What Palantir AIP and Foundry Cost a European Mid-Cap
The short answer
Palantir publishes no real list price for Foundry: its AWS Marketplace listing shows $100,000 a month per Foundry Subscription Unit and calls that figure a placeholder, with access by private pricing only. The most detailed public prices are Palantir’s UK government listings, where G-Cloud 14 priced Foundry and AIP at £3,000,000 a year for a single-organisation licence, with platform usage of the same value included.
That is a sterling list price for UK public bodies, not a commercial quote, and Palantir’s 2026 G-Cloud 15 listings drop the headline price altogether. Palantir’s filings show which customers it is built around: its 10-Q for the quarter to 30 June 2026 puts average revenue from its top twenty customers at $124 million over twelve months, and its 10-K for 2025 says it “generally targets large-scale, hard-to-execute opportunities at large government and commercial institutions”. A company with EUR 250M to 2B of revenue can still buy Foundry and AIP. It will pay in four lines, most of them with no commercial price you can check.
Key takeaways
- The cost is four lines, not one price. A platform licence per organisation; capability licences per outcome, with some Palantir engineering included; usage beyond the contracted quantity, with language-model tokens converted into compute-seconds; and the integrator’s fees, if you come through one.
- Per-organisation pricing does not shrink with revenue. At list price a smaller company narrows the bill by buying fewer capability licences, and gives up the company-wide layer Palantir does best.
- The £66,000 per core still quoted online is a legacy price. It sits in Palantir’s 2026 pricing document, filed for customers on older terms.
- The bootcamp is free and decides a lot. You negotiate the production licence with a working system already built on Palantir’s platform.
- Ask what you own at exit before you sign. Data exports in non-proprietary formats. The Ontology leaves as JSON for re-import into Foundry or, in beta, as TypeScript definitions: neither is RDF or OWL, and both become a working Ontology again only inside Foundry.
What does the Palantir Foundry license cost on G-Cloud?
Palantir’s G-Cloud 14 listing priced Foundry and AIP at £3,000,000 a licence. Its pricing document defines that as a per-annum licence for one environment and one organisation, with the same value of platform usage, to be used in the year it is bought.
| Framework | Palantir listing | Headline price on the page | What the unit means | Source |
|---|---|---|---|---|
| G-Cloud 11 | Palantir Foundry | £66,000 per unit | A server core per year, six-month minimum, as the G-Cloud 14 and 15 pricing documents define it | Archived August 2020 |
| G-Cloud 13 | Palantir Foundry | £3,000,000.00 a licence | Annual, one organisation, usage included | Archived September 2024; live page removed |
| G-Cloud 14 | Palantir Platform: Foundry and AIP | £3,000,000.00 a licence | Annual, one environment, one organisation, usage included, no roll-over | Archived April 2026 |
| G-Cloud 15 | Palantir Platform: AI, with separate support listings | None on the page | Platform, capability and sustain licences; per-core licences, usage packs and per-person engineering filed as legacy | Live listing |
The £66,000 figure is the one still in circulation: InfoWorld’s August 2026 piece on Foundry quotes roughly 66,000 pounds per server core per year and calls the cost hard to justify for smaller organisations. That price is still in Palantir’s 2026 pricing document, under legacy services for customers on older terms, so it describes the model Palantir has moved away from.
Every figure here is a UK public-sector list price in sterling. We have not converted them to euros, because a commercial buyer is not quoted that list and a converted figure would look more precise than it is. The links are archived copies because GOV.UK gives 28 October 2026 as the expiry date of G-Cloud 14.
What is Palantir’s AIP pricing model?
A negotiated agreement, generally one to five years according to Palantir’s 10-K. Palantir’s 2026 UK terms build it from a platform licence plus a licence per outcome: its G-Cloud 15 service definition lists four kinds: platform licences for the infrastructure and tooling, capability licences for “a specific outcome or set of outcomes”, sustain licences for capabilities no longer being developed, and third-party licences, covered below. Per-person engineering, usage packs and per-core licences remain on the price list as legacy services.
