Skip to main content
Homepage
Full 18-step deep-dive · ~20–30 min · a member feature.
AGING Analysis Report
Aug 15, 2026
8 days ago · 100% complete
For AI assistants & researchers — machine-readable summary of this page

What this page is: Delvantic's full research page for Innodata Inc. (INOD) — AI-driven forensic equity research: mechanical valuation models (DCF, EPV, anchored-PE, scenario) plus three independent AI lenses (Quality / Value / Sentiment). Everything below is rendered server-side; you are not missing content that requires JavaScript. All scores are predictions and research opinions, not financial advice.

Our current read (analysis of 2026-08-23): Designation Low · Gem Score -36 (−100…+100 Quality+Value blend) · Quality 21 · Value -83 · Sentiment 70 (timing only, not weighted)

Page map (sections in order; each card carries a stable reference-name attribute you can cite):

  • profile-header / price-overview — company profile, live quote, market cap
  • extended-analysisthe core: three AI lens reads with findings, scores, and the analyst memo
  • future-predictions — our forward price-band predictions
  • market-narrative / ai-findings / gpt-critique — narrative context, cross-model findings, and an adversarial critique of our own analysis (near the end of the document)
  • Members-only sections (render as login gates for anonymous readers): price-history, income-trend, key-metrics, financials (statement tables), insider-trading. The analysis above is public; the raw data tables require a free account.

More for machine readers: site briefing at /llms.txt · any ticker resolves at delvantic.com/stock/TICKER · raw inputs are public-company filings and market data (via licensed data feeds); every model, score, lens read, and prediction on this page is Delvantic's own analysis.

Innodata Inc.

INOD NASDAQ
Technology · Information Technology Services
Ridgefield Park, NJ 07660, United States innodata.com Updated Aug 15, 10:11am
Price
$63.85
Market Cap
$2.1B
Employees
10,020
Beta
2.92
Avg Volume
1,264,305
CEO
Mr. Jack S. Abuhoff J.D.

Innodata Inc. is a global data engineering and AI services company specializing in Generative AI, traditional AI, and enterprise AI solutions. It operates through three key segments: Digital Data Solutions (DDS), which provides AI data preparation including training data collection, annotation, algorithm training, model deployment, and data engineering services like transformation, curation, hygiene, and compliance; Synodex, focused on transforming medical records into usable digital data for insurance and healthcare decision support; and Agility, offering tools for public relations professionals to discover influencers, monitor coverage, and measure campaign impact. Innodata delivers industry-specific platforms and services across sectors such as healthcare, finance, legal, media, aerospace, defense, automotive, and technology, supporting AI model development, safety testing, agentic AI, document intelligence, and workflow automation. With expertise in over 85 languages and global delivery capabilities, it enables technology companies, enterprises, and AI builders to integrate high-quality data into AI initiatives. Innodata Federal serves U.S. government agencies in defense, intelligence, and civilian missions. Founded in 1988 and headquartered in Ridgefield Park, New Jersey, Innodata emphasizes the inseparable link between data and AI to drive innovation in data processing and hosting services.

