Palantir Foundry Pricing: What It Costs, How Compute Seconds Work, and Real Price Points

Last updated: October 8, 2026

By Omar J, Director, PAL4C Labs

TL;DR Palantir doesn’t publish a Foundry price list. You negotiate an annual platform fee, usually on a 1 to 5 year term, and usage is metered in compute-seconds and storage. AIP turns LLM tokens into compute-seconds too. Public anchors run from a free Developer Tier to a $2.2M a year federal module on AWS Marketplace. Budget for integration and enablement on top of the license, because that’s where surprises live, and it’s the part our Palantir Foundry and AIP implementation team scopes first.

$2,200,000. That’s the only Foundry-family price tag you can find on a public storefront right now, and it’s for one narrowly scoped federal finance module with a 200-user cap. Not exactly a menu.

Everyone shopping for Palantir Foundry pricing hits the same wall. The sales team asks about your use case before it’ll discuss numbers, the analyst sites repeat ranges nobody can source, and one widely shared figure (more on that below) is a 2014 government line item for a different product. So buyers walk into the first commercial call blind.

We implement Foundry and AIP for a living, so we see contracts, compute bills, and the costs that land after signature. This post sticks to what can be verified. Palantir’s own documentation on how usage gets metered, its SEC filings, and the handful of public listings that actually carry a number. Then it covers the part nobody prices in, which is everything outside the license. If you’re still working out whether Foundry is the right platform at all, our Foundry implementation services page explains how we scope that question before money changes hands.

Why Is There No Palantir Foundry Price List?

Palantir sells Foundry as a negotiated platform subscription, not a per-seat product. Price depends on scope, deployment, term, and expected usage, so it’s set per contract. Its SEC filings describe terms of one to five years.

The company says as much in its own words. Palantir’s Q2 2026 quarterly report states that it generally offers contract terms of one to five years and recognizes revenue ratably over the term. Ratable means the money is earned month by month across the contract, which is how subscription software works, and why Palantir cares a great deal about how big the first contract is and how fast it expands.

Same filing, one more detail buyers miss. Palantir runs pilots and bootcamps “generally at our own expense.” So the first step usually costs you time, not cash.

Why no list price, then? Partly strategy. Foundry deals vary wildly. A single on-premises deployment for a defense agency and a cloud-hosted analytics workspace for a mid-size manufacturer share a product name and almost nothing else, which makes a rate card more of a liability than a sales tool. And honestly, a published number would anchor every negotiation. No vendor in Palantir’s position wants that.

How Is Foundry Usage Actually Measured?

Foundry meters work in compute-seconds, a unit of processing work rather than clock time, plus storage in gigabyte-months. Palantir publishes the formula. The dollar rate per compute-second is set in your contract, not published.

Palantir’s usage types documentation is blunt about this. A compute-second “is not a measurement of time, but rather a unit of work.” For a batch job, the simplified math looks like this.

compute-seconds = max(vCPUs, GiB of RAM ÷ 7.5) × number of executors × seconds the job runs

Run it once with real numbers. Say a nightly pipeline uses 4 executors, each with 4 vCPUs and 30 GiB of RAM, and the build takes 10 minutes. RAM divided by 7.5 is 4, which ties the vCPU count, so each executor counts as 4. Four executors times 4 times 600 seconds gives you 9,600 compute-seconds for one run. Nightly, that’s roughly 288,000 compute-seconds a month from a single pipeline that most people on the team never think about.

That’s the shape that matters. Not the per-run number.

Storage follows a separate meter. Dataset storage and Ontology data are measured in gigabyte-months, so a 500 GB dataset kept for a full month counts as 500 gigabyte-months. Ontology indexing has its own cost, because objects you expose to apps and agents get indexed separately from the raw datasets underneath them.

Interactive tools burn compute too. Palantir documents compute usage per application, including Contour, Code Repositories, SQL, and Code Workspaces, which count compute-seconds for as long as a workspace is starting or active. Read that twice if your data scientists like leaving notebooks open over a long weekend.

How Does Palantir AIP Pricing Work?

AIP converts LLM usage into the same compute-second meter. Each model has a published rate of compute-seconds per 10,000 input and output tokens, and output tokens cost roughly 4 to 5 times more than input tokens.

