INOD · Innodata Inc. · Stock research

Current Thesis

Two weeks after the 2026-08-06 Q2 print answered the concentration question in the bulls' favour, the tape has not paid for it. The stock closed $63.85 on 2026-08-14 and $64.21 on 2026-08-21, with RSI(14) drifting from 57.2 to 51.9; it sits 47.2% below the 52-week high of $121.50 and the shares are down 32.8% over three months. The narrative leg on offer is unchanged: an AI training-data, RLHF and model-evaluation pure-play growing revenue 58% YoY with adjusted gross margin at 49% against a 40% internal target. What has changed since the last note is the packaging — a live $300M at-the-market equity program (424B5, 2026-08-06) and a founder handing over the CEO seat on 2026-09-30 — and the fact that eleven sessions of post-print trade produced no directional resolution.

The narrative is maturing. The dating: fundamentals still beat on 2026-08-06; the 2026-08-07 session took the stock to an intraday $78.05 and closed it at $64.35, down 1.7%; the two weeks since have gone sideways inside a few dollars. Coverage is established and the aggregate sell-side mark ($122.75 average across four analysts, stockanalysis.com as of 2026-08-22) sits roughly 91% above the last close, which is a gap the marks have not yet been cut to close. It is not dead — revenue is accelerating, the balance sheet holds $250.4M of cash and short-term investments, and price has stayed above the 2026-07-17 close of $60.81 for five weeks. It is not accelerating — a revenue beat of roughly $4–6M against the various wire tallies bought exactly one session of upside.

Bullish and bearish views on Innodata Inc.

The model's bull view on Innodata Inc. (INOD), in brief: Q2 2026 (2026-08-06): revenue $92.142M, +58% YoY, described in the release as the 12th consecutive quarter of year-over-year growth. The bear view: The print was absorbed, not chased. 2026-08-07: intraday high $78.05, close $64.35 (-1.7% on the day). Ten sessions later the close was $64.21. A beat that doubles the consensus EPS line and leaves the price flat two weeks on is the participation datapoint. The $300M… Both cases follow in full.

Bull Case

• Q2 2026 (2026-08-06): revenue $92.142M, +58% YoY, described in the release as the 12th consecutive quarter of year-over-year growth. Diluted EPS $0.41 against a consensus carried at $0.21 on one wire and $0.22 on another; revenue against estimates of $86.3M and $88.0M on those same tallies.
• Profitability scaled faster than the top line: adjusted EBITDA $25.4M, +92% YoY, 27.5% of revenue; adjusted gross margin 49% versus the company's stated 40% target; net income $14.4M (2026-08-06 release).
• Concentration halved in one quarter: the largest customer fell to 37% of Q2 revenue from 56% in Q1 2026, while a second Big Tech account went from 17% to 34%. The single-point-of-failure framing that defined this name in May is a smaller argument now.
• Cash and short-term investments $250.4M at 2026-06-30; H1 2026 operating cash flow $164.4M, with roughly $134M of customer prepayments flagged against that balance.
• FY2026 revenue growth guidance of "40% or more" reiterated on 2026-08-06, with management stating that large programs from new and prospective customers are not in the number.
• Sell-side marks stayed high after the print: average 12-month target $122.75 across four analysts, high $140 (BWS Financial, raised from $110), low $111, per stockanalysis.com on 2026-08-22 — against the 2026-08-21 close of $64.21.

Bear Case

• The print was absorbed, not chased. 2026-08-07: intraday high $78.05, close $64.35 (-1.7% on the day). Ten sessions later the close was $64.21. A beat that doubles the consensus EPS line and leaves the price flat two weeks on is the participation datapoint.
• The $300M at-the-market common-stock program was filed the same day as the results (424B5, 2026-08-06), with Goldman Sachs, Craig-Hallum, Wells Fargo Securities, Maxim Group and Wedbush as sales agents. The company now has a continuous mechanism to sell equity into strength. Shares outstanding showed 34.38M on the stockanalysis.com company page as of 2026-08-22 — the number to compare against the Q3 10-Q cover page.
• Guidance was reiterated, not raised. Against H1 revenue of $182.2M, ts2.tech calculated that the reiterated guide implies an average H2 quarterly pace near $85.0M, roughly 7.7% below Q2's $92.1M.
• The top two customers still represent about 71% of Q2 revenue. The percentage table alone does not settle whether the largest account's dollar spend grew or shrank; that requires the 10-Q, and diversification driven by an account rolling over is a different fact from diversification driven by a second account scaling.
• Founder selling preceded the handover: a 294,059-share disposal on 2026-06-16 (roughly $32M per the GuruFocus insider tally) and a further 200,000-share filing dated 2026-07-16, inside a disclosed disposal run exceeding $126M in the three months to mid-July.
• The competitive set — Scale AI, Surge, Mercor — is private and unpriced, so pricing pressure shows up only in Innodata's own gross margin, with a lag.

