Intelligence analysts worked across information held in disconnected systems, with restrictions on who could see and use particular data. Useful analysis required bringing relevant information together while preserving those controls and human judgment. Existing databases and government IT projects supplied alternatives, but assembling them into a workable analytical environment remained a substantial implementation problem PL2 PL1.
Palantir
Palantir made difficult data projects usable, then built a commercial route around deployments and reusable software.
Arena: Market Conditions Before Palantir
What shaped the outcome
Step 1 of 4 · 2003–20 · Shared platform
Engineers connected data to work that customers needed done
Palantir’s first problem was getting useful analysis out of information held in separate systems with different permissions. A software license alone left much of that work with the customer. Co-founder Joe Lonsdale describes sending engineers into customer organizations, then incorporating what they built into the common product. That choice gave the supplier responsibility for whether an application worked in its actual setting. PL2
The distinction matters commercially. A customer evaluating a large internal development project has to carry integration risk before receiving useful software. Engineers who can connect existing systems and demonstrate a working use give the buyer something concrete to judge. The method is expensive; its economics improve only when later deployments can reuse earlier work.
EasyJet’s 2017 account of Skywise supplies a specific operational example. Its engineering and IT staff worked with Airbus on predicting technical faults, using a platform developed with Palantir. The trial put aircraft data into maintenance decisions before faults disrupted service. The immediate alternative was less-informed, more reactive maintenance. Airbus supplied aviation access and expertise, so this was a joint achievement. PL3
Marc Andrusko’s a16z analysis helps explain why simply hiring embedded engineers is insufficient. He argues that Palantir’s developed software components distinguish it from open-ended customization. We adopt that distinction: field work can create an application, but reusable software determines how much of the next application must be rebuilt. PL10
Rivals Customer integration or a new internal project leaves more implementation work with the buyer. EasyJet compares predictive with reactive maintenance, not Palantir with a named software finalist.
Step 2 of 4 · 2015–25 · Government
A usable product still needed a route through procurement
Government users and government purchasers did not always want the same buying process. The record in Palantir’s Army litigation includes operational requests for its existing software. Yet the Army pursued a development solicitation without adequately considering whether commercial products could meet the requirement. In 2018, the appeals court upheld the challenge to that process. PL4
The consequential choice was to contest the purchase specification. Demonstrating software to users could generate demand, but that demand would not produce a sale if procurement required a different kind of supplier. The judgment reopened consideration of commercial alternatives; it did not itself select Palantir as the best product.
By July 2025, the Army had consolidated 75 existing contracts into one purchasing framework with volume discounts. That changed the effort required to buy more software across programs. Its $10 billion ceiling was an option to purchase over as much as ten years, with no promise to spend the ceiling. The evidence of an installed government business lies in the contracts being combined. PL5
This history complicates a simple explanation based on privileged government access. Relationships may help a supplier understand a buyer, but the documented procurement obstacle was real. Palantir had to make commercial software eligible for consideration as well as make it useful to operators.
Rivals The 2015–2018 alternative was an Army development procurement; by 2025 the comparison was fragmented contracting versus a common framework.
Step 3 of 4 · 2019–25 · Shared platform
Palantir paid to prove value before accounts became profitable
Palantir accepted losses on early deployments. Its 2020 filing separates accounts into acquisition, expansion and scale phases and describes pilots funded by the company. This transferred some evaluation risk from the buyer to the seller, while giving Palantir access to problems that could support larger contracts. PL1
The financial tradeoff was substantial. Accounts categorized as being in acquisition at the end of 2019 produced only $0.6 million of that year’s revenue against a $65.4 million contribution loss. Those same accounts generated $18.8 million in the first half of 2020 but still lost money on that measure. Conversion created revenue before it necessarily created a profitable account. PL1
Beth Kindig’s IPO analysis emphasized the difficulty of scaling a specialized, labor-intensive business. That is a stronger challenge than assuming every pilot eventually pays back. Our reading keeps the concern and separates it from the article’s sweeping conclusion about product-market fit. Customers can obtain useful software while the supplier still has poor acquisition economics. PL11
The later business did achieve aggregate profitability: FY2025 brought $4.48 billion in revenue and $1.41 billion of GAAP operating income. U.S. commercial revenue was $1.47 billion, alongside $1.86 billion from U.S. government customers. Those results weaken the claim that the model could never scale profitably. They leave open which cohorts, product changes and cost decisions did most of the work. PL6
Rivals An upfront paid implementation places more initial financial risk on the buyer. This is an economic reconstruction, not a matched rival-price comparison.