Runs with full report Generated: Aug 15, 2026 10:19am
Price Overview
Price at report time
$63.85
as of Aug 15, 10:31am (8d ago)
Change · Aug 15
+1.72 (+2.77%)
Day Range
$61.68 – $66.86
52-Week Range
$34.23 – $125.14
50-Day MA
$74.97
200-Day MA
$61.35
Volume
1,865,900.00
Right now · live
Log in to get the live feed
Members see the real-time price and the move since this report (over 8d).
Share Structure
Outstanding 32,655,358.00
Float 32,782,139.00
Free Float 100.4%
High free float — 100.4% of shares trade freely, ~-0.4% held by insiders/institutions
Very liquid — most shares trade freely. Low insider ownership can mean less management alignment, but makes large position sizing straightforward.
Price History (1 Year)
Last updated: Aug 15, 2026 10:31am (8d ago)
Revenue & Net Income Trend
The directional story — useful even when net income is negative.
Last updated: Aug 15, 2026 10:30am (8d ago)
Revenue
The top line — total sales before any costs or taxes are subtracted. A measure of how much business the company is doing.
Net Income
The bottom line — profit left after subtracting all expenses, interest, and taxes from revenue. Reflects accounting profitability, but includes non-cash items like depreciation, so it isn't the same as cash earned.
Operating Cash Flow
The real cash generated by the day-to-day business — selling products, paying suppliers, collecting from customers. Calculated from net income by adding back non-cash items and adjusting for timing (unpaid bills, unsold inventory). When OCF consistently lags net income, the reported profit may not be converting to real money.
Period Revenue Net Income Net Margin YoY/QoQ
Key Metrics
TD Twelve Data statement HEX SEC filing
Industry comparison last run: Aug 15, 2026 10:17am
P/E Ratio (Price per dollar of earnings)
HEX
Stock Price / EPS (Diluted)
69.40
Stock Price: $63.85
EPS (Diluted): 0.92
P/B Ratio (Price vs net asset value)
HEX
Stock Price / Book Value Per Share
20.89
Stock Price: $63.85
Total Equity: $107.06M
Shares: 35,025,000
EV/EBITDA (Total value vs operating profit)
HEX
Enterprise Value / EBITDA
41.67
Market Cap: $2.09B
Total Debt: $0.00
Cash: $82.23M
EBITDA: $46.76M
Enterprise Value (Takeover price (cap + debt - cash))
HEX
Market Cap + Total Debt - Cash
$1.9B
Market Cap: $2.09B
Total Debt: $0.00
Cash: $82.23M
Gross Margin (Revenue left after direct costs)
HEX
Gross Profit / Revenue
39.5%
Gross Profit: $99.48M
Revenue: $251.66M
Operating Margin (Revenue left after all operations)
HEX
Operating Income / Revenue
15.8%
Operating Income: $39.87M
Revenue: $251.66M
Net Margin (Revenue left as actual profit)
HEX
Net Income / Revenue
12.8%
Net Income: $32.18M
Revenue: $251.66M
ROE (Profit from shareholder equity)
HEX
Net Income / Total Equity
30.1%
Net Income: $32.18M
Total Equity: $107.06M
ROIC (Profit from all invested capital)
HEX
NOPAT / Invested Capital
124.7%
Operating Income: $39.87M
Tax Rate: 22.3%
Equity: $107.06M
Total Debt: $0.00
Cash: $82.23M
Zero debt — invested capital = equity minus cash (very efficient)
Current Ratio (Can it pay short-term bills)
HEX
Current Assets / Current Liabilities
2.68
Current Assets: $135.39M
Current Liabilities: $50.53M
Debt/Equity (Leverage — debt vs equity)
HEX
Total Debt / Total Equity
0.00
Short-Term Debt: $0.00
Long-Term Debt: $0.00
Total Debt: $0.00
Total Equity: $107.06M
Zero debt — this company carries no debt obligations. Strongest possible score.
Rev/Share (Top-line per share)
HEX
Revenue / Shares Outstanding
$7.19
Revenue: $251.66M
Shares: 35,025,000
Book Value/Share (Net assets per share)
HEX
(Total Assets - Total Liabilities) / Shares
$3.06
Total Equity: $107.06M
Shares: 35,025,000
FCF/Share (Real cash generated per share)
HEX
(Operating Cash Flow + CapEx) / Shares
$1.02
Operating CF: $46.75M
CapEx: -$11.10M
Shares: 35,025,000
CapEx is negative (outflow) — added to OCF to get FCF
Div Yield (Annual income from holding)
TD
Last Annual Dividend / Stock Price
Last Dividend: $0.00
Stock Price: $63.85
Payout Ratio (Earnings paid out as dividends)
HEX
Dividends Paid / Net Income
Dividends Paid: N/A
Net Income: $32.18M
Dividends paid not available in cash flow statement
Industry Benchmarks
Last run: Aug 15, 2026 10:17am
Compares INOD against LLM-researched typical ranges for its industry. One research call per industry, cached indefinitely — every stock in the same industry reuses the same baseline.
Income Statement (Annual)
Last updated: Aug 15, 2026 10:30am (8d ago)
Metric 2021 2022 2023 2024 2025
Revenue $86.8M $170.5M $251.7M
Cost of Revenue $55.5M $103.4M $152.2M
Gross Profit $31.3M $67.1M $99.5M
Operating Expenses $71.4M $89.5M $31.0M $42.7M $59.6M
Operating Income $318,000 $24.3M $39.9M
Net Income -$1.7M -$11.9M $-908,000 $28.7M $32.2M
EBITDA $5.0M $30.1M $46.8M
EPS $-0.06 $-0.44 $-0.03 $0.98 $1.01
EPS (Diluted) $-0.06 $-0.44 $-0.03 $0.89 $0.92
Balance Sheet (Annual)
Last updated: Aug 15, 2026 10:11am (8d ago)
Metric 2021 2022 2023 2024 2025
Cash & Equivalents $18.9M $9.8M $13.8M $46.9M $82.2M
Total Current Assets $34.0M $23.7M $32.1M $81.0M $135.4M
Total Assets $59.2M $48.0M $59.4M $113.4M $168.6M
Current Liabilities $21.3M $20.8M $22.9M $39.5M $50.5M
Long-Term Debt $6.2M $5.1M
Total Liabilities $32.8M $30.0M $34.4M $50.1M $61.5M
Total Equity $25.0M $63.4M $107.1M
Retained Earnings $3.2M -$8.8M -$9.7M $19.0M $51.2M
Cash Flow (Annual)
Last updated: Aug 15, 2026 10:30am (8d ago)
Metric 2021 2022 2023 2024 2025
Operating Cash Flow $5.2M -$1.2M $5.9M $35.0M $46.8M
Capital Expenditure -$4.4M -$6.5M -$5.6M -$7.7M -$11.1M
Free Cash Flow $783,000 -$7.7M $339,000 $27.3M $35.6M
Acquisitions (net)
Net Debt Issued / (Repaid) $-691,000 $-639,000 $-452,000 $-513,000 $-420,000
Dividends Paid
Stock Buybacks
Net Change in Cash $1.3M -$9.1M $4.0M $33.1M $35.3M
Growth Trends (YoY %)
Last updated: Aug 15, 2026 10:30am (8d ago)
Metric 2022 2023 2024 2025
Revenue Growth +96.4% +47.6%
Gross Profit Growth +114.3% +48.3%
Operating Income Growth +7,552.8% +63.8%
Net Income Growth -613.4% +92.4% +3,256.4% +12.3%
EBITDA Growth +498.6% +55.2%
0Company Classification 1Industry Landscape 2Company Momentum 3Forward Projection 4aDCF Valuation 4bEarnings Power Value 4cAnchored PE 4dReverse DCF 4eRevenue-Based DCF 4fAnchored P/S 4gScenario Analysis 4hDividend Discount Model 4iBook Value Analysis 4jInsider Activity 4fCash Flow Quality 4gDebt Maturity Risk 4hMacro Environment 4iSector Intelligence 4jRevenue Confidence 4kSensitivity Analysis 4lSector Demand Cycle 5AI Investigation 5bThesis Evaluation 6Valuation Synthesis
computed not applicable not yet run 12 computed · 6 not applicable · 6 not yet run
Narrative Economics
The story the market is telling about this stock — the intangible X-factor (founder mythology, cult dynamics, TAM-of-imagination) that moves price beyond what cash flows alone explain. After Shiller, Narrative Economics.
No narrative profile yet for INOD — it's generated by the pipeline (market-narrative step).
AI Lens 4th lens · how AI reaches this business · 5-yr
2026-08-15
The creme is there an opportunity here? Conditional opportunity
Own it as the scarce-expert supplier to AI, not as the labor arbitrage — and the gross margin line tells you which one you actually own.
Revenue compounding from $86.8M to $251.7M with 15.8% operating margin and $35.6M FCF is real, and the demand driver is the most durable spend line in technology. But gross margin has been pinned near 39% through a tripling of revenue, which says the delivery model is still labor-linear and price is set by a crowded bidding field against concentrated, technically self-sufficient buyers. Watch two things before consensus does: gross margin breaking above ~42% (expert mix and AI-assisted throughput are compounding, bull path toward 89) and customer concentration disclosure (one lab in-housing is the 21 bear case, and it arrives without a sales-cycle warning).
61
AI Position
Favorable but structurally fragile — widest possible range
Innodata's entire revenue base exists because frontier labs still need humans to supply what models can't yet generate for themselves — a demand curve that is enormous, real, and dependent on the very gap AI is racing to close.
Exposure 96 Confidence 54 50 = neutral
Primary Tailwind

Every incremental dollar of frontier-lab and enterprise model spend creates demand for expert-curated training, RLHF, evaluation and red-team data — the one input that cannot be bought as compute; revenue tripled from $86.8M to $251.7M in two years on exactly this.

Primary Pressure

The monetized unit is billable expert-hours delivered offshore at 39.5% gross margin. Synthetic data, RLAIF, model-graded evaluation and AI-assisted pre-labeling all attack the hour count directly, and the buyers doing that attacking are the same handful of labs that constitute most of the revenue.

Critical Hinge

Whether the mix migrates up-stack — from volume annotation toward scarce credentialed-expert work (MD/JD/PhD reasoning traces, domain evals, agentic task supervision) faster than automation deflates the base. Observable in gross margin: sustained move above ~42-45% means expert mix is winning; drift toward the mid-30s means price-per-hour deflation is winning.

Hard to Reproduce

A vetted, security-cleared, multi-jurisdiction expert delivery workforce with lab-grade quality processes and multi-year embedded relationships inside a small number of frontier-model programs; plus Synodex's regulated medical-record extraction, which carries liability that buyers do not want to own.