Setup & Price Structure

• Reference points: last completed daily close $64.21 (2026-08-21); 52-week high $121.50, so -47.2%; three-month price change -32.8%; RSI(14) 51.9, down from 57.2 on 2026-08-14.
• The closes on the record here — $60.81 (2026-07-17), $64.35 (2026-08-07), $63.85 (2026-08-14), $64.21 (2026-08-21) — span under four dollars across five weeks. That is compression around a shelf, and a shelf is not yet a base: nothing in the sequence has tested the 2026-08-07 spike high of $78.05, which is the level that would show the print finally being paid for.

• Crowding and positioning observables, stated as observables: four analysts publish targets averaging $122.75 (2026-08-22), about 91% above the last close, and none of them had cut into the drawdown as of that page; an at-the-market program is live and can issue into any rally; insider disposals are disclosed through mid-July; and there is no scheduled earnings date inside the next 30 days, so no dated binary is compressing positioning right now.
• Market capitalisation showed $2.21B on stockanalysis.com as of 2026-08-22 against a reiterated FY2026 guide implying revenue above $352.328M on the 2026-08-06 wire framing.

Catalyst Calendar (next 30 days)

• 2026-08-23 → 2026-09-22: no scheduled company-specific event identified. The window is empty; drift and sell-side revisions are the only inputs.
• 2026-09-30 (dated, just outside 30 days): CEO transition takes effect — Rahul Singhal becomes CEO, Jack Abuhoff becomes Executive Chairman and remains Board Chair (Form 8-K, 2026-08-06).
• ~2026-11-05 (est., early November in prior years): Q3 2026 results and the 10-Q. First read on the H2 pace against Q2's $92.142M, on whether the largest customer holds near 37%, and on any share count growth from the ATM versus 34.38M.

Elapsed catalysts

• Continuous: at-the-market sales under the 2026-08-06 424B5 are disclosed at the quarterly filing, not in real time, so the dilution datapoint arrives only with the Q3 report. (passed 20d ago)

What Would Change Our Mind

The structure that matters is the five-week shelf between the 2026-07-17 close of $60.81 and the 2026-08-07 spike high of $78.05. Losing the low end takes out the only piece of evidence that the post-print absorption was accumulation rather than distribution; On the fundamental side, three specific prints would break the leg: Q3 revenue coming in below $92.142M sequentially while the FY guide is framed at the low end of "40% or more"; adjusted gross margin retreating toward or below the 40% company target from the 49% posted in Q2; or a Q3 10-Q cover page showing materially more than 34.38M shares outstanding, which would date the ATM as an active seller rather than an option. In the other direction, the case strengthens on a weekly close back above $78.05 with the Q3 concentration table holding the largest customer near 37% on higher total revenue — that combination would argue the narrative is re-accelerating rather than maturing. The 2026-09-30 handover passing without incremental founder disposals removes one overhang; further Form 4 sales after that date add one.

Correlation Notes

• There is no listed pure-play comparison for AI data labelling and model evaluation; the closest competitors are private. Read-across therefore arrives from customer behaviour, not from a peer group tape.
• With the top two customers at about 71% of Q2 revenue, the revenue line is levered to two hyperscaler-scale model-training budgets. Capex commentary from large AI platform companies is a more relevant input than any AI-services index.
• The name trades with high-beta AI mid-caps on risk-off days: the last note flagged that a weekly close below $58 on no company-specific news, alongside a broad drawdown in AI-linked mid-caps, would be multiple compression rather than a company event.
• This is a services business with project-based, lumpy revenue — moves are driven by earnings, guidance and customer disclosure, not by memory pricing, GPU lead times or datacenter build headlines.