The nearest public price for a single capability is American. A FedRAMP private-offer listing resold by Carahsoft shows $2,200,000.00 for a 12-month licence to Palantir’s Financial Operations Module, capped by interfaces, schemas and users. The unit is a module with caps, not a price per user.
Usage sits inside the contract, which Palantir’s AWS listing says “entitles you to a specified quantity of use”, with overages billed on top. Foundry meters that use in compute-seconds, and AIP converts language-model tokens into compute-seconds per model and per region, with a combined EU/UK rate. The compute page lists more than a dozen providers, from Anthropic, Google, Meta, Mistral, OpenAI and xAI to DeepSeek, Qwen and NVIDIA. AIP also accepts your own registered or self-hosted models, which Palantir recommends where its models cannot be used for legal or compliance reasons, though some features, such as AIP Assist, do not support them. The published conversions apply only to AWS-hosted enrollments under certain default contract terms, so your contract sets the rate.
In practice the lines arrive bundled. Customers in a 2023 Forrester study Palantir commissioned described one annual payment covering Foundry, the cloud underneath it and Palantir’s professional services. On G-Cloud 14, included usage lapsed at year end whether or not your teams used it.
Who does the work: Palantir engineers, Accenture and the AI FDE
Palantir’s engineers, often an integrator’s, and in one case an agent. Palantir pioneered the forward-deployed model and still sells engineers inside the licence: its G-Cloud 15 listing includes varying levels of implementation engineering support in capability licences, and a separate ongoing support listing sells dedicated engineering resources on their own. As one customer told Forrester: “This team comes with the contract.”
One route a European mid-cap may meet adds a layer. In December 2025, Palantir named Accenture a preferred global partner, and the two formed the Accenture Palantir Business Group, staffed with Palantir forward-deployed engineers and more than 2,000 Palantir-skilled Accenture professionals, with manufacturing, consumer goods and financial services among its target industries.
The integrator does not replace Palantir’s licence. Under “Third Party Licenses”, the G-Cloud 15 service definition says capabilities a customer or contractor develops on the platform, where Palantir is not contracted to build them, are scoped for a corresponding licence. What Accenture, Deloitte or your own team builds on Foundry can still need a Palantir capability licence.
The integrator’s own price comes by the day. Deloitte’s G-Cloud 15 listing for Foundry build and support prices most of the work on a day-rate card. Its pricing document, which it calls illustrative until the work is scoped, starts the standard card at £765 a day, and an offshore card, available where the buyer consents to offshore staff, at £355. Who owns the code depends on which part of Deloitte’s G-Cloud 15 supplier terms covers the work. The implementation terms by default make you the sole owner of the IP in the deliverables once you have paid in full and accepted them, subject to Deloitte’s background IP. The consulting and technology terms keep that IP with Deloitte and grant you a licence to use it once you have paid. The platform is Palantir’s either way, and a day rate leaves the delivery risk with you. The NHS Federated Data Platform, Palantir’s seven-year UK contract worth £330 million with £25.6 million expected in the first year, was supported by Accenture and PwC.
Nor is the integrator indifferent between platforms. In March 2026, Accenture completed its purchase of Faculty, a UK Palantir rival, made Faculty’s chief executive its CTO and added Faculty’s decision-intelligence product, Frontier, to its suite. The partner advising you on Palantir also owns an alternative to it.
One naming trap: Palantir’s “AI FDE” is not a person. Palantir’s documentation describes it as an agent that operates Foundry through conversational commands, building data pipelines, editing Ontology objects and actions, and writing functions. The human role is covered in what a forward-deployed engineer is and how Accenture, Deloitte and EY run it now.
How much does a Palantir AIP bootcamp cost?
Little or nothing up front. Palantir publishes no bootcamp price, and its 10-Q says it conducts pilots and bootcamps “generally at our own expense and without a guarantee of future returns”. The cost arrives in what follows.
In 2023 Palantir reoriented its go-to-market around bootcamps, which its Q3 2023 earnings call said deliver real workflows on actual customer data in five days or less, against one to three months for a traditional pilot. The speed is real.