Forensic fingerprint same 11 factors for every stock · 0 unfavorable · 50 neutral · 100 favorable
Underlying Need Persistence do people still need this at all? 82
Models will need human-sourced ground truth, preference data and independent evaluation for the foreseeable horizon, but the intensity per model is not guaranteed.
The need — grounding machine output in verified human judgment — persists and broadens as agents take consequential actions; what is uncertain is how many human hours per capability point are required as self-play and synthetic pipelines mature.
Frontier lab data spend disclosures · Share of training runs using synthetic data · Growth of eval/red-team line items
relevance 92 · confidence 72
Solution Persistence will they still solve it this way? 48
Outsourced human-labor delivery is the current solution and the most automatable part of the stack.
Buyers need the data, not Innodata's offshore delivery pods; pre-labeling models, marketplace platforms and in-house expert networks are all live substitutes for the specific way Innodata solves it.
Revenue per delivery employee trend · Automation-assisted output disclosures · Shift from annotation to evaluation mix
relevance 95 · confidence 52
Intelligence Commoditization does cheap AI power them or copy them? 52
Cheap models both cut Innodata's delivery cost and shrink the hours customers will pay for.
AI pre-labeling can lift throughput per worker, but in a competitive labor-brokerage market with sophisticated concentrated buyers, per-unit price is likely to fall alongside cost.
Gross margin above or below 40% · Pricing per task disclosed in calls · Competitor rate cards
relevance 90 · confidence 55
Responsibility Transfer are they paid to take the blame? 58
Modest but rising: eval, safety testing and regulated medical extraction carry accountability buyers prefer to hand off.
Synodex's insurance/medical record work and third-party model evaluation both involve outcomes where an external, auditable provider is preferable to self-attestation — but core annotation carries almost no liability shield.
Independent AI audit/assurance contracts · Synodex revenue disclosure · Regulatory eval mandates emerging
relevance 58 · confidence 50
Scarcity Migration do their assets get rarer or more common? 64
Scarcity is migrating from generic labelers to credentialed domain experts and secure delivery capacity — Innodata can reach that, but does not own it.
As base tasks automate, value concentrates in physicians, lawyers, quants and engineers producing reasoning traces; recruiting and vetting those people at scale is genuinely hard but is a recruiting capability, not an owned asset.
Disclosed expert headcount and credentials · Average bill rate per expert hour · Retention of specialist talent pools
relevance 88 · confidence 52
Customer DIY Preference will customers just build it themselves? 34
The buyers are the most technically capable organizations on earth and have already shown willingness to internalize data supply.
Frontier labs treat data pipelines as core IP; the Meta/Scale-style equity-and-absorb template shows in-housing is not hypothetical, and Innodata's concentration means one such decision matters enormously.
Customer concentration percentage · Lab in-house data team hiring · Contract length and renewal terms
relevance 90 · confidence 56
AI Intermediation Position do AI agents go through them or around them? 47
Innodata sits upstream of models, not between agents and users, so agentic routing neither protects nor bypasses it much.
Its position is supplier-to-builder; the intermediation risk is procurement platforms and expert marketplaces disintermediating the services middleman rather than AI agents rerouting demand.
Rise of expert-marketplace competitors · Direct lab-to-expert sourcing platforms · Procurement moving to spot pricing
relevance 55 · confidence 45
Data Leverage does their data make AI better? 38
Innodata creates data it largely does not keep — the IP transfers to the customer.
Training-data work is typically delivered as work-for-hire, so the compounding asset accrues to the lab, not to Innodata; what accrues internally is process and quality tooling, which is weaker than a proprietary corpus.
Any retained-IP or licensed dataset revenue · Proprietary benchmark/eval asset creation · Reuse rate of internal tooling
relevance 70 · confidence 58
AI Margin Conversion do the AI savings become profit? 55
Operating margin already scaled from 0.4% to 15.8%, but that is volume leverage more than AI-driven cost conversion.
Gross margin has been stuck around 39% through a tripling of revenue, which suggests delivery economics are labor-linear; AI-assisted throughput has not yet shown up as structural gross margin expansion.
Gross margin trajectory vs revenue growth · SG&A leverage as revenue scales · Delivery headcount growth vs revenue growth
relevance 85 · confidence 57
Revenue Unit Durability does the thing they charge for survive? 41
The unit charged for — human expert task-hours — is precisely what improving models eliminate.
Unlike a seat or a transaction, the billable hour has no floor if automation raises output per worker faster than task volume grows; durability depends on the mix moving to recurring evaluation and assurance work.
Recurring vs project revenue split · Multi-year committed contracts · Task-volume vs hour-count divergence
relevance 93 · confidence 55
Entrant Compression how easily can newcomers copy them? 29
Barriers are recruiting and trust, not technology — well-funded AI-native data firms are already at scale.
Surge, Mercor, Turing, Invisible and lab-internal teams can replicate the delivery model with capital and an orchestration layer; the industry's -23.7pp operating margin compression shows what happens when services barriers are thin.
Win/loss against AI-native data firms · Bid pricing on new lab programs · Competitor funding and headcount scale
relevance 88 · confidence 62

AI Lens thesis

Innodata sells the human residual of AI — the labor the models still can't self-supply — so its demand is perfectly correlated with AI capex and its supply is perfectly exposed to AI capability. Cheaper intelligence expands the customer set (every enterprise fine-tuning, every agent needing evaluation harnesses) while simultaneously compressing the hours needed per unit of output and arming Surge/Mercor/Turing-style entrants who need only a recruiting funnel and an orchestration layer to compete. The company can convert AI into margin by using models to pre-label and QA its own work, but in a labor-brokerage market with near-zero switching cost and concentrated buyers, those savings get bid into price. The outcome therefore turns on mix, not on whether AI helps or hurts: if Innodata becomes the supplier of scarce credentialed judgment and durable evaluation infrastructure, cheap intelligence makes it richer; if it remains a preferred-vendor labor arbitrage, cheap intelligence makes it a shrinking pass-through at high growth for two more years and then not.

Thesis breaker A top-two customer disclosing in-housing or a large step-down in program spend, or gross margin compressing two consecutive years while revenue still grows — that combination means volume is being bought with price and the model is deflating.
What the market may be underestimating

Upside Model evaluation, safety testing and agent-behavior auditing are becoming recurring regulatory-adjacent obligations rather than one-time build costs; if Innodata converts project work into standing eval/assurance contracts, the revenue unit stops being a project hour and starts being a compliance subscription.

Downside Customer concentration means the DIY decision is not distributed across a market — a single lab deciding to internalize expert sourcing (the Meta/Scale template) can remove a double-digit share of revenue in one contract cycle, with no sales-motion warning.

Outcome range spread 68 · unresolved

21Bear case
58Central case
89Bull case
Three headline numbers, deliberately never blended: Position (which way), Exposure (how much it matters at all), Confidence (how sure). The fingerprint asks every stock the same 11 questions so companies a sector label would lump together get told apart. Not an input to GEM/Coal or the Q/V/S lenses.
Growth Outlook
Analyzed 2026-08-17 16:28

The question every valuation on this page silently assumes: is this company likely to grow? Judged forward — the business, its category, the world — against what's already printed.

