Palantir Loses Money on Purpose at the Start. That's the Whole Plan.

Picture a software engineer flying to a military base, badge clipped on, sitting next to the analysts who actually use the tool - not to sell anything, but to wire the product into the way the place already works. Palantir calls these people Forward Deployed Engineers, and its own registration filing says they travel to customer sites, including military bases, to embed the platform.² That image is not a perk. It is the entire strategy. By the time the engineer flies home, the agency isn't running Palantir software. It's running on Palantir.

The official story is that Palantir is a great government salesforce that keeps winning contracts. That undersells what's happening. Palantir does not win contracts the way a vendor wins a bid; it lands inside an institution and lets the institution slowly forget how it functioned before. The company didn't bolt this on after the fact - it wrote the playbook into the document it filed to go public.

Why losing money at the start is the point

Here is the mechanism most readers miss. Palantir's S-1 discloses something accountants find unremarkable and strategists should find revealing: costs are incurred upfront, while revenue is recognized over time.² In plain terms, the expensive part - the engineers, the integration, the months of embedding - hits the books before the money arrives. For a normal vendor that's a problem to minimize. For Palantir it's the toll you pay to get inside. The acquire phase is supposed to be unprofitable. You spend to prove the platform works on the customer's hardest problem, on their data, in their building. Once it does, the expand phase begins, and expansion is where the economics flip - because by then nobody wants to rip it out.

The evidence that the flip is real sits in the disclosures. The average revenue from Palantir's top 20 customers climbed from $54.6 million in 2023 to $64.6 million in 2024.⁶ That is not new logos doing the work; it is existing customers spending more, year over year, on the same embedded platform. Land once, expand forever. The acquire phase is a cost. The expand phase is an annuity.

When 75 contracts become one

The clearest proof that this model works at scale arrived in mid-2025. The U.S. Army consolidated 75 contracts - 15 prime and 60 related - into a single enterprise agreement with Palantir, with a ceiling of up to $10 billion over ten years.⁴ The headline reads like a $10 billion windfall. It mostly isn't. It is a procurement restructuring: existing, scattered work pulled into one agreement, with middleman fees stripped out.⁴ Read it correctly and it's more impressive than a windfall, not less. An institution that has 75 separate Palantir engagements has already been thoroughly landed. Consolidating them is the customer formally admitting it now runs on one supplier and deciding to stop pretending otherwise. That is the expand-and-scale phase reaching its logical end: the buyer reorganizes its own procurement around the vendor.

The growth nobody expected: the commercial side

If the model only worked on governments it would be a clever niche. The interesting development is that the same playbook is now firing in the private sector. In FY2024, government revenue was 55% of Palantir's $2.865 billion and commercial was 45% - already a narrower gap than the 'defense contractor' label implies.¹ By the third quarter of 2025, U.S. commercial revenue grew 121% year over year, while U.S. government revenue grew 52% to $486 million in the quarter, with total revenue up 63%.⁵ The land-and-expand machine, it turns out, doesn't care whether the building it embeds in is a military base or a corporate data center. The same Forward Deployed motion that captured agencies is now capturing enterprises.

Isn't this just lock-in dressed up as strategy?

The fair objection is that 'land-and-expand' is a flattering name for vendor lock-in, and that the model's success is also its liability. That objection has teeth. A UK procurement-law analysis documented exactly this pattern inside the Ministry of Defence: the land-and-expand tactic produced a second contract worth more than three times the first, and a former central-government procurement lawyer concluded the MoD had 'commercially surrendered to a single service provider.'⁷ That is the dark mirror of the top-20 expansion number - the same compounding that delights investors is what 'surrender' looks like from the buyer's chair. The honest read is that Palantir has built genuine sovereign-dependency risk, and oversight bodies are only beginning to price it. But here is the part that makes it durable rather than merely controversial: the dependency is real because the utility is real. An agency doesn't get captured by software that doesn't work. The lock-in is downstream of having become genuinely load-bearing - which is precisely why it's so hard to undo, and precisely why regulators worry.

Palantir is often described by where it came from - a 2003 incorporation, early venture money including roughly $1.25 million from the CIA's investment arm in exchange for equity, no board seat attached.³ But the origin story explains less than the operating model does. The company that listed directly on the NYSE in 2020⁸ put its whole thesis in writing: acquire, expand, scale. The genius was never a single contract or a clever pitch. It was deciding that the goal of the first deal is not to make money - it's to make leaving unthinkable. Win that, and the second contract triples the first, and the customer eventually files the paperwork to make the dependence official.