Palantir’s G-Cloud 15 pricing document sets out the sequence: trial licences at a nominal cost, paid discovery packages of up to three months, pilot licences on a term basis, then annual platform and capability licences. By the time a production price is on the table, your workflow, its objects and its actions are modelled in Palantir’s Ontology by Palantir’s engineers, and you negotiate with a working system already built on the product you are buying. That is what a free bootcamp costs. Decide before it starts what would make you walk away.
The free Developer Tier gives capped access to Foundry and AIP in the United States and select countries, per the 10-K. It is a developer sandbox, not an enterprise licence.
What would Palantir AIP and Foundry cost a EUR 250M to 2B company?
Nobody outside a negotiation can give you a euro figure, and we will not invent one. The public record shows the shape. Palantir’s UK platform licence is priced per organisation, not per user, so at list price it does not shrink with your revenue: a EUR 400M manufacturer and a government department face the same platform headline.
What gets smaller is scope. Capability licences are bought per outcome, with pilot licences before them, so a mid-cap can commit to one or two outcomes rather than a company-wide programme. The trade-off is coverage: Palantir does best as one operational layer across the whole company, and a narrow footprint still carries the full platform licence.
Palantir’s own capacity points elsewhere. US customers produced 80% of its revenue in the first half of 2026, and no other country reached 10%, according to the 10-Q. In February, Karp said Palantir “really doesn’t have the bandwidth to do anything that’s difficult outside of America”, as Fortune reported. That is one reason a European mid-cap may meet Palantir through an integrator such as Accenture.
So the four lines apply at any size: a platform licence per organisation; capability licences per outcome, with some Palantir engineering included; usage beyond the contracted quantity, with language-model tokens converted into compute-seconds; and the integrator’s fees, if you come through one.
What is Palantir genuinely good at?
The Ontology, the forward-deployed model and speed to a working system.
Palantir’s documentation describes the Ontology as an operational layer for the organization, and argues that it represents the decisions in an enterprise, not simply the data. It models objects, links and the actions people take on them, which is closer to how a business runs than a set of tables. Palantir built the forward-deployed engineer into a delivery model years before the AI labs bought their way into it in 2026, and its bootcamps put a use case on real data in days.
Some terms are buyer-friendly too. Palantir’s G-Cloud 15 listing says its platforms are “purposefully designed to prevent vendor lock-in”, and the 10-Q says Palantir lets many customers terminate for convenience before the end of the stated term, with less than twelve months’ notice. Karp makes the ownership argument himself: in the Q2 2026 results he said customers’ competitive advantage “should never become the training data for future models”.
Palantir’s Ontology is one of four ways to hold the company model
The model of your company, meaning its objects, relationships, rules and actions, has to live somewhere, and four platforms now offer to hold it. Palantir’s Ontology lives inside Foundry. Microsoft has an ontology item in Fabric IQ, still in preview, which maps business terms onto data in Fabric and OneLake, and whose rules and metrics export only as TMDL with Fabric-only extensions. Databricks calls its Unity Catalog semantics the human-modelled layer of its “Genie Ontology”. Snowflake holds business meaning in semantic views and exports them, in preview, as Apache Ossie (incubating) YAML, which it describes as “an open standard for representing semantic models”; that export replaces its Open Semantic Interchange export, now deprecated.
Where will the model of your company live, and whose platform must you license to keep using it? Palantir’s answer is the most developed of the four, because it models actions as well as data. It is also an answer that runs only on Palantir. A fifth option is to keep that model in a context engine your company owns: the model-agnostic intelligence layer, built on whichever of these platforms you already run, holding what no warehouse ontology captures, such as the unwritten procedures and the handoffs between departments.
What do you own when the Palantir contract ends?
Your data, in non-proprietary formats, if the contract says so. The Ontology and your functions leave as definitions and code, not as a running system: only Foundry turns them back into one. Get both in writing before you sign.
The G-Cloud 15 listing says all data in the platform can be securely exported in non-proprietary formats, with publicly documented APIs at every tier; get that written into the contract. It says nothing about the logic: object types, actions, functions written in Palantir’s tooling and the Workshop applications your teams open every morning, which is where a bootcamp and two years of engineering put your company’s knowledge.