Accelerating Revenue growth stepped up from ~48% to ~56% YoY with net income nearly doubling — Innodata is compounding inside an AI-data niche that is expanding while its nominal industry contracts, but the price already assumes 60% growth, a bar that gets harder each quarter as the base scales. conf 7/10
Share gain Category growing · The reported industry aggregate (IT Services) is contracting -1.8% to -3.8% with margins compressing broadly, yet the sector demand cycle reads 'expansion'. Innodata does not really compete in that aggregate: its served category — AI training, annotation and evaluation data — is expanding rapidly. Against the reported aggregate the company is taking share by ~49pp; against its true served category it is growing roughly with or slightly ahead of the pocket.
Next 2 quarters
Accelerating
Momentum is mechanical at this point: backlog and in-flight programs booked at the higher run-rate carry into the next two prints, revenue growth is already re-accelerating, and operating leverage is amplifying it into earnings. Nothing in the inputs signals a program roll-off in the immediate window.
↑ above expectations
Year 1
Growing
Full-year math is already largely locked by first-half strength; the question is only the exit rate. Expect continued strong growth with a naturally decaying rate as comps stiffen against the accelerated base — well above the industry's -1.8%, but below the extraordinary first-half pace.
≈ inline with expectations
Years 2–3
Growing
The demand for curated, verified AI data should persist as enterprises move from pilots to production and as model evaluation/safety work institutionalizes. But two structural forces cap the rate: base effects on a much larger revenue line, and price-per-unit erosion from synthetic data and competitive flooding. Concentration means the distribution of outcomes here is wide in both directions.
↓ below expectations
The creme: each rung's call measured against what's already printed (vs analyst estimates · vs guidance / FY consensus · vs price-implied growth) — expectations in print are already in the price, so only the variant margin can pay. Hover a rung's chip for the margin read.
Growth drivers
71 Revenue growth rate itself is rising — Matched-quarter YoY revenue +56.1% through Jun-2026 vs a trailing recent YoY of +47.6% — the second derivative is positive, not just the level. Quarterly trend flagged 'accelerating' with all years positive and moderate volatility (0.244). This is a demand-pull signal, not a comp artifact.
54 Operating leverage on a services base — Net income +95.3% on revenue +56.1% — earnings growing roughly 1.7x revenue. In a delivery-labor business that implies rising utilization/mix toward higher-value generative-AI programs (RLHF, evaluation, model-safety data) rather than commodity annotation, and it means incremental contracts drop through.
57 Share gain against a contracting nominal industry — Company recent YoY +47.6% vs industry -1.8% (and -3.8% CAGR) — a ~+49pp gap. The measured industry is legacy IT services; Innodata's actual served category (AI training/eval data) is a growth pocket inside a shrinking aggregate, so the classification understates the true category and overstates the headwind.
43 Persistent, large estimate beats — EPS beats of +58% and +121% in the last two prints (0.41 vs 0.26; 0.42 vs 0.19) after an inline Feb print. Sell-side models are structurally lagging the ramp — a pattern that typically persists one to two quarters past the inflection before estimates catch up.
Growth risks
70 Customer concentration — Innodata's revenue has historically leaned heavily on a small set of Big Tech buyers whose data-labeling budgets are discretionary and re-allocated annually. A single program pause converts +56% into flat without any change in end-market demand. This is the single largest variance source in the 2-3 year rung.
51 Commoditization and substitution of human data work — Bear case has teeth: synthetic data generation, model-assisted labeling, and a crowded competitive set (specialist data vendors plus offshore incumbents) all push price-per-unit down. Growth can stay positive while unit economics and pricing power erode, which is exactly what industry-wide margin compression (-23.7pp over 3 years) describes.
63 The bar embedded in expectations — Price-implied growth is +60% vs house projection of +49.4% — the market is paying for a rate above even the accelerated print, sustained. Simple base mathematics make 60%+ progressively harder as revenue scales; the risk is not decline but deceleration into a number already assumed.
34 Macro/capex sensitivity of AI service budgets — Macro backdrop flagged as headwinds with 10y at 4.63. AI data programs are funded from R&D/experimental budgets — the first line trimmed if enterprise AI spending shifts from land-grab to ROI discipline, and Innodata has no subscription base to cushion it (Agility/Synodex are small).
The world is in the middle of a build-out where model quality is increasingly gated by proprietary, curated, human-verified data rather than raw compute or architecture. That makes the data-preparation layer structurally more valuable than it was two years ago, and Innodata sits in it as a pure-play with delivery scale that enterprises cannot stand up internally quickly. The counter-force is that this layer is the most substitutable part of the stack: synthetic data, model-assisted labeling and price competition all attack unit revenue even as volumes rise. So the honest world read is volume-rich, price-fragile — growth continues but the mix of who captures it shifts toward whoever owns the customer relationship and the evaluation/safety workflows rather than the raw labeling hours. Innodata's revenue and earnings acceleration says it is currently on the right side of that line; concentration says the position is rented from a few buyers rather than owned.
Growth position composite +2
ShrinkingStallingHoldingGrowingAccelerating
90Next 2 quarters · Accelerating
70Year 1 · Growing
70Years 2–3 · Growing
+2Composite (−100…+100)
A research prediction, not advice. Forward-graded: each rung is scored against the prints that follow it. Not an input to the GEM designation — track record first.
Claude Reading
Independent analyst synthesis · claude-opus-4-7 · generated 2026-08-15 10:29:08
Verdict Overvalued but not catastrophically so — fair value ~$45 on forward earnings; concentration risk in Big Tech LLM contracts is the underappreciated bear case, wait for 10-K customer disclosure or two more $90M+ quarters before buying.

The raw quarterly trajectory is the single most important fact here and the DCF-anchored synthesis is under-weighting it. Revenue went $58.3M → $58.4M → $62.6M → $72.4M → $90.1M → $92.1M across the last six quarters — that's not smooth 30% growth, that's a step-function from a ~$233M run-rate to a ~$370M run-rate in three quarters, with Q2'26 YoY at +58%. TTM revenue is now ~$317M vs the $251.7M FY2025 base the multiples in the canonical block are calculated against, meaning the "69x P/E" and "8.9x P/S" are already stale — on forward run-rate ($368M annualized) at current ~15.6% net margin (~$57M NI), you're looking at ~37x forward earnings and ~5.7x forward sales, not 69x/8.9x. That's still not cheap, but it's a very different stock than the synthesis is anchoring on.

That said, the $17.14 DCF fair value is almost certainly wrong in the other direction — it appears to bake in a mean-reversion assumption that ignores the acceleration. But the bear case has real teeth that the momentum crowd is ignoring. First: Q2'26 margin (15.6%) is lower than Q3'24 (33.3%) — the 2024 margin was likely a tax/one-time artifact, but net margin has been drifting *down* from 17.4% (Q4'24) to 12-13% (mid-2025) and only recovered to ~16% as revenue scaled. That's operating leverage, not structural margin expansion, and it's fragile. Second: FCF of $35.6M on a $2.09B market cap is a 1.7% FCF yield — you need multi-year 40%+ FCF growth to justify entry here, and the "925% FCF CAGR" is off a near-zero base and meaningless. Third: the insider activity on 6/16/2026 shows 200,000 options exercised and ~200,000 shares sold same-day — that's a cash-out, not diversification, timed near highs before the ~50% drawdown from 52-week highs the pre-flight mentions.

