Make the decision the data was waiting for.

The problem Palantir says it solves

Palantir's own description starts with data that cannot be used where decisions are made. Its documentation calls the Ontology "an operational layer for the organization" that connects datasets to their real-world counterparts, from plants and equipment to customer orders and financial transactions. The same logic explains how it sells. Its annual filing for 2025 says Palantir is "fundamentally competing with the internal software development efforts of our potential customers." Read plainly, the first alternative to buying Palantir is a customer's own engineering team building the connection itself, not another vendor.

The belief that follows is that this work is done best by engineers who sit with the customer. Palantir calls them forward deployed engineers, FDEs for short. The Seoul engineering postings say the company pioneered the role, and the experienced one says it "isn't just a job title: it's the blueprint." The annual filing says "We embed directly with customers across numerous industries." The chief technology officer put the claim more bluntly on the Q2 2026 earnings call.

Remember, only Palantir has FDEs. Everyone else has sparkling sales engineers.

Shyam Sankar · Chief Technology Officer, Palantir · Q2 2026 earnings call · 2026-08-03

Why now: bootcamps and model control

Palantir gives two reasons this works at scale now. The first is the bootcamp. Its own overview describes sessions where customers "go from zero to use case in one to five days," building live alongside Palantir engineers, and its annual filing says bootcamps "deliver real workflows on actual customer data in days." The same filing says a sales cycle often lasts six to nine months and can run a year or more. Read together, a bootcamp lets a customer judge the product on its own data before that long process ends.

The second reason is language models, and who controls them. Palantir argues that enterprises now want to own the data, logic, actions and security around any model they use, rather than hand them to a model lab. That is the job it gives AIP. The number to watch is US commercial revenue, recognized in the quarter: $214M in Q4 2024, $306M in Q2 2025 and $764M in Q2 2026. Palantir presents bootcamps as part of how it wins customers.

Demand for AI sovereignty has now been unleashed. And Palantir is the only company that has demonstrated it can transform tokens into actual economic value.

Alex Karp · Co-founder and Chief Executive Officer, Palantir · Q2 2026 earnings release · 2026-08-03

What the work looks like, step by step

Take a manufacturer with a late part, as an illustration and not a real customer's project. First, data engineers and the customer's system owners bring inventory, order and schedule records into Foundry and reconcile the identifiers, keeping track of which system each number came from. Second, engineers and planners define in the Ontology the things the planner manages, such as parts, suppliers and production orders, and the actions allowed on them. Third, AIP can query that context and propose a response through tools connected to the Ontology, while domain experts test realistic failure cases. Fourth, a workflow application turns the chosen response into an authorized change to the plan, and Apollo keeps the software delivered in the cloud, on premises or in more rugged environments.

In this illustrative workflow, people agree on object definitions, test AI failures and approve the proposed change. That final human approval is a choice in the example. Palantir's Ontology documentation describes actions and granular security and governance for changes. Organizations configure who can access objects and execute actions; the documentation does not make a named person's approval mandatory for every AIP action.

In Seoul this chain is the job. The engineering postings put engineers with the customer and in front of stakeholders from technical teams to executives, the strategist postings send strategists onsite to customer analysts, and every posting requires fluent Korean and English. The first test of the idea in Korea is therefore human: can you agree with a Korean planner or plant manager on what a part, a shipment or a quality defect means, and then build the application that acts on it? Chapter 02 shows what Palantir sells, chapter 04 whether the numbers hold, and chapter 06 what is known about Korea.

Three products that stack

Palantir's architecture documentation describes three integrated platforms that stack. Foundry is the base, AIP attaches language models to it, and Apollo keeps the software delivered. The annual filing lists a fourth, Gotham, used across government functions, which this chapter leaves aside. The filing calls Foundry the "foundational data operations platform" and describes the Ontology as "the heart of our platforms."

Foundry integrates an organization's data, such as datasets, virtual tables and models, and maps it in the Ontology into objects, properties and links, which Palantir calls the semantics of the organization. Action types and functions define what people are allowed to change, which it calls the kinetics. Teams then build analyses and applications on top, and the Workshop screen in the documentation is one example, built on demonstration data. Buyers are operations leaders, data teams and application builders. Palantir says its first rival is the customer's own software team, and describes the other rivals by category.

AIP connects language models, tools and agents to the Ontology so a model can look up an organization's own data and propose actions within the controls that organization has set. Teams can test how the model behaves before it reaches users. Palantir sells it through bootcamps, where customers build use cases live with its engineers in one to five days. The screen shown is AIP Chatbot Studio from Palantir's documentation, with a demonstration chatbot built on Federal Reserve press conference transcripts. Palantir frames its rival here as the language labs behind the models, and its CEO has said customers "have declined to become vassal states of the language labs."

Apollo is the delivery side of the platform. Palantir's architecture documentation describes it as managing continuous delivery of software across different environments, so applications stay current where a customer's data and operations sit. The documentation screens show an API explorer and a documentation explorer for queries and types. Buyers are platform and software operations teams.

One customer, and the limits of what is reported

HD Hyundai is the best-documented customer in Korea. The work began in 2021 at HD Hyundai Oilbank, the year HD Hyundai also started its Future of Shipyard project, which aims by 2030 for a largely automated shipyard built around a digital twin. In 2022 two shipbuilding affiliates ran pilots on Palantir's platform, starting in January and May, joining the data they had already accumulated to look for ways to improve design and production. In September 2022 the group signed a Foundry contract covering its shipbuilding affiliates, valued at $20M over five years, which Palantir said brought the total value of its contracts with the group to $45M over five years. Korean trade press describes the software being used for schedule management on ship construction and for material management, financial planning and equipment operation.

Together with Palantir Technologies, (we are) drawing on the latest innovation in AI, digital twins, and so much more to transform our shipyards, unlocking a whole new level of productivity and safety

Chung Ki-sun · Executive Vice Chairman, HD Hyundai · Palantir's customer video, as reported in Korean English-language press · 2025-02-28

What was reported afterwards is a claim, not a measurement. The companies say HD Hyundai is manufacturing ships around 30% faster with Palantir software. A separate report gives a 2030 target of 30% higher productivity. In January 2026 the companies announced a group-wide expansion, which Palantir called its largest and longest partnership of its kind in Korea. Its value was reported as hundreds of millions of dollars over several years, but that figure came from an unnamed person and Palantir declined to comment on terms. The announcement also names areas where HD Hyundai Oilbank saw results. Distinguish the company's 30% performance claim, anonymous-source reporting of January contract value, and the $45M contract value announced in 2022.

A second named customer: Airbus

A further customer example is Airbus. Palantir and Airbus began collaboration on A350 manufacturing in 2015 and developed the Skywise aviation platform. Palantir says more than 50,000 people used Skywise in daily operations as of the February 2026 renewal. The initial collaboration date and the later platform adoption are separate milestones.

HD Hyundai announced group-wide expansion in January 2026, while LG CNS announced broader LG Group adoption plans in March. Announced expansion and live deployments are separate stages.