Innodata vs Palantir: 85.9% vs -22.9% YTD Splits the AI Data Trade

There is a 108-point performance gap between two stocks supposedly riding the same wave: Innodata is up 85.9% year to date in 2026 while Palantir has dropped 22.9%, against an S&P 500 that has gained 8.1%. That spread, between a $90 million quarterly revenue specialist and an enterprise platform doing $1.63 billion a quarter, is the real story. It tells us the AI infrastructure trade is bifurcating along valuation lines, not along quality or growth lines.

The Numbers

Start with the headline quarters. Innodata's Q1 2026 revenue grew 54% year over year to a record $90.1 million, with adjusted EBITDA up 96% to $25 million and adjusted gross margin at 47%, seven points above its own 40% long-term target, as TradingView reported. Palantir's same quarter was on a different scale entirely: $1.63 billion in revenue, up 85% year over year, with U.S. revenue up 104% and U.S. commercial revenue up 133%.

On profitability, Palantir is in a class that almost no public software company occupies right now. Adjusted operating margins of 60%, adjusted EBITDA margins of 61%, and free cash flow margins of 57% in a single quarter. Roughly $8 billion in cash and short-term Treasuries on the balance sheet. By contrast, Innodata's 47% gross margin is impressive for a services-heavy business but sits below Palantir's operating margin, which is the more demanding metric.

Now the valuation, which is where the YTD divergence starts making sense. Innodata trades at 7.85x forward 12-month price-to-sales. Palantir trades at 36.89x. That is a 4.7x multiple gap on companies both growing more than 50% year over year. Consensus 2026 revenue growth sits at 40.6% for Innodata and 71% for Palantir on management's own guidance, with Palantir's U.S. commercial segment projected to grow at least 120%.

The forward EPS picture for Innodata is more nuanced. The Zacks 2026 EPS estimate moved to 99 cents from 93 cents in the last 30 days, implying only 7.6% year-over-year growth, which is far below the 40.6% revenue growth estimate. The 2027 analyst EPS growth estimate then jumps to 79.2% on 29.5% revenue growth. That gap between 2026 and 2027 earnings trajectories is where the bull case lives or dies. The source does not disclose the cost structure driving the muted 2026 EPS, which matters because it could be either heavy reinvestment (good) or margin pressure as the new Big Tech customer ramps (less good).

What's Actually New

Two things in this filing cycle deserve attention from data and platform leads, separate from the share-price noise.

First, Innodata disclosed that a single Big Tech customer that contributed zero revenue a year ago could deliver roughly $51 million in 2026 and become its second-largest account. That is a step-function customer event, not organic growth. Meanwhile, revenues from other big tech customers grew 453% year over year in Q1 collectively. When you net those two facts, the picture is a vendor that has gone from selling to frontier labs to selling to multiple hyperscalers in parallel, which materially changes the concentration risk profile even if the absolute customer count is still small. We do not know the gross retention rate on these accounts, but the bound is informative: if even one hyperscaler pulls back, the 453% number cuts in both directions.

Second, Innodata launched an Evaluation and Observability Platform for AI agents in production, and landed a first $1 million engagement with a hyperscaler plus 15 additional companies in evaluation. This is the more interesting development for engineering audiences. Evaluation tooling for agentic systems is the unsolved layer of the stack right now. Traces, scoring rubrics, drift detection on tool-use behavior, none of it has a dominant vendor. If Innodata converts even a third of those 15 evaluations into paid contracts at $1 million-plus ACV, the product line transitions from services revenue (linear, headcount-bound) to platform revenue (margin-expanding). That is the lever that justifies the 79.2% 2027 EPS growth estimate.

On the Palantir side, the genuinely new disclosure is the magnitude of the U.S. commercial acceleration: 133% growth in a segment that was already large. ShipOS and Maven continue to anchor the government story, but the commercial inflection is what drove the 71% full-year guide. The question is not whether AIP is winning deals. It clearly is. The question is whether it is winning them at margins that justify 36.89x sales.