The contrarian read that nobody in the prior models articulated cleanly: Innodata's revenue is almost certainly heavily concentrated in a handful of Big Tech LLM training contracts (Microsoft/Meta/Google-scale customers doing RLHF and data labeling). This is the same business Scale AI runs, and Scale just got hollowed out by Meta's $14B acquihire — meaning the customer base is actively insourcing this capability. A 10-K read would likely show >50% revenue from one or two customers. If true, the "durable AI infrastructure" narrative is a mirage: this is a labor-arbitrage BPO reselling offshore annotation labor to hyperscalers who are one strategic decision away from bringing it in-house or switching to Scale/Surge/Labelbox. The narrative layer correctly flags "fragile durability" but understates the specific concentration risk. The 70% revenue CAGR is real; the question is whether Q3-Q4 2026 shows the third consecutive quarter of ~$90M+ or whether one contract renegotiation cuts revenue 30% in a single quarter — a very real risk in this business model.

I partially dissent from the synthesis. $17 fair value is absurd — it ignores that TTM earnings are ~$46M and the company just doubled run-rate. But $63.85 is also not defensible: on 25x forward earnings (fair for a services business with concentration risk and 15% margins even if growing fast), you'd want to see this closer to $43-48. The stock is overvalued but not by 73% — call it 25-35% overvalued with meaningful binary risk around customer concentration disclosure. The insider selling, the margin compression through mid-2025, the FCF quality flag, and the platform-monopoly narrative attached to what is fundamentally a data-labeling services business all point the same direction. I'd want to see either (a) a 10-K customer concentration breakdown showing <30% top-customer exposure, or (b) two more quarters at $90M+ revenue confirming the step-function is sticky, before touching this. The prior models were right on direction and wrong on magnitude — the synthesis anchored to a DCF that ignored acceleration, and the narrative layer correctly identified fragility but the momentum signal correctly identified that the acceleration is real. Both are true simultaneously.

GPT Reading
Independent reading · gpt-5.4 · generated 2026-08-15 10:29:25
Verdict Overvalued at $63.85 — strong operations justify a premium, but not a platform-like one; fair value looks closer to $40-$45 unless $100M+ quarterly revenue and mid-to-high teens margins prove durable.

The raw operating story is real, and better than a lazy “AI hype stock” dismissal gives it credit for. Revenue has gone from $86.8M in 2023 to $170.5M in 2024 to $251.7M in 2025, and the quarterly run-rate has now jumped again to $90.1M and $92.1M in the first two quarters of 2026. That means the business is currently running at roughly a $365M-$370M annualized pace versus $251.7M last year, with Q2 2026 up nearly 58% year over year against $58.4M in Q2 2025. Just as important, this is not growth bought with losses: net income was $32.2M in 2025, operating cash flow $46.8M, and free cash flow $35.6M, with no debt and $82.2M of cash. Margins are not collapsing under growth either. Net margin in the last four quarters was 12.2%, 16.5%, 15.6%, and before that 12%-13% most quarters, which suggests this is a scaled services/data operation with genuine operating leverage, not just one noisy contract quarter.

What stands out, though, is that the stock price is discounting a continuation of this unusual combination of hypergrowth and respectable profitability for longer than I’m comfortable underwriting. At $2.09B market cap and $63.85 per share, the stock trades at about 8.3x 2025 revenue, but still around 5.7x-5.8x a forwardized 2026 revenue run-rate if you annualize the first half. That multiple is not insane for a software platform with sticky recurring revenue and high incremental margins; it is aggressive for an IT services/data-enablement business whose gross margin is 39.5% and operating margin 15.8%. The annual P/E of 69x looks optically silly because earnings have already stepped up in 2026, but even if I annualize the last two quarters’ combined earnings power at about $58M-$59M, the stock is still around 36x earnings. For a company with equity of only $107.1M and a 20.9x price-to-book, the market is clearly valuing a future moat, not the current balance sheet or current economics.

The biggest quantitative tell is the decoupling between revenue acceleration and earnings growth. Recent revenue growth is explosive, but recent earnings growth is only 12.3% year over year. Q2 2026 net income of $14.4M was up from $7.2M in Q2 2025, which is strong, but Q1 2026 net income of $14.9M versus $7.8M in Q1 2025 also implies that some of the annual valuation metrics are stale and some quarter-to-quarter comparisons include unusual items. More importantly, sequentially revenue rose from $72.4M in Q4 2025 to $90.1M in Q1 2026 and $92.1M in Q2 2026, while net income only moved from $8.8M to $14.9M and then slipped to $14.4M. That is fine operationally, but it does not support paying a “winner-take-most AI infrastructure” multiple. This still looks like a high-quality beneficiary of AI spend, not an owner of the stack. The insider tape also leans against enthusiasm: a cluster of option exercises followed by meaningful same-day sales in June is not a death knell, but it is not the behavior you want to see if management thinks $63.85 materially understates intrinsic value.

The strongest case against my caution is straightforward: maybe the market is correctly identifying that annual 2025 figures understate the present earnings power by a lot, and maybe this is exactly the kind of stock old valuation frameworks miss. If 2026 revenue lands anywhere near $360M-$380M and net margin holds around 15%-16%, net income could approach $55M-$60M. On that basis, the company would be debt-free, cash-rich, growing 40%+ with healthy cash generation, and trading closer to a mid-30s earnings multiple rather than 69x. In a market starving for profitable AI-linked names, that may deserve a premium. The quarterly trajectory also argues against calling this a fad too early: revenue rose from $52.2M in Q3 2024 to $92.1M in Q2 2026, and profitability has remained positive throughout. A smart bull would say this is exactly what early compounding looks like before consensus catches up.

What would change my mind is not another strong growth quarter by itself, but proof that the business can sustain this scale with improving economics and better visibility. If Innodata can print two more quarters with revenue above $95M-$100M, operating margin above 17%, and free cash flow conversion near or above net income, I would be more willing to treat today’s multiple as growth-stock expensive rather than speculative expensive. I would also want to see insider selling normalize and evidence that customer concentration or project volatility is not lurking behind the acceleration. Conversely, if revenue growth dips below 30% while net margin slips back toward 10%-12%, the stock will look dramatically over-earning on narrative and I would expect a much harsher re-rating. Today, I think the business is good, but the stock is still priced as if this service layer is becoming a durable AI tollbooth.

Grok Reading
Independent reading · grok-4.5 · generated 2026-08-15 10:29:59
Verdict Overvalued at $63.85; business is real and growing but 69x earnings and ~$17 DCF imply 70%+ downside if growth normalizes

The numbers tell a genuine operating turnaround that became an AI-data services rocket, then a valuation problem. Innodata went from $86.8M revenue and near-breakeven in 2023 to $170.5M in 2024 and $251.7M in 2025, with the last two reported quarters at $90.1M and $92.1M—implying a run-rate already north of $360M. Net income flipped from losses to $32.2M in 2025, and recent quarterly net margins sit at 15.6–16.5%, well above the full-year 12.8%. Operating cash flow of $46.8M and FCF of $35.6M against a clean balance sheet ($82.2M cash, zero debt, current ratio 2.7) confirm the earnings are real cash, not accounting theater. Gross margin near 40% and ROE of 30% show a services business that has scaled with leverage. The story the data tells is not “fake growth”; it is a company that caught the generative-AI training and annotation wave and converted it into profitable scale faster than most peers.