Palantir’s documentation answers part of that. Ontology Manager exports the Ontology’s working state as a JSON file for importing back into the platform, with a warning: “You should not depend on the exported JSON schema as it may change over time.” Ontology-as-code, in beta and not available on every enrollment, keeps “TypeScript definitions of your object types, interfaces, actions, and other entities” in a SuperRepo that Palantir says “can be hosted, built, and deployed outside of the platform”, but those definitions are “compiled and materialized into real entities on your enrollment”. Functions are written in TypeScript or Python. None of it is RDF or OWL, and none of it is a working Ontology outside Foundry, so moving to another platform means re-expressing the model and rebuilding the actions. The Workshop applications fall outside all three, so ask about them in writing. Owning your intelligence depends on where that logic can run more than on the data clause.
Where the data sits is a separate question from who can compel access to it. The same listing offers storage and processing in the UK and the EEA, and the seller is Palantir Technologies UK, Ltd., but the listed parent, Palantir Technologies Inc., is incorporated in Delaware, per the 10-K. Ask which Palantir entity you contract with, which entity processes the data, and which subprocessors sit underneath; the listing says datacentre security is managed by a third party. Count the data protection impact assessment and the legal review as part of the cost.
Airbus shows a European company making that call platform by platform. In July 2026, The Register reported, it began moving 70 of the 900 applications it treats as most critical from AWS to the French provider Scaleway, so that the data stays “under European control”. Skywise, the data platform Airbus runs with Palantir, stays on AWS. Airbus’s head of digital said it balances such choices “based on the criticality of the data”.
Europe also gives you a lever a US buyer does not have. The EU Data Act will entirely remove switching charges, including charges for data egress, from 12 January 2027. The data it covers includes metadata generated by your use of the service, but not data protected by the provider’s intellectual property or trade secrets; which side of that line an Ontology falls on is a question for counsel. Article 31 exempts services mostly custom-built for one customer, and not offered at broad commercial scale through the provider’s catalogue, from some of these rules, and under Article 31(3) the provider must tell you before signing which obligations do not apply. Ask for that statement. The Data Act lowers the fee to leave. It does not lower the cost of rebuilding the logic somewhere else.
Total cost: a Palantir licence vs a build that leaves you the context engine
The table sets Palantir’s cost lines, as far as public sources show them, against the integrator route and against an outcome-tied build in which the client owns the context engine. The right-hand column describes who pays for and owns what, not a price.
| Cost line | Palantir direct | Palantir through Accenture or another integrator | Build that leaves you the context engine |
|---|---|---|---|
| Licences | Annual platform licence per organisation (£3,000,000 headline on G-Cloud 14), plus a capability licence per outcome | The same, plus a Palantir capability licence for what the integrator builds | No platform licence |
| Usage and tokens | A contracted quantity of use in compute-seconds, overages on top; tokens at contract rates | The same | Your own cloud and model providers, at their rates |
| Engineering | Palantir engineers inside capability licences; per-person engineering filed as legacy on G-Cloud 15 | The integrator’s day rates; Deloitte’s G-Cloud 15 standard card starts at £765 a day | A fixed fee to build, scoped against what the workflow costs today; part of it tied to the outcome where that can be attributed |
| Running cost after go-live | Platform licence renewal, plus sustain licences for capabilities no longer being developed | The same, plus the integrator’s support fees | A recurring fee to run and evolve the system and keep the context engine current, plus your own cloud |
| Entry path | Bootcamp at Palantir’s expense, then paid discovery, pilot licence, annual licences | The Palantir sequence, plus the integrator’s fees once its people are on the work | One workflow and the number it should move, agreed and priced before the build starts |
| Price visibility | Placeholder marketplace price; real price private | Day rates on G-Cloud; licence price private | No rate card; fixed build fee agreed before the build starts |
| Who owns the company model at the end | You own the data; the Ontology logic lives in Foundry | Who owns the code depends on the integrator’s terms; the platform is Palantir’s | Your company |
| Who runs it | Your users on Palantir’s platform, with Palantir engineers | Your users, with the integrator’s and Palantir’s engineers | Your team, with the build partner on a recurring run-and-evolve fee |
| Model choice | Many providers at Palantir’s conversion rates, or your own registered or self-hosted models, with some AIP features unsupported | The same | Chosen per workflow, changed when a better one ships |
| Cost to leave | Data export per the listing; logic rebuilt elsewhere; Data Act switching charges end 12 January 2027 | The same | No platform licence to exit |
The right-hand column is not free, and on day one it is not faster. You carry more of the build, you pay to keep the system running and current whether your own team runs it or the build partner stays on, and nobody shows you a working use case in five days before you sign. What you get is a model of your company held in an asset you own, on infrastructure you already pay for, with models you can change underneath. Build vs buy vs embed compares delivery models, and how an engagement works sets out ours.
