Platform Moat Analysis

The Ontology is Palantir's core intellectual property and the foundation of its switching cost moat. Once an enterprise maps its operations into the Ontology (modeling business objects, their relationships, logic rules, and action triggers), migration requires rebuilding this entire semantic layer from scratch. The process of initial ontology creation takes weeks to months, representing significant upfront investment that creates sunk-cost inertia. Data is stored in proprietary shape and format — exportable as CSV/JSON but not readily usable by equivalent systems.

The question is ROI — at what point does the cost of staying with Palantir exceed the cost of migrating?

Stock-based compensation is the persistent bear case that undermines the moat's value to shareholders. FY2025 SBC was $684M (15% of revenue), creating a $840M gap between GAAP operating income ($1,414M) and adjusted operating income ($2,254M) when adding $156M employer payroll taxes on equity. Shares outstanding have grown 45% cumulatively since the 2020 IPO (from ~1,770M to ~2,565M diluted), and annual dilution remains ~4.7%. The $1B buyback program was terminated in January 2026 after utilizing only $75M (7.5%), suggesting management considers the stock fully valued.

The trend is clearly positive — the question is whether it improves fast enough relative to the extreme valuation.

Palantir does NOT have true data network effects in the classical sense (Uber: more drivers = more riders, Facebook: more friends = more value). Each customer's Ontology deployment is fundamentally independent — Customer A's data does not make Customer B's deployment better. What Palantir DOES have are four weaker but real moat mechanisms that are often mislabeled as 'network effects': (1) Implementation learning effects: Forward-deployed engineers accumulate domain expertise across deployments, enabling faster time-to-value for new customers in similar industries. This is institutional knowledge transfer, not a network effect.

The FedStart marketplace is the most promising true network effect but serves only the government segment. The ~711-customer base is far too small for meaningful data network effects compared to platforms with true network dynamics.

Palantir publishes 25+ case studies across government, commercial, and international customers. The evidence suggests real, quantifiable value — but with important caveats. The strongest evidence comes from third-party validated studies (Forrester TEI: 315% ROI; Nucleus Research: Swiss Re 170% ROI in 7.3 months) and operational metrics (Airbus Skywise: 10,500+ aircraft, 33% A350 production acceleration; BP digital twin: 30,000 additional bbl/day globally from APEX system, 2M+ sensors integrated). Military results are dramatic (Maven: 20,000+ active users, 2,000-person targeting cell replaced by ~20 analysts) but harder to independently verify.

CRITICAL FINDING: Most case studies demonstrate ONGOING platform value, not one-time implementations. However, virtually all quantified case studies are Palantir-commissioned or self-reported, and independent verification of specific dollar savings is sparse. The bear case is not that the platform delivers no value, but that the value may not justify the $114/share platform premium.

Palantir's net dollar retention rate reached 139% in Q4 2025, up from a trough of ~107% in Q3 2023, representing one of the most dramatic NRR recoveries in enterprise software. The 139% figure means existing customers are spending 39% more year-over-year. Three primary drivers explain this expansion: (1) AIP module cross-sell — the 2023 launch of AIP as an AI layer atop Gotham/Foundry creates a natural upsell path, with customers like Lear expanding from 100 users/4 use cases to 16,000 users/280 use cases; (2) Use-case proliferation within organizations — customers start with one department and expand across the enterprise, with ACV growth examples of $7M-to-$31M (utility) and $4M-to-$20M+ (energy) within a single year; (3) The bootcamp go-to-market motion — 75% conversion rate compresses sales cycles and accelerates initial deal sizes, which then expand further. Notably, management stated that NDR 'does not yet fully capture the acceleration' because it excludes customers acquired in the past 12 months.

However, PLTR's NRR is inflating off a low base (107% in 2023) and may not sustain at these levels once the AIP cross-sell wave saturates existing customers. The key question: is 139% a new steady state or a one-time AIP uplift?