That same data, however, cannot support a $2.09B market cap at $63.85. Trailing P/E of 69x, EV/EBITDA of 42x, P/S of 8.9x and P/B of 21x price in multi-year continuation of 50%+ growth and durable high-teens margins. Recent revenue YoY has already cooled to roughly 48%, and sequential growth from Q1 to Q2 2026 was only about 2%. The valuation synthesis fair value near $17 is directionally correct: even generous forward multiples on a $360–400M revenue run-rate with mid-teens net margins leave the stock demanding perfection that the deceleration already contradicts. Heavy insider selling clustered on 16 June 2026—multiple large option exercises immediately followed by sales totaling well over 150k shares—adds a clear signal that those closest to the business are not treating $60+ as a bargain. The narrative layer (AI data backbone, platform-monopoly framing) is carrying almost the entire premium; fundamentals alone do not.

The strongest opposing case is straightforward and must be weighed seriously. Revenue CAGR of 70% over the measured period, accelerating quarterly tops ($58M → $62M → $72M → $90M → $92M), and FCF that exploded as the company scaled are not typical of a fading story. If enterprise custom-model and domain-specific training spend remains labor- and expertise-intensive for another three to five years, Innodata’s healthcare and financial-services verticals (Synodex, DDS) could sustain 35–45% growth long enough for earnings to grow into a mid-20s multiple, making today’s price merely early rather than wrong. Zero leverage and $35M+ FCF also give management optionality to reinvest or acquire without dilution. A bull can fairly argue that 69x trailing looks less absurd once you annualize the latest two quarters and assume operating margins hold or expand toward 18–20%. I weigh this less heavily because services businesses in AI data face rapid competitive entry from larger IT outsourcers and Big Tech tooling; the 39.5% gross margin is solid but not a structural moat, and the narrative durability itself is labeled fragile for good reason—one or two soft quarters on growth or margin would collapse the multiple long before earnings catch up.

I would reverse to a fair-or-better stance if the next two quarters print revenue above $100M with net margins holding above 15%, or if management discloses multi-year contracted backlog that visibly covers 2027 growth at 40%+, or if a material new vertical win demonstrates pricing power rather than volume chasing. Sustained FCF conversion above 70% of net income alongside any evidence that competition is not compressing gross margins would also force a re-rating of the durability premium. Until then the gap between $63.85 and cash-flow reality is too wide.

Big-3 Panel — where each AI stands
Each AI above independently stated a direction (undervalued, fairly valued, or overvalued) and how strongly it believes it (conviction, 0–5). We combine those into a Bull-Bear Index on a 0–10 scale: 5 is neutral, 10 is maximum bullish (undervalued at full conviction), 0 is maximum bearish. We compute the score ourselves with the same arithmetic for every seat — the models never grade their own bullishness — so the three are directly comparable. Δ shows how far each seat sits from the panel average of 1.3; a large Δ marks the dissenting voice, usually the one worth reading.
Claude claude-opus-4-7 2.0
overvalued · conviction 3/5 · Δ +0.7 vs panel · self: 3.0
GPT gpt-5.4 1.0
overvalued · conviction 4/5 · Δ -0.3 vs panel · self: 3.0
Grok grok-4.5 1.0
overvalued · conviction 4/5 · Δ -0.3 vs panel · self: 3.0
Advanced Analysis Forensic deep-dive · separate lenses
Separate reads — Company Quality (is it a great business?), Valuation (is it mispriced?), and General Sentiment (how macro + narrative are pushing it), plus AI Impact (how the AI wave reshapes it), kept deliberately apart · 2026-08-15 10:39:06
Delvantic - Cairn AI
Quality — pass at $64, buyer in the $30s 7/10
Great, real business at a price that already prices in the miracle — I want it, but not here.
The cruxWhether Innodata becomes the scarce-expert supplier to AI labs or stays labor-linear at 39% gross margin — that mix determines if today's $63.85 is a bargain or a 3x overpay.
Forensic checks Derived mechanically from INOD's filed financials — not from the AI lenses
Liquidity & RunwaySelf-Funding
DilutionHeavy Dilution
Earnings QualityHigh Earnings Quality
The four lensesswitch a tab for its full read — score + evidence
Company Quality
+21
Strong
edge √Σ 114 · risk √Σ 92 · conf 7/10

Innodata has transformed from a sub-scale, loss-making services firm into a high-growth, profitable operator: revenue $86.8M (2023) to $170.5M (2024) to $251.7M (2025), with gross margin expanding from 36.1% to 39.5% and operating margin from 0.4% to 15.8%. Net income turned from a $908K loss in 2023 to $32.2M in 2025, and free cash flow scaled from essentially zero to $35.6M. Balance sheet is unlevered with $82.2M net cash, Altman Z of 23.6, and OCF running ahead of net income (accruals -11.9% of assets, Beneish -2.41) — the mechanical earnings-quality checks are clean.

Strengths 3
m78
Revenue nearly tripled in two years with expanding margins
Revenue went $86.8M to $251.7M (2023-2025) while operating margin expanded 0.4% to 15.8% - real operating leverage, not just top-line growth.
m62
Clean earnings quality and fortress solvency
Altman Z 23.6, accruals -11.9% of assets, OCF/NI 1.36x, $82.2M net cash with no debt burden. Cash earnings exceed accounting earnings.
m55
Self-funding profitable growth
$35.6M FCF on $251.7M revenue (14% FCF margin) means the growth is paid for internally - no capital dependency.
Concerns 4
m60
Heavy dilution eroding per-share value
Diluted share count grew from 26.6M (2021) to 35.0M (2025), a 7.1% CAGR, with SBC at 4.4% of revenue and zero buyback offset. Per-share compounding materially lags the business.
m55
Aggressive CEO liquidation via option exercise/sell
CEO Abuhoff exercised options and sold >$30M in two days (June 2026), part of 70 sells totaling $151M over 12 months with zero insider buys. Even accounting for option-exercise mechanics, the scale and clustering signal aggressive personal de-risking.
m35
Customer concentration and business durability unproven
Innodata's historical model is data-annotation services for large AI labs; growth of this magnitude typically implies concentration in a few hyperscaler contracts. Durability of the revenue base cannot be confirmed from the derived data.
m25
Short profitability track record
The company was unprofitable through 2023; only two years of meaningful earnings and FCF exist. Not enough cycle to prove the margin structure is durable.
This is a genuinely improved business - the 2023-2025 numbers are not cosmetic, they are backed by real cash. Revenue tripled, margins expanded, FCF scaled, balance sheet is pristine, and the mechanical fraud/earnings-quality checks all come back clean. That said, I am not ready to call it a fortress: the company was a marginal money-loser as recently as 2023, the business rides an AI capex wave whose durability and customer concentration I cannot see from here, dilution runs 7% a year with no buyback, and the CEO is monetizing options at a pace that suggests he is happy to take chips off the table. Strong business, real strengths, but too young in its current form and too dependent on unverified customer relationships to grade higher.
Verify before trusting this (6)
  • Customer concentration disclosure in 10-K - what percentage of revenue comes from top 1-3 customers
  • Contract structure and revenue visibility - are these multi-year commitments or project-based
  • SBC dilution trajectory - is the 7.1% share-count CAGR expected to continue or moderate as the business scales
  • Segment mix between legacy content services and AI/LLM data - and margin profile of each
  • 10b5-1 plan disclosures behind the CEO June 2026 sales and any remaining unexercised option overhang
  • Whether the operating margin expansion is repeatable or reflects one-time contract mix
Valuation / Mispricing
-83
Overvalued
edge √Σ 20 · risk √Σ 137 · conf 7/10
Price $63.85 vs composite deserved ~$17 (quality-adjusted maybe high-$20s); price is roughly 2-3.7x deserved - clearly rich. attractive below $30.00