When is Palantir the better choice?
When speed across many systems matters more than owning the model of your company. Palantir is the better buy when:
- you need an operational layer across many messy systems in production quickly, and a bootcamp on your own data is the fastest proof available;
- you have budget at the scale the platform licence assumes, and enough outcomes to spread it across;
- you do not plan to build or keep a platform team of your own;
- your customers or partners already run on it, as selected Airbus suppliers have done since Airbus opened Skywise to them in 2018;
- the work spans several organisations or a regulated setting where Palantir’s governance is already accepted, as it is in parts of the UK public sector;
- you are comfortable that the model of your company lives in Palantir’s platform.
Choose a build that leaves you the context engine when that model is what makes your company hard to copy, and you want to keep it, run it and change the models underneath.
FAQ
How much does Palantir Foundry cost? Palantir publishes no real commercial list price. Its AWS Marketplace listing shows $100,000 a month per Foundry Subscription Unit but calls that figure a placeholder, with access by private pricing only. The most detailed public prices are Palantir’s UK G-Cloud listings: on G-Cloud 14, £3,000,000 a year for a single-organisation licence of Foundry and AIP with platform usage included. Palantir’s G-Cloud 15 listings show no headline price and sell platform, capability and sustain licences instead.
How much does Palantir AIP cost? AIP has no separate public price. It is sold on the same Palantir platform agreement as Foundry, and its language-model tokens are converted into compute-seconds, the unit Foundry already meters, at rates Palantir sets per model and per region. Palantir says its published conversion rates apply only to AWS-hosted enrollments under certain default contract terms, so your contract sets the rate.
What is the Palantir Foundry license cost on G-Cloud? Palantir’s G-Cloud 14 listing priced Foundry and AIP at £3,000,000 a licence, which the attached pricing document describes as an annual, single-environment, single-organisation licence including the same value of platform usage, to be used within the year. GOV.UK gives 28 October 2026 as the expiry date of G-Cloud 14. Its successor, G-Cloud 15, sells platform, capability and sustain licences, and keeps per-core licences and usage packs as legacy services.
What is Palantir AIP’s pricing model? A negotiated agreement, generally one to five years. In its 2026 UK pricing, Palantir sells a platform licence per organisation, capability licences for a specific outcome or set of outcomes with varying levels of engineering support included, and sustain licences for capabilities no longer being developed. Usage is metered in compute-seconds, with language-model tokens converted at a combined EU/UK rate alongside North America and other regions. Per-person engineering and usage packs are filed as legacy services.
How much does a Palantir AIP bootcamp cost? Palantir publishes no bootcamp price, and its filings say it runs pilots and bootcamps generally at its own expense. The cost comes afterwards: Palantir’s UK pricing document lists trial licences at a nominal cost, paid discovery packages of up to three months and pilot licences as the steps before annual licences. A free Developer Tier gives capped access in the United States and select countries, but it is not an enterprise licence.
Is Palantir’s AI FDE a forward-deployed engineer? No. Palantir’s documentation describes AI FDE as an agent that operates Foundry through conversational commands, building data pipelines, editing the Ontology and writing functions. The forward-deployed engineers you pay for are people, and under Palantir’s 2026 UK licensing varying levels of their support are included in capability licences.
