Report contents
Industry landscape
Where the company competes: US government buyers in a crowded field of data software
Palantir sells software that pulls an organization’s scattered data into one place, models it as objects that match how the operation actually runs (a hospital bed, a police officer, a waiting list), and lets users act on those objects. Gotham, the original product, serves intelligence and military analysis. Foundry is the version sold to health services, police forces and companies. Palantir reported $4.5 billion of revenue in fiscal 2025, up from $2.9 billion a year earlier, about 55% growth. Government customers supplied 54% of that revenue, US customers 74%, and the company reported 954 customers.
Upstream sit systems Palantir does not own: databases, APIs, spreadsheets, sensor feeds and legacy applications. Downstream sit the analysts, clinicians, police staff and operations managers who use the output. Palantir’s claim to be hard to replace rests on the middle of that chain, the Ontology, which maps data to real-world objects and can write changes back into the source system. Computer Weekly’s July 2026 comparison of Foundry’s components found plenty of suppliers that do the ingestion and modeling work, and no identically described read-write semantic layer among them. Absence of a named equivalent in one trade-press comparison is not proof of uniqueness, but it locates where any advantage would have to live.
Three submarkets, each with its own geography. First, US defense and intelligence data fusion, including AI-assisted targeting: Palantir took over the Pentagon’s Project Maven in 2019 after Google withdrew, and its software runs across all six US military branches and 36 federal agencies. Second, operational data platforms bought by civilian agencies and large companies. Third, public-sector work outside the US, currently concentrated in the UK, where the company runs the NHS Federated Data Platform on a £480 million, seven-year contract and piloted a workforce-analytics system for the Metropolitan Police.
Geography bounds each of these. Government procurement, security clearance and data-residency rules make the buying unit national, often agency by agency, so scale does not travel freely across borders the way it does in commercial software. With 74% of revenue from the United States, the arena Palantir actually contests today is mostly US federal, with the UK its largest beachhead abroad.
| Layer | Function | Suppliers named in the comparison |
|---|---|---|
| Data integration | Ingests, cleans and models data from databases, APIs, spreadsheets and sensors | Informatica, Fivetran, dbt, Azure Data Factory, AWS Glue, SnapLogic, Matillion |
| Semantic model | Maps data to business objects; writes back to source systems | Palantir’s Ontology; no direct equivalent identified |
| Application layer | Low-code operational apps built on the model | Palantir Workshop |
Market structure: crowded below, thin at the top. The integration layer alone has seven credible vendors, and two of them are hyperscaler services bundled with cloud contracts. By the practical test that barriers are weak when leaders cannot be counted on one hand, entry at that layer is easy and price competition there should be assumed. The barrier sits higher up, in the semantic model and in the accreditation that lets a vendor touch classified or health data at all. Neither is quick to replicate, though only the accreditation is genuinely exclusive.
Market shares are hard to pin down. Palantir does not publish a share of any submarket, and the integration vendors do not report a comparable segment, so concentration has to be inferred from the shape of the customer base. It is barbell-shaped. The top 20 customers averaged $64.6 million each in the prior fiscal year, the company reported, while 954 customers against $4.5 billion of revenue implies roughly $4.7 million per customer on average, an estimate from reported figures. A handful of agencies carry the book; a long tail supplies growth.
Remaining deal value grew 40% to $5.4 billion, the company reported, or about 1.2 times fiscal 2025 revenue. That is a modest book relative to current billings, which makes the renewal calendar more important than any backlog as a leading indicator.
Two things would change this picture. The UK contract carries a break clause the government can exercise in February 2027, and two parliamentary committees have already urged it to do so; a exit there would show the non-US public-sector market is smaller and more political than the contract value suggests. And if a hyperscaler or an integration vendor ships a comparable semantic layer, the crowded lower layers would start to set Palantir’s price. So far the evidence establishes a strong position with US government buyers; it does not establish how far that position travels beyond them.
Competitive advantage
Does the company have a moat? Fairly strong, and concentrated in US government work
Palantir earns excellent returns and has earned them for only a few years. Overall rating: Fairly Strong. The company clears the profitability test on current numbers with room to spare, and clears the concentration test only inside US federal work. Six years of public reporting, four of them profitable, is not long enough to show that either advantage survives a full competitive cycle.
The moat test is narrow. Barriers to entry show up in two places: concentration that holds over time, and returns above the cost of capital that persist for a decade or more. High margins on their own prove nothing, because a well-run company can earn them for a stretch and lose them once rivals copy the playbook. Palantir's margins arrived fast. Gross margin went from 78.0% in fiscal 2021 to 84.9% in the first nine months of fiscal 2026, and operating margin went from negative 26.6% to positive 46.6%. That is what operating leverage on a fixed cost base looks like. Whether it is structural or cyclical is the open question.