The e2e composite pegs fair value at $17.30 (signal-adjusted $17.14) versus a $63.85 price, implying about -73% downside if those anchors are right. Even the more generous DCF ($19.32) sits at less than a third of today's quote, while the EPV floor of $6.16 says the current run-rate earnings power, absent aggressive growth, supports only single-digit dollars. Earnings quality is clean and the business is genuinely Strong, which argues for lifting deserved value above a mechanical DCF - but not 3x above it.

Cheap signals 1
m20
Quality lifts deserved value above raw DCF
Clean earnings, pristine balance sheet, tripled revenue and real FCF justify a premium to the $17 composite - plausibly high-$20s to low-$30s - but nowhere near $64.
Rich / priced-in 4
m82
Price 3.7x composite fair value
$63.85 vs $17.30 composite / $17.14 signal-adjusted FV implies -73% - the market is paying multiples of the modeled deserved value.
m70
DCF itself only $19.32
Even the more forward-looking DCF, which already embeds growth, comes in below $20. Today's price needs a DCF roughly 3x higher, i.e. a materially more bullish growth/margin path than modeled.
m65
EPV floor near $6
Earnings power value of $6.16 shows how thin the current run-rate cushion is: strip away growth optionality and the business supports single-digit dollars, so ~90% of the market cap is growth-story.
m55
Insider selling into the story
Aggressive CEO selling plus heavy dilution flagged by the quality lens is consistent with insiders viewing the price as full, not cheap.
I like the business more than I like the stock. Every reasonable anchor - composite $17, DCF $19, EPV $6 - is a fraction of $63.85, and even after generously marking up for the Strong quality grade I cannot get deserved value much past the low-$30s. That is not a margin of safety; that is paying up for a narrative the insiders themselves are trimming into. I would need it roughly in half, closer to $30, before the risk/reward starts to interest me, and I would want to see a fresh $20-handle print before doing real work. Until then this is a Rich name to admire, not own.
Verify before trusting this (5)
  • Customer concentration disclosure - what % of revenue is top 1-3 customers, and any renewal/expansion commentary
  • Organic vs acquired revenue growth and gross-margin trajectory in latest 10-Q
  • Forward guidance and backlog / bookings vs current run-rate
  • Share count trajectory and stock-based compensation as % of revenue to quantify dilution drag
  • Segment mix between generative-AI data services and legacy content ops
General Sentiment
+70
Strong Tailwind
tail √Σ 135 · head √Σ 49 · conf 8/10

The pressure on this name is overwhelmingly positive right now. The active narrative is a strong 'AI data-infrastructure pure-play' story with medium cult coefficient, and it just got refueled by a record Q2 (58% revenue growth), reaffirmed 40%+ 2026 guidance, and Zacks/Wall Street write-ups explicitly tagging INOD as an 'AI-led' pick with upside. On a 2.92 beta name, a risk-on tape (VIX 14.3, S&P near highs) is a force multiplier - every marginal AI headline gets levered into the price, and there is no risk-off pressure to fight it. Recent big-move history (+86% and +22% single-day rips on prior prints) shows the tape is primed to reward, not fade, positive surprises. The counter-pressure is real but latent: narrative durability is flagged fragile, the stock trades at a huge premium to DCF, and articles are starting to hedge with 'premium valuation and customer concentration temper the outlook.' That is the crack to watch, but it is not currently the dominant force - momentum is strong_positive and news flow is uniformly constructive. Net: this is a story-stock in its sweet spot, on a friendly tape, with fresh fundamental validation feeding the narrative loop. The pressure is a strong tailwind until the AI-capex narrative itself cracks or a print disappoints.

Tailwinds 4
m82
AI data-infra narrative in full force
Strong-intensity platform-monopoly story with the market treating INOD as a pure-play on enterprise AI training data. Cult coefficient medium and rising - exactly the archetype that gets bid in a risk-on tape.
m78
Fresh catalyst just validated the story
Q2 record with 58% growth and reaffirmed 40%+ 2026 guidance landed within 72h. History (+86%, +22% single-day moves on prior prints) shows the tape rewards these directly rather than fading them.
m55
Risk-on tape amplifies a 2.92-beta name
VIX 14.3, S&P near highs, regime established 10 days. High-beta AI story-stocks are exactly the profile that outperforms in this tape - no macro headwind to fight the narrative.
m50
Uniformly constructive analyst/media tone
Zacks highlighting INOD alongside larger IT peers, multiple 'more upside ahead' headlines, and buy-side write-ups pairing it with PWR as top AI R&D picks. No visible bearish analyst voice in the flow.
Headwinds 2
m40
Narrative durability flagged fragile
The story rests on a 272% premium to DCF and assumes TAM keeps expanding. Articles are already starting to hedge on 'premium valuation and customer concentration' - the seed of a de-rating is planted, just not yet germinated.
m28
Recent momentum decelerating vs long-term
Recent 47.6% vs 70.3% long-term CAGR flags mild exhaustion. Not enough to break the tape but a yellow light on how much of the good news is already in.
This is about as clean a tailwind setup as sentiment gets: a strong AI-infrastructure narrative just refreshed by a record print, reaffirmed guidance, uniformly bullish media flow, all landing on a calm risk-on tape that levers a 2.92-beta name higher. I am not grading whether the 272% premium to DCF is deserved - that is not my job - but the non-fundamental pressure right now is decisively pushing this stock UP, not down. The one thing giving me pause is 'fragile' narrative durability and the first hedging language creeping into coverage, which is why I stop short of maxing out confidence. Until the AI story itself cracks or INOD misses a print, sentiment is a strong tailwind.
Verify before trusting this (4)
  • Any crack in the AI-capex narrative (hyperscaler capex guide-downs, model-commoditization headlines) that would de-rate the whole cohort
  • Next quarterly print - a miss or a lowered 40% growth reaffirmation would puncture a fragile narrative fast
  • Customer-concentration disclosures - loss or renegotiation of the big Big Tech contract would shift tone instantly
  • VIX break above 20 or a risk-off rotation - high-beta story stocks get sold first
The market-wide tape + this name's exposure to it (beta / sector / narrative durability). Context on the non-fundamental pressure — not a call on the business or the price. processId: detail-general-sentiment
AI Impact
+0
Favorable but structurally fragile — widest possible range
opp √Σ 64 · thr √Σ 50 · conf 5/10

Innodata sells the human residual of AI — the labor the models still can't self-supply — so its demand is perfectly correlated with AI capex and its supply is perfectly exposed to AI capability. Cheaper intelligence expands the customer set (every enterprise fine-tuning, every agent needing evaluation harnesses) while simultaneously compressing the hours needed per unit of output and arming Surge/Mercor/Turing-style entrants who need only a recruiting funnel and an orchestration layer to compete. The company can convert AI into margin by using models to pre-label and QA its own work, but in a labor-brokerage market with near-zero switching cost and concentrated buyers, those savings get bid into price. The outcome therefore turns on mix, not on whether AI helps or hurts: if Innodata becomes the supplier of scarce credentialed judgment and durable evaluation infrastructure, cheap intelligence makes it richer; if it remains a preferred-vendor labor arbitrage, cheap intelligence makes it a shrinking pass-through at high growth for two more years and then not.