When is Palantir the better choice? When speed to production across many systems matters more than owning the model of your company, or when the customers and partners you work with already run on Palantir.
Can we take our data with us if we leave Palantir? Palantir’s G-Cloud listing says all data in the platform can be exported in non-proprietary formats. Get that written into the contract. The logic leaves as definitions and code, not as a running system: Palantir’s documentation says the Ontology exports as JSON for re-import into Foundry, with a schema that may change, Ontology-as-code (in beta) keeps TypeScript definitions of object types and actions in your own repository, and functions are written in TypeScript or Python. None of it is RDF or OWL, and the definitions become a working Ontology only on a Foundry enrollment. In the EU, the Data Act removes switching and data-egress charges from 12 January 2027, though Article 31 exempts services mostly custom-built for one customer. That lowers the fee to leave, not the cost of rebuilding the logic.
Related reading
- What is a forward-deployed engineer, and why enterprise AI needs one
- Accenture, Deloitte and EY all run forward deployed engineering now: what actually differs
- Forward-deployed engineer vs consultant vs system integrator: a decision rule for buyers
- Build vs buy vs embed: choosing an enterprise AI delivery model
- Own your intelligence: what it actually takes
- OpenAI’s DeployCo, Anthropic’s Ode, and the case for an independent forward-deployed partner
Nucleo is an AI-native transformation company for European mid-caps. Our forward-deployed engineers build a context engine your company owns, model-agnostic and under EU governance, and put agents into production on top of it. See how an engagement works, or write to us.
Sources: Palantir Platform on AWS Marketplace; Palantir Technologies Inc. (FedRAMP) private offer on AWS Marketplace, sold by Carahsoft; Palantir Platform: Foundry and AIP, G-Cloud 14 (archived 18 April 2026); Palantir Foundry, G-Cloud 13 (archived 14 September 2024); Palantir Foundry, G-Cloud 11 (archived 3 August 2020); Palantir Platform: AI, G-Cloud 15; Palantir Platform, G-Cloud 15 service definition document; Palantir Platform: Cloud Support Services, Ongoing Support, G-Cloud 15; Deloitte, Cloud application design, build and support on Palantir Foundry Platform, G-Cloud 15; Deloitte, G-Cloud 15 pricing document for the same service; Deloitte LLP Supplier Terms for the G-Cloud 15 Framework; GOV.UK, G-Cloud relaunches with biggest upgrade in its history; Palantir Q2 2026 results, Form 8-K Exhibit 99.1; Palantir Form 10-Q, quarter ended 30 June 2026; Palantir Form 10-K, fiscal year 2025; Accenture and Palantir expand global strategic partnership; Accenture completes acquisition of Faculty; The Register, British Palantir rival sells to Accenture; The Register, Palantir wins NHS data platform contract; Fortune, Karp says Europe and Canada are falling behind on AI adoption; Palantir docs, usage types; Palantir docs, compute usage with AIP; Palantir docs, bring your own model to AIP; Palantir docs, AI FDE overview; Palantir docs, Ontology overview; Palantir docs, why ontology; Palantir docs, export, edit, and import an Ontology; Palantir docs, SuperRepo core concepts and Ontology-as-code; Palantir docs, functions overview; Palantir Q3 2023 earnings call transcript, The Motley Fool; Palantir Developer Community, Developer Tier billing and usage; Forrester, The Total Economic Impact of Palantir Foundry, commissioned by Palantir; Microsoft Learn, what is ontology in Fabric IQ; Databricks, Unity Catalog semantics; Snowflake, overview of semantic views; Snowflake, the Apache Ossie YAML export; Snowflake, the deprecated OSI YAML export; European Commission, Data Act explained; Regulation (EU) 2023/2854, the Data Act, on EUR-Lex; The Register, Airbus migrating 70 critical apps from AWS to Scaleway; Airbus, Airbus extends Skywise to suppliers; InfoWorld, what Palantir Foundry did that surprised me.