Market position: the leading US federal vendor, with no disclosed share
Market position: Fairly Strong. Palantir holds the Pentagon's flagship AI program, taken over in 2019 after Google withdrew, and sells into agencies across the federal government. No competitor has displaced it from that position in five years. What is missing is a share number. Palantir does not publish one, and the submarkets it sells into, from defense data fusion to hospital operations software, have no authoritative sizing that would let an outsider compute it. The best available evidence is indirect: revenue of $4.5 billion in fiscal 2025, with the top 20 accounts averaging $64.6 million each, or roughly 29% of revenue, an estimate from the average spend the company reported for that group.
Stability: Fairly Strong. The share-stability rule of thumb, where a change of under 2 percentage points over five years suggests strong barriers and a change above 5 points suggests none, cannot be applied without a share series. The available proxies all point one way. Net revenue retention reached 157% in the second quarter of 2026, the company reported, so existing accounts spent 57% more than a year earlier before any new logo was signed. US commercial revenue grew 150% year over year in the same quarter. Those numbers show gains inside accounts.
They do not show everything. Net retention says nothing about churn in the long tail of smaller customers, and Palantir does not disclose gross retention, which is the figure that would show whether customers stay at all. A business adding logos quickly can hold revenue share while losing share of each account.
Profitability: a near-zero five-year average, then a step change
| Fiscal year | Revenue (USD bn) | Gross margin (%) | Operating margin (%) | Net margin (%) | ROE (%) |
|---|---|---|---|---|---|
| 2021 | 1.54 | 78.0 | -26.6 | -33.7 | -27.3 |
| 2022 | 1.91 | 78.6 | -8.4 | -19.6 | -15.1 |
| 2023 | 2.23 | 80.6 | 5.4 | 9.4 | 6.8 |
| 2024 | 2.87 | 80.2 | 10.8 | 16.1 | 10.8 |
| 2025 | 4.47 | 82.4 | 31.5 | 36.3 | 26.2 |
| 9M 2026 | 5.88 | 84.9 | 46.6 | 52.0 |
Profitability: Fairly Strong. The five-year average return on equity is close to zero, about 0.3% from fiscal 2021 through fiscal 2025. On the framework's scale, where 15% to 25% signals a clear advantage and 6% to 8% signals none, that average reads as no moat at all. It is misleading about the current business and useful as a caution. The losses came in 2021 and 2022, when operating expenses ran above revenue on a base of $1.5 billion to $1.9 billion. They disappeared as revenue scaled.
Fiscal 2025 ROE was 26.2%. The first nine months of fiscal 2026 produced $3.06 billion of net income on $11.23 billion of equity, an annualized rate above 50%, an estimate from reported quarterly figures. ROIC is harder to read because Palantir carries no debt and held $3.62 billion of cash against $13.41 billion of assets at September 30, 2026. On equity plus debt less cash, fiscal 2025 NOPAT of roughly $1.64 billion against about $6.0 billion of capital implies a return near 27%. Treat short-term investments as excess cash and invested capital shrinks toward zero, at which point the ratio stops meaning anything.
The cash backs up the accrual. Free cash flow was $2.10 billion in fiscal 2025 against $1.62 billion of net income, and about $3.3 billion in the first nine months of fiscal 2026. Part of that gap is stock-based compensation added back, a real cost settled in shares. No listed rival reports a comparable segment, so peer margins cannot be lined up directly; Morningstar, which rates the moat narrow, puts Palantir at the top of its software coverage on growth and margin combined.
Overall rating: Fairly Strong. Profitability carries the most weight, and it clears the bar comfortably on current numbers. Market position is strong where it can be observed, but the observation window is short. What keeps the rating below Strong is duration: consistent profitability dates only to 2023, and none of the excess return has yet been tested by a well-capitalized entrant selling an equivalent layer at a bundle price. A gross margin below 80% or net retention sliding toward 120% would point to customers finding a cheaper substitute. Another three to four years of returns at these levels through a full renewal cycle would support a higher rating.
Where the advantage comes from and how long it can last: switching costs, clearances and limited scale
Two of the candidate sources hold up, and both sit on the demand side or in the institution rather than in the product. The third, scale, is real but smaller than the revenue base makes it look. Brand and balance-sheet strength are not barriers here: $3.62 billion of cash and no debt let Palantir absorb losses and fund a long sales cycle, but capital is available to any credible competitor.
Switching costs inside the Ontology: Fairly Strong
The mechanism runs through deployment, not through the license. A Palantir install maps a customer's source data to objects that mirror how the operation actually runs (a hospital bed, a police case, a supply order) and writes results back into the source systems.
