The arrangement has been running since 2020, and for most of that time the Department of National Defence treated its details as something the public did not need to see. Documents obtained by the Investigative Journalism Foundation and published by CBC News on Nov. 7 describe an agreement for Palantir’s Gotham software that began at $14 million and can reach $63 million if every option is exercised. An internal notation described the file as not for public disclosure.
Gotham is not a spreadsheet. It is a system for fusing information held in separate databases so that an analyst can see relationships across them, and the department says it uses the software to do precisely that, on closed classified networks inside secure facilities in Canada. Officials declined to say what data Gotham actually processes, citing national security, which is the answer a defence ministry gives and also the answer that makes the rest of the discussion difficult.
The assurances the department offered are specific. A security assessment concluded that no government data, prompts, outputs or metadata handled under the contracts could be stored, accessed or processed outside Canada. On the facts as Ottawa describes them, that is a strong position: the data sits on Canadian soil, on networks that do not touch the public internet, administered under Canadian security rules by people with Canadian clearances.
The counterargument is about who can compel whom, not about where the servers sit. The CLOUD Act allows American authorities to require U.S. companies to produce data in their possession, custody or control, and lawyers who have examined the arrangement note that the statute’s reach is defined by corporate control rather than geography. A Canadian subsidiary of a Delaware company remains inside that perimeter. Whether a server in Ottawa changes the answer is the question a court would face, and the honest response is that it has not been tested in this configuration.
Procurement specialists make a second point, which is about time. A contract that runs for six years with options is not a purchase; it is a relationship, and relationships accumulate switching costs. The pool of vendors able to deliver a system of this kind, cleared for classified work at allied standards, is short. Two Canadian firms, Vancouver-based Caseway and Ottawa-based Anvil, have been cited by experts as alternatives, and the same experts concede that neither approaches the breadth of what Palantir sells. That gap is the actual exposure, and it is not unique to Canada. Ukraine’s dependence on Starlink became an operational question in October, when officials accused SpaceX of throttling speeds near the front line. The lesson researchers draw is not that any single company is untrustworthy. It is that a military which buys a foreign platform for a core function has handed a foreign government a lever, whether or not that government ever pulls it.
The dependence has a second form that no statute addresses. Software of this kind spreads through an institution by being useful: analysts build workflows on it, data models are shaped to its schema, and the queries that produce a morning briefing are written by people who have stopped thinking about the vendor at all. Replacing Gotham in 2028 would mean rebuilding the analysis, not swapping a license, and departments that have tried to migrate away from systems like it describe the process in years rather than months. Dependence in procurement is rarely announced. It accumulates in the interval between the first pilot and the second renewal.
Palantir’s own positioning sharpens the concern. The company has described itself in increasingly national terms, its chief executive has appeared in advertising arguing for Western values, and a list of principles it published in June included the assertion that Silicon Valley has an obligation to participate in the defence of the nation. None of that is improper. All of it is relevant to a customer asking whether a vendor’s interests will align with its own in five years, when the contract renews and the alternatives are thinner than they were at signature.
The classification marking is the part that will outlast the contract. Such markings are lawful, and defence departments use them routinely, but they are also why the file surfaced through a journalism foundation instead of a procurement notice. In Ottawa, as in Washington, the political cost of a contract is set less by its value than by how it became known. A $63 million deal that nobody could find is harder to defend than a $63 million deal that was tendered in public, even when the security reasoning behind the secrecy was sound.
For Palantir the money is immaterial. Canada is a small line in a business that has grown on U.S. government work, and a $63 million ceiling does not move a quarterly report. What the file establishes is a template: allied militaries adopting American data platforms quietly, on classified networks, under assurances that borders still determine who can read what. The next country to sign will point to Canada’s security assessment as the model. The next legal challenge will be about whether that assessment survives a subpoena, and the answer will not be written by a defence ministry.