Step 4 of 4 · 2023 · Commercial
Existing workflows gave AI a place to be used
A connected operational system gives a new analytical tool a place to act. Customer-specific definitions, permissions and applications can be reused when another use is added; they also create work for a replacement supplier. Palantir’s value to an established customer therefore includes the functioning deployment around the software. PL1
That position helps explain the appeal of applying AI through existing systems. Andrusko describes AIP as connecting models with an organization’s data and operations. In this account, model access is one input. The surrounding software determines which information is available and how an output can enter a controlled workflow. PL10
There is an important limit to the incumbent’s defense. A customer can remain because its applications work well, because migration is burdensome, or because a current contract is convenient. Those reasons have different implications for a challenger. A supplier that offers a sufficiently better outcome can justify rebuilding integrations. The usefulness of the installed system establishes a plausible obstacle, rather than a measure of how many customers it retains.
AI demand is also an alternative explanation for recent commercial growth. It can raise spending across suppliers without proving that Palantir alone has a superior model. The earlier maintenance deployment and procurement record explain capabilities that predate the AI boom. They make the later growth intelligible without treating a new category’s expansion as evidence of permanent competitive protection.
Rivals Rivals can supply models or alternative data platforms; replacing an operational deployment differs from substituting a model.
Key dates
- 2003–2008Build for intelligence work
- 2015Users request an existing capability
- 2017-06-20Put aircraft data into maintenance decisions
- 2018-09-13Reopen consideration of commercial software
- 2019–H1 2020Carry early deployment losses
- 2020-09-30 retrospectiveReturn customer work to a shared product
- 2023–2024; analyzed 2026Apply AI through operational software
- 2025-07-31Consolidate government purchasing
- 2025; reported 2026-02-02Reach aggregate operating profitability at larger scale
Sources
Oldest first.
- PLX-3 Military leaders urgently push for new counterterrorism software. Trade report on a DoD memo
- PLX-4 The War Over Soon-to-Be-Outdated Army Intelligence Systems. Trade report on court filings
- PL3 easyJet deploys Airbus Skywise platform for gains in predictive maintenance. Customer announcement
- PLX-10 Raytheon, Palantir win $876M battlefield tech contract. Trade report
- PL4 Palantir USG v. United States, 17-1465. Court judgment
- PLX-2 Palantir's Army win is important, but not quite the watershed moment. Trade analysis
- PLX-1 Palantir — who successfully sued the Army — has won a major Army contract. Trade report
- PL1 S-1/A registration statement. IPO filing
- PL11 Palantir IPO: Deep Dive Analysis. Public-market investor analysis
- PL2 Why We Built Palantir. Founder retrospective
- PLX-9 Army Awards Palantir $823M Contract For Enterprise 'Data Fabric'. Trade report
- PL5 Army awards enterprise service agreement. Buyer procurement announcement
- PLX-11 Palantir CEO Alex Karp credits contrarian strategy. Trade report of CEO remarks
- PL10 The Palantirization of everything. VC analysis
- PL6 Fourth-quarter and full-year 2025 results. Earnings release
- PLX-8 Palantir Q4 2025 earnings call transcript. Earnings call transcript
- PLX-7 DOD components face 'aggressive' timeline for Maven Smart System transition. Trade report on a DoD memo
- PLX-5 Peter Thiel transcript. Founder interview
- PLX-6 The Patriot: Shyam Sankar of Palantir. Executive profile with first-person quotes