Palantir’s AIP compute usage page gives the conversion as tokens multiplied by the model’s compute-seconds per 10,000 tokens, divided by 10,000. The rates differ by model and by region. Two examples from the North America table, as published at the time of writing, are below.

Model lineups change fast, so treat these as a worked example of the method, not a price sheet. Check the live table in your own enrollment before you budget.

Look at the right two columns. A chatbot that reads a lot and writes a little is cheap. An agent that writes long summaries, drafts emails, or narrates its reasoning back to users is not, because output is where the multiplier sits. Two agents can handle the same request volume and land very different bills.

AIP Logic adds its own floor. Palantir’s Logic documentation describes minimum compute-second charges per LLM block and per tool execution, so a function that chains six tool calls costs more than one that makes a single call, even when the token counts look similar. If AIP itself is still a fuzzy concept, our plain-language guide on what Palantir AIP is covers the components before you get into metering.

What Public Palantir Price Points Actually Exist?

Four public anchors exist. A free AIP Developer Tier, a $2.2M per year federal module on AWS Marketplace, private-offer-only listings for the main platform, and a U.S. Army enterprise agreement with a $10B ceiling.

A few notes on each, because the details change how you read them.

The Developer Tier is real and genuinely free. A Palantir staff member confirmed on the Palantir Community forum that “AIP Developer Tier is completely free,” with capacity limits you can’t exceed. It’s a sandbox for learning, not a place to run production workloads. Great for a skeptical engineer who wants to poke at the Ontology before a sales call.

The $2.2M FedRAMP module on AWS Marketplace is the closest thing to a sticker price in public. It’s scoped tightly to 3 interfaces, 5 schemas, 200 users, and one financial operations use case, with no refunds and AWS infrastructure on top. Don’t divide it by 200 and call that a per-user rate. Palantir doesn’t sell that way, and the number mostly tells you where federal platform pricing starts.

And the Army deal? The U.S. Army announced an enterprise agreement that consolidates 75 contracts (15 prime and 60 related) into one vehicle, worth up to $10 billion over as long as 10 years. The Army itself describes that figure as the maximum potential value, “not any specific obligations.” Headlines that turned it into $1B a year got the math wrong.

Now the number to ignore. Some cost sites still quote Palantir at $141,000 per server core. That figure comes from a 2014 GSA price list for Gotham, Palantir’s government intelligence product, which Nextgov reported had been posted to Hacker News, and a GSA spokeswoman told Nextgov the data was “not even accurate.” Wrong product, wrong decade, disputed at the time. Please stop quoting it.

What Do Palantir’s SEC Filings Say About Deal Size?

They show a skewed customer base. Palantir reported 1,049 customers at June 30, 2026, while its top 20 customers averaged $124M each in trailing twelve-month revenue. Most buyers spend far less than that.

Both numbers sit in the Q2 2026 10-Q. Customer count grew from 849 a year earlier. Top-20 average revenue grew 67%, from $75 million. Read together, they say Palantir lands accounts and then expands them hard, and that a small group of very large customers pulls the average way up.

So what’s a typical Foundry contract? Nobody outside Palantir can answer that from public data, and anyone quoting a precise median is guessing. What the filings support is a shape rather than a figure. Pilots and bootcamps at Palantir’s expense, a first contract scoped to a use case, then expansion as more teams and workflows come onto the platform.

That shape has a practical consequence. Your first contract is the cheapest you’ll ever sign for this platform, and also the one that sets the unit economics for every expansion after it. Negotiate the compute rate, not just the headline fee.

Which Costs Sit Outside the Palantir License?

Integration engineering, Ontology design, enablement, cloud infrastructure on marketplace deals, and ongoing platform operations. In our projects, these decide whether year one feels expensive or reasonable. They rarely show up on the order form.

Here’s where the money goes once the contract is signed.

• Source integration. Every ERP, EHR, WMS, or homegrown database needs a connection, a sync schedule, and someone who knows which table is the real one. Our engagement with a global logistics operator connected 18 disconnected warehouse and inventory systems before the first AIP agent did anything useful.
• Ontology design. Someone has to decide what a Supplier, a Shipment, or a Patient Encounter actually is, which properties matter, and which actions people are allowed to take. Rushing it is the most expensive shortcut we see.
• Enablement, which is the cost buyers trim first and regret first. A platform your team can’t extend becomes a consultant dependency.
• Infrastructure. On marketplace deals like the AWS module above, cloud compute and storage are billed separately from the Palantir license.
• Run costs after go-live. Pipeline monitoring, permission reviews, model updates, and the occasional 2 a.m. failed build.