First, the model sits in the operational path. Analysts and staff work through the ontology, so replacing it means re-integrating the sources, rebuilding the object model and retraining the people who use it. The customer pays that cost, not the vendor.
Second, deployments widen on their own. Net revenue retention of 157% in the second quarter of 2026 means the installed base spent 57% more than a year earlier. Every added use case lengthens the list of workflows that break on migration, so the cost of leaving rises after the sale, not at it.
Third, search costs are high. The realistic alternative is usually internal IT, which Morningstar describes as patchwork dashboards that are hard to improve and costly to scale. To compare a rival product, a buyer has to rebuild a working system first.
None of that makes the position airtight. The ontology is a layer above systems of record Palantir does not own; the data sits in cloud warehouses and databases whose owners already hold the customer relationship. The retention evidence is one disclosed metric with no gross churn alongside it. Some captivity is contractual and revocable: the NHS Federated Data Platform contract carries a break clause the UK government can exercise in February 2027. And pricing power shows up mainly as expansion inside existing accounts, not as price held against a determined discount. The evidence so far only shows that customers expand; it does not show what they would do if a credible alternative arrived at half the price.
Clearances and procurement: Fairly Strong
Classified and sensitive work requires cleared facilities, cleared staff and a record of past performance, and each takes years to assemble. A cleared vendor running a live program therefore has an edge at every re-compete: the customer's alternative is to absorb the delay and the program risk of starting over. Palantir began running Maven in 2019 and has held positions across the armed services since, which is a reference no new entrant can manufacture.
The barrier changes the economics in a specific way. Renewals and expansions in this segment face less open competition, and procurement officers who already trust a system are slow to swap it. That supports the 83% to 87% gross margin and the 46.6% operating margin reported for the first nine months of fiscal 2026 more than any product claim does.
It is not exclusive. Systems integrators and defense primes hold clearances too, and the barrier is jurisdictional. US customers supplied 74% of fiscal 2025 revenue, and the same accreditations do not travel to allied governments without separate approval. Budget politics also cap what the position is worth: continuing resolutions, program cancellations and price pressure at renewal all bite, and a barrier that protects incumbency does not necessarily protect price.
Scale in research and forward-deployed staff: Moderate
Software fixed costs, from core platform engineering to the industry templates that shorten a deployment, spread across $5.9 billion of revenue in the first nine months of fiscal 2026. An 84.9% gross margin says serving the next customer costs little at the margin.
Relative share is what matters, and here the picture is mixed. Palantir is large next to specialist data-integration vendors, which is where the pricing power in its niche comes from. It is small next to the hyperscalers: Microsoft, Amazon and Google each spend more on R&D in a quarter than Palantir generates in a year. Palantir also puts engineers on customer sites, which adds cost with every account. A gross margin below 90% is consistent with shipping people alongside code, and that limits how far scale can lever the model.
Scale on its own would rate weaker, because an entrant with capital can buy its way to size. It earns a place here only because customer captivity already exists, so each increment of scale reinforces a barrier that is independently there.
Durability: substitutes are aimed at the layer, not the product
The clearest threat is not a rival data platform. It is a foundation model that infers an object model directly from a customer's schemas, which would erode the build work that justifies the ontology in the first place. Morningstar calls the ontology differentiated and hard to reproduce today, and still assumes growth normalizes beyond ten years because model capability is advancing quickly.
The second threat is packaging. AWS, Snowflake and ServiceNow already bill the buyer for storage and analytics, so a semantic layer sold as an add-on would compress Palantir's price without needing to be better. Morningstar names major cloud platforms as the main competitive tail risk. Third, open-source modeling frameworks give smaller buyers a good-enough option, and internal IT tooling keeps improving, which slowly raises the floor the patchwork alternative sets.
None of this has shown up in the numbers. Gross margin has risen every year since 2021, which is what a protected franchise looks like. That makes gross margin the first place a substitute would register: falling gross margin while revenue still grows would say a rival is setting the price.
Taken together, the switching costs hold for the life of a deployment, roughly three to five years, and weaken at each re-compete. Clearances last as long as the program and the classification regime, bounded to the US and its allies. Scale is the least durable of the three. The moat looks real for the next five years with reasonable confidence; over ten, it depends on whether the ontology remains a layer customers cannot cheaply rebuild, and the public evidence does not yet settle that.
Generated by DeepVal AI. Data and models may contain errors or be outdated. For research only; not personalized investment advice.
Continue reading the full report
- Comparison and summaryHow Palantir Technologies Inc. compares with its closest rivals, and our moat rating
Free to sign up · 20 research points · unlocking this report uses 10
20 points cover 2 full reports. No card needed.
Research reflects its generation date. Preview the first two modules.
More on Palantir Technologies Inc.
Palantir Technologies Inc.
Business, moat and key questions
Open company page