Innodata's Customer Diversification Effort: Is It Finally Paying Off?

Innodata's Q2 2026 revenue rose 58% year over year to $92.1 million, while adjusted EBITDA jumped 92% to $25.4 million and adjusted gross margin expanded 600bps to 49%. Management reiterated at least 40% revenue growth for 2026, supported by AI data-engineering demand and new opportunities in agentic AI, evaluation, cybersecurity and physical AI. Customer concentration improved as its largest account fell to 37% of revenue from 56% in Q1, although the top two customers still represented 71%; 2026 and 2027 EPS estimates increased to $1.18 and $1.67, respectively.

Analysis

The relevant change is not diversification per se but the implied durability of AI-data spend: two customers now represent distinct demand pools, reducing single-account cancellation risk while leaving concentration high enough that quarterly procurement timing can dominate reported results. The margin profile suggests reusable datasets and higher-complexity evaluation work are becoming a larger mix; if sustained, this can lift incremental EBITDA conversion above the market's services-company framework. The key diligence item is whether revenue is tied to recurring model-refresh/evaluation workflows versus finite pretraining projects, since the latter would create a material 2027 growth-air-pocket risk.

INOD's premium multiple already discounts a substantial portion of the operating-leverage narrative. At roughly 36x forward earnings, a single customer ramp pause, lower utilization, or mix shift back toward labor-intensive annotation could compress both estimates and the multiple over the next 1-3 months. Conversely, independently disclosed multi-year awards or evidence that the new frontier-lab customer becomes meaningful would justify a rerating because it would convert management's pipeline narrative into backlog visibility.

Competitive read-through is more favorable for TASK than PLTR: expanding demand for AI evaluation, trust-and-safety, and human-in-the-loop workflows broadens the addressable services pool, while PLTR monetizes a different enterprise software budget. The contrarian view is that diversification may actually increase pricing pressure if large platforms dual-source data work; INOD must demonstrate that its proprietary datasets and specialized evaluation capabilities retain gross margin as customer mix broadens. QBTS has no direct fundamental read-through and should not be traded on this development.

Key Decisions for Investors

• Maintain a tactical long INOD only on post-results weakness or after verification of contracted backlog; target a 6-12 month 20-30% upside if revenue growth remains above 40% and adjusted gross margin holds near 49%. Exit if either top-two customer exposure rises above 75% or gross margin falls below 45%, which would undermine the reusable-data thesis.
• Use a defined-risk structure rather than chasing spot: buy 6-9 month INOD call spreads financed only after implied volatility is reviewed against its post-earnings range. The trade requires confirmation that next-quarter guidance embeds continued sequential growth; absent that data, treat as a watch item.
• Pair a modest long INOD / short TASK only if INOD demonstrates another quarter of margin stability and TASK's AI-services growth fails to accelerate; this isolates specialized-data monetization from broad AI-services demand. Cover the pair if TASK reports superior AI margin expansion or INOD guides below 40% growth.
• Do not infer a PLTR catalyst from this news. PLTR's valuation and earnings sensitivity remain driven by enterprise/government software conversion, not outsourced model-data spending; avoid using INOD's results as a read-through for PLTR.