How big is that slice? It varies with how messy your sources are, so we won’t pretend there’s a ratio. For a sense of scale, that 18-system logistics build ran 5 months from kickoff to deployment and now saves the client $12M a year with 34% fewer procurement delays, according to our Foundry case studies. Most of those 5 months went into pipelines and the Ontology, not the agent.

Bias disclosed. Integration and enablement are exactly what we sell, so of course we’d tell you to budget for them. Ask your Palantir account team the same question and compare the answers. Our data integration and pipeline engineering work and Palantir training and enablement programs are the two lines we see underfunded first.

How Do You Keep Foundry Compute Spend Under Control?

Watch idle interactive sessions, oversized builds, full rebuilds that should be incremental, and LLM output tokens. Set resource queues and alerts early. Test agents in evals before scaling them to every user.

Compute bills don’t spike from one dramatic mistake. They creep. A notebook left running, a pipeline that rebuilds three years of history every night, an agent that writes 800 words when 80 would do. Each one looks harmless in isolation.

Here’s what we check in the first month after go-live.

• Code Workspaces and notebooks that stay active overnight or over weekends
• Is that nightly build incremental, or does it reprocess the full history every time?
• Executor sizes set for a test run and never revisited
• Ontology object types indexed but not used by any app or agent
• Agents with long, chatty outputs. Shorten the system prompt’s output instructions before you shorten anything else
• Resource queues and usage alerts per project, so one team’s experiment can’t eat another team’s budget

None of this is glamorous. All of it shows up on the invoice. Ongoing cost hygiene is part of our platform operations and MLOps work, and it’s usually the quickest win we find on accounts we inherit from someone else.

Run evals before you roll an agent out to 300 people, too. A test suite tells you whether a cheaper model handles the task well enough. That single comparison has paid for itself more than once.

Is Palantir Foundry Worth the Money?

Foundry earns its price when a frequent, expensive decision depends on data spread across several systems and someone will act on the output. It’s overkill for single-system reporting or a team that only needs dashboards.

Here’s the plain version, from people who only do Palantir work and therefore lose a deal every time we say no.

Foundry tends to be worth it in these situations.

• A decision made weekly or daily crosses three or more systems and costs real money when it’s wrong
• Operational teams need to act inside the tool, with changes written back to source systems, not just view a chart
• Governance and permissions matter, as in healthcare, defense, regulated finance, or critical infrastructure
• You plan to add use cases over time, so the Ontology gets reused

And it’s probably the wrong buy in these.

• Your data lives mostly in one warehouse and the goal is BI. A warehouse plus a dashboard tool will cost far less
• Nobody owns the decision you want to change. Agents can’t fix an org chart
• You need a fixed, predictable bill and can’t tolerate usage-based compute

If you’re weighing Foundry against a lakehouse or a warehouse, our Foundry vs Databricks vs Snowflake comparison breaks down how each platform charges and what each is built for. Plenty of enterprises end up running Foundry on top of one of the other two, which changes the cost math again.

Size the Whole Program Before You Negotiate

Three things to take away. Foundry pricing is negotiated, so the compute rate and term matter as much as the headline fee. AIP costs follow token volume, especially output tokens. And the license is only part of year one, because integration, Ontology design, and enablement decide how fast you see value.

PAL4C Labs is a Palantir-exclusive consultancy. We help operations, supply chain, and healthcare teams scope use cases, build pipelines and Ontologies, and get AIP agents into production.

Before your next call with Palantir, bring us the use case, the source systems, and the people who’ll use it. We’ll estimate the compute and services side of the budget with you so you negotiate from a real number instead of a blank page.

Pricing Questions Buyers Ask

How Much Does Palantir Foundry Cost per Year?

Is Palantir Foundry Free to Try?

So What Exactly Is a Foundry Compute-Second?

Does Palantir Charge for AIP per Token?

Is There a Minimum Contract Length?

Can You Buy Palantir Through AWS Marketplace or GSA?

Do Palantir Bootcamps Cost Money?