What's Priced In for Data Teams

For CTOs and platform leads making vendor decisions, the priced-in expectations matter as much as the headline numbers.

What is already priced into Palantir: continued 70%-plus growth, sustained 60% operating margins, and uninterrupted government spending through any political transition. The 22.9% YTD drawdown suggests the market is starting to discount at least one of those three assumptions slipping. For engineering teams evaluating Foundry or AIP, the practical implication is that Palantir has every incentive to defend its ACVs aggressively, which usually translates into longer commitment terms and bundled pricing. Compare that to Snowflake-style consumption pricing or a Databricks lakehouse build, where unit economics are visible per query. The Databricks and Snowflake stacks have become the natural reference points when budgeting against AIP, even though the feature surface is different.

What is already priced into Innodata: the new Big Tech account ramps cleanly to $51 million, hyperscaler revenue keeps compounding off the 453% base, and the evaluation platform converts pilots to production. That is a lot of execution assumed in an 85.9% YTD move, even at 7.85x sales.

What is not priced in for either name: a slowdown in frontier model training spend. Both companies are downstream of the same capex cycle. Innodata sells data engineering and evaluation services into model builders. Palantir sells the application layer on top of those models. If hyperscaler training budgets compress in late 2026, Innodata sees it first in pipeline conversion and Palantir sees it second in commercial deal velocity. The source does not disclose either company's exposure to specific frontier labs, which matters because customer concentration at the lab level is a different risk than concentration at the hyperscaler level.

Contrarian View

The consensus framing here is that Innodata is the value play and Palantir is the overpriced winner. I'd push back on that.

Innodata at 7.85x sales looks cheap relative to Palantir, but it is not cheap in absolute terms for a business that is still primarily services revenue with customer concentration risk. The 2026 EPS estimate growing only 7.6% while revenue grows 40.6% says operating use is not yet showing up. Either margins compress as the new accounts ramp, or reinvestment is heavy, or both. Neither scenario is a disaster, but neither is the cleanly compounding software profile that 85.9% YTD implies.

Palantir at 36.89x sales is expensive, but it is expensive on the back of 60% operating margins and $8 billion in cash. The drawdown this year may be the multiple compression that needed to happen, not the start of a longer slide. If U.S. commercial really does grow 120% in 2026, the multiple looks different by year-end.

The contrarian read: the better risk-adjusted trade may be the one the YTD chart says is losing. Innodata has more to prove. Palantir has already proven it and is being repriced.

Unanswered Question

The source does not disclose Innodata's gross retention on its top three hyperscaler accounts, nor Palantir's net revenue retention on the U.S. commercial segment that just grew 133%. Those two numbers would settle the entire debate. Testable bound: if Innodata's $51 million Big Tech ramp materializes on schedule and the evaluation platform converts at least 5 of the 15 pilots to paid contracts by Q4 2026, the 79.2% 2027 EPS growth estimate becomes credible and the 7.85x multiple looks justified. If either milestone slips by a quarter, expect consensus 2027 estimates to compress meaningfully.

Key Takeaways

• Innodata Q1 2026: $90.1 million revenue (up 54%), 47% adjusted gross margin, 96% adjusted EBITDA growth. Palantir Q1 2026: $1.63 billion revenue (up 85%), 60% adjusted operating margin, $8 billion cash.
• The valuation gap is 4.7x: Innodata at 7.85x forward sales versus Palantir at 36.89x. YTD performance has been 85.9% versus -22.9%.
• The single most important Innodata catalyst is a Big Tech customer ramping from zero to ~$51 million in 2026, becoming the #2 account. Concentration risk is improving but still real.
• The evaluation platform with one $1 million hyperscaler contract and 15 evaluations is the lever that could re-rate Innodata from services multiple to platform multiple.
• Prediction: if Palantir's U.S. commercial segment hits the 120% growth guide for 2026, expect the multiple compression to reverse and the YTD performance gap to narrow by at least 30 points by year-end.