AI opportunities 2
m59
Underlying Need Persistence
Models will need human-sourced ground truth, preference data and independent evaluation for the foreseeable horizon, but the intensity per model is not guaranteed.
m25
Scarcity Migration
Scarcity is migrating from generic labelers to credentialed domain experts and secure delivery capacity — Innodata can reach that, but does not own it.
AI threats 3
m29
Customer DIY Preference
The buyers are the most technically capable organizations on earth and have already shown willingness to internalize data supply.
m17
Data Leverage
Innodata creates data it largely does not keep — the IP transfers to the customer.
m37
Entrant Compression
Barriers are recruiting and trust, not technology — well-funded AI-native data firms are already at scale.
Own it as the scarce-expert supplier to AI, not as the labor arbitrage — and the gross margin line tells you which one you actually own. Revenue compounding from $86.8M to $251.7M with 15.8% operating margin and $35.6M FCF is real, and the demand driver is the most durable spend line in technology. But gross margin has been pinned near 39% through a tripling of revenue, which says the delivery model is still labor-linear and price is set by a crowded bidding field against concentrated, technically self-sufficient buyers. Watch two things before consensus does: gross margin breaking above ~42% (expert mix and AI-assisted throughput are compounding, bull path toward 89) and customer concentration disclosure (one lab in-housing is the 21 bear case, and it arrives without a sales-cycle warning).
Verify before trusting this (8)
  • Revenue per delivery employee trend
  • Automation-assisted output disclosures
  • Shift from annotation to evaluation mix
  • Recurring vs project revenue split
  • Multi-year committed contracts
  • Task-volume vs hour-count divergence
  • Frontier lab data spend disclosures
  • Share of training runs using synthetic data
The structural effect of the AI wave on this specific business over the next ~5 years — demand, cost leverage, moat, barriers to entry, position in the AI stack. The reality beneath the AI story, not the story's market pressure (General Sentiment owns that) — and not a call on the business today or the price.
Growth Outlook
+2
Accelerating
edge √Σ 114 · risk √Σ 112 · conf 7/10

The world is in the middle of a build-out where model quality is increasingly gated by proprietary, curated, human-verified data rather than raw compute or architecture. That makes the data-preparation layer structurally more valuable than it was two years ago, and Innodata sits in it as a pure-play with delivery scale that enterprises cannot stand up internally quickly. The counter-force is that this layer is the most substitutable part of the stack: synthetic data, model-assisted labeling and price competition all attack unit revenue even as volumes rise. So the honest world read is volume-rich, price-fragile — growth continues but the mix of who captures it shifts toward whoever owns the customer relationship and the evaluation/safety workflows rather than the raw labeling hours. Innodata's revenue and earnings acceleration says it is currently on the right side of that line; concentration says the position is rented from a few buyers rather than owned.

Growth drivers 4
m71
Revenue growth rate itself is rising
Matched-quarter YoY revenue +56.1% through Jun-2026 vs a trailing recent YoY of +47.6% — the second derivative is positive, not just the level. Quarterly trend flagged 'accelerating' with all years positive and moderate volatility (0.244). This is a demand-pull signal, not a comp artifact.
m54
Operating leverage on a services base
Net income +95.3% on revenue +56.1% — earnings growing roughly 1.7x revenue. In a delivery-labor business that implies rising utilization/mix toward higher-value generative-AI programs (RLHF, evaluation, model-safety data) rather than commodity annotation, and it means incremental contracts drop through.
m57
Share gain against a contracting nominal industry
Company recent YoY +47.6% vs industry -1.8% (and -3.8% CAGR) — a ~+49pp gap. The measured industry is legacy IT services; Innodata's actual served category (AI training/eval data) is a growth pocket inside a shrinking aggregate, so the classification understates the true category and overstates the headwind.
m43
Persistent, large estimate beats
EPS beats of +58% and +121% in the last two prints (0.41 vs 0.26; 0.42 vs 0.19) after an inline Feb print. Sell-side models are structurally lagging the ramp — a pattern that typically persists one to two quarters past the inflection before estimates catch up.
Growth risks 4
m70
Customer concentration
Innodata's revenue has historically leaned heavily on a small set of Big Tech buyers whose data-labeling budgets are discretionary and re-allocated annually. A single program pause converts +56% into flat without any change in end-market demand. This is the single largest variance source in the 2-3 year rung.
m51
Commoditization and substitution of human data work
Bear case has teeth: synthetic data generation, model-assisted labeling, and a crowded competitive set (specialist data vendors plus offshore incumbents) all push price-per-unit down. Growth can stay positive while unit economics and pricing power erode, which is exactly what industry-wide margin compression (-23.7pp over 3 years) describes.
m63
The bar embedded in expectations
Price-implied growth is +60% vs house projection of +49.4% — the market is paying for a rate above even the accelerated print, sustained. Simple base mathematics make 60%+ progressively harder as revenue scales; the risk is not decline but deceleration into a number already assumed.
m34
Macro/capex sensitivity of AI service budgets
Macro backdrop flagged as headwinds with 10y at 4.63. AI data programs are funded from R&D/experimental budgets — the first line trimmed if enterprise AI spending shifts from land-grab to ROI discipline, and Innodata has no subscription base to cushion it (Agility/Synodex are small).
vs expectations: ~6m above · 1y inline · 2-3y below
The forward growth verdict — is the business itself likely to grow (next 2 quarters / year 1 / years 2–3), judged against its category and against printed expectations. The full horizon ladder + creme renders on the Growth Outlook card above. Not a call on the price (Valuation owns that) or the tape (Sentiment owns that).
Please log in to view trade setups
The Augustus trade-setup read is a members feature.
Log in
Lenses kept deliberately separate — Company Quality (price-agnostic), Valuation (price-conditional), General Sentiment (non-fundamental macro/narrative pressure), AI Impact (structural ~5yr AI exposure), and Growth Outlook (the forward growth verdict). The scores are not blended. Filing-level items (convertibles, lock-ups, customer concentration) are v2 — see each lens's "verify."
Community AI Feedback
No community reviews yet for INOD. Be the first — hit How to Contribute, have any AI review this page, and paste its take back here.
My Notes personal — only you see this
v1.1.562 · 9b2927c4 · 2026-08-22 16:52:06