaluating models on realistic extended multimodal conversations with revision tracking, buried context, and failure-mode analysis (Source: @innodata X, July 2026)
• ICAB (Innodata Cultural Alignment Benchmark) — multimodal benchmark testing implicit cultural understanding across workplace norms, family dynamics, visual prompts, cross-locale nuance (Source: @innodata X, May-June 2026)
• Agentic AI platform beta secured $1M hyperscaler engagement; 15+ companies evaluating, 2 hyperscalers discussing channel partnership (Source: CEO Q1 2026 letter, May 2026 EX-99.1)
• SHIELD program (defense/security clearances) opens federal AI data pipeline work through CDAO Tradewinds (Source: @innodata X, March 2026)
• New CFO Jayant Chauhan (ex-Mphasis/Blackstone, OYO/SoftBank) appointed June 17, effective July 6 — signals institutionalization/scaling phase (Source: June 17 8-K EX-99.1)
• Q2 2026 earnings scheduled for August 6, 2026 (5 days from this research date) — near-term catalyst
• Two new Schedule 13G filings (July 28 and July 31, 2026) indicate continued institutional accumulation

• Q2 revenue of $92.1M was up 58% YoY, marked the 12th consecutive quarter of YoY growth, and management said it beat consensus by 7%. (SEC EX-99.1, 2026-08-06)
• In Q2, the largest customer was 37% of revenue, the Big Tech customer announced last quarter scaled to 34%, and a new customer was landed from a fastest-scaling frontier lab. In H1, the two largest customers were 46% and 26% of revenue, and 66% of accounts receivable was due from two customers. (SEC EX-99.1 + 10-Q, 2026-08-06)
• The 10-Q says revenue increased primarily because of higher volume for AI data engineering services from existing customer programs; direct operating costs were $49.7M in Q2, up 40%, mainly because of headcount. (SEC 10-Q, 2026-08-06)
• The company entered a $300M equity distribution agreement on Aug. 6, which adds growth-capital flexibility but also creates dilution risk if heavily used. (SEC 8-K Item 1.01, 2026-08-06)
• Management also announced a planned leadership transition effective September 30, 2026: Rahul Singhal to become President and CEO, Jack Abuhoff to become Executive Chairman. (SEC EX-99.1 / 8-K, 2026-08-06)

This worked calculation shows the unit-economics logic: Innodata’s actual gross margin is highly sensitive to price per task and cost per task. The actual values should be input from the SEC filing before any conviction score is assigned.

\[

\Delta \text{Revenue} = 52.2 - 22.1 = 30.1 \text{ million}

\]

\[

\Delta \text{Adjusted EBITDA} = 19.6 - 3.2 = 16.4 \text{ million}

\]

\[

\text{Incremental adjusted EBITDA margin} =

\frac{16.4}{30.1} = 0.545 \text{ or } 54.5\%

\]

For every incremental dollar of year-over-year revenue in that comparison, approximately $0.545 flowed to adjusted EBITDA. This indicates significant operating leverage, likely because the cost base includes fixed personnel, platform, and management capacity that did not need to scale proportionally with revenue.

This is not a forecast. The incremental margin can decline if:

• New programs require large upfront training/hiring.
• Higher-cost domain experts replace lower-cost labelers.
• Customer concentration causes lumpy volume.
• Pricing competition compresses gross margin.

This sensitivity table shows why the high incremental margin assumption is material to the company’s operating-profit trajectory. The high case is [SINGLE-SOURCE] and should be treated with lower confidence until corroborated by a second originating source.

Contribution profit per hour = 80 - 32 - 8 = $40 per hour

Monthly contribution per FTE = 40 × 140 = $5,600

Monthly revenue per FTE = 80 × 140 = $11,200

Contribution margin = 40 / 80 = 50%

This 50% contribution margin would be the delivery-level margin before corporate SG&A. It is not the company’s reported gross margin, which would likely include additional delivery overhead and allocation differences.

Case A: Bill rate declines 15%

P = 80 × 0.85 = $68

Contribution per hour = 68 - 32 - 8 = $28

Monthly contribution per FTE = 28 × 140 = $3,920

Contribution margin = 28 / 68 = 41.2%

This shows that a modest bill-rate compression of 15% drops delivery-level margin by roughly 8.8 percentage points.

Case B: Bill rate declines 30%

P = 80 × 0.70 = $56

Contribution per hour = 56 - 32 - 8 = $16

Monthly contribution per FTE = 16 × 140 = $2,240

Contribution margin = 16 / 56 = 28.6%

This suggests that if the market commoditizes and pricing falls sharply, the unit economics become much thinner.

Case C: Direct labor cost rises 20%

C = 32 × 1.20 = $38.40

Contribution per hour = 80 - 38.40 - 8 = $33.60

Monthly contribution per FTE = 33.60 × 140 = $4,704

Contribution margin = 33.60 / 80 = 42.0%

This illustrates that labor inflation matters, but bill-rate compression is often the larger margin risk.

Case D: Utilization drops from 140 to 110 hours

Contribution per hour = 80 - 32 - 8 = $40

Monthly contribution per FTE = 40 × 110 = $4,400

Monthly revenue per FTE = 80 × 110 = $8,800

Contribution margin = 40 / 80 = 50%, but monthly contribution falls by 21.4%.

This shows that even with stable pricing and stable unit margin, lower utilization reduces absolute economics.

The worked calculation supports three conclusions:

1. Utilization and bill rates are the key swing factors in a labor-intensive data services model.
2. Gross margin can remain high in a high-demand environment, but the model has significant downside if contracts are repriced or work is insourced.
3. Actual Innodata margins cannot be inferred from this model; they must be pulled from the 10-K.