Surveillance capitalism & The State
When Shoshana Zuboff named surveillance capitalism in 2019, the worry was that companies were harvesting your behaviour to sell ads. Targeted shoes. Manipulated clicks. A creepy but bounded commercial arrangement.
That framing is now obsolete. The data economy stopped being about advertising and became the operating system of state power. The line between Silicon Valley and Washington, the thing that let each pretend to check the other, no longer exists.
Here is the mechanism, layer by layer.
The state became a customer, not a regulator
The political takeover runs through one network. JD Vance worked at Mithril Capital, Peter Thiel's fund, and was made partner in 2015. Thiel introduced him to Trump, then put a record $15 million into his 2022 Senate race — the largest single donation to a Senate candidate in US history. Vance is now Vice President.
The network fans out from there. David Sacks, a PayPal co-founder, is the administration's AI and crypto czar. A Thiel protégé runs the Office of Science and Technology Policy. The people who spent twenty years arguing the state was an obstacle now operate it.
And the state has become their best client. Since the 2025 inauguration, Palantir has been awarded over $1.3 billion in federal contracts spanning Defense, Homeland Security, Treasury, Justice and a dozen more agencies. Its US government revenue grew 66% year-over-year in Q4 2025. It signed $81 million+ with ICE and a $30 million contract to build a prototype "ImmigrationOS." Anduril, backed by Thiel's Founders Fund, signed a ten-year Army deal in March 2026 worth up to $20 billion.
The doctrine is explicit: to hold American hegemony, the state must guarantee the dominance of its technology champions. The regulator became the customer, and the customer pays.
The new substance of surveillance is the agent
While the politics consolidated, the thing being captured changed.
Meta and Google no longer merely log which pages you open. The frontier is the autonomous agent — the model that reads your email, sits in your meetings, drafts your decisions, and executes them. To be useful it must ingest everything: your correspondence, your company's institutional memory, your financial reasoning. Ad-tracking saw your clicks. The agent sees your intent.
This runs through a handful of centralised APIs. Three or four firms own the compute and the training data, which means they own the context. Every enterprise that routes its work through a frontier model is feeding the most complete behavioural record ever assembled into infrastructure it does not control. Surveillance is no longer a byproduct of the service. It is the service.
Surveillance becomes lethality
The clearest view of where this leads is military, because there the abstraction collapses into a kill chain.
Palantir's Maven Smart System fused nine separate intelligence systems into one interface and compressed the time from detecting a target to striking it from hours to minutes. It can generate 1,000 targeting recommendations per hour. One unit reportedly matched the output of a 2,000-person targeting cell from the Iraq war with about 20 people. In March 2026 the Pentagon made Maven a formal program of record; its funding has grown to $13 billion from $480 million in 2024, and it is deployed across every US combatant command.
The live conflicts of 2024–2026 became the proving ground. Systems trained on terabytes of civilian data — social posts, geolocation, communications — now propose who is a target and which munition to assign. Mass surveillance and automated lethality turned out to be the same pipeline pointed in different directions.
The friction has to be removed
A machine like this cannot run with legal friction, so the friction is being dismantled — at home and abroad.
In the US: the executive order of 11 December 2025, "Ensuring a National Policy Framework for Artificial Intelligence," moves to pre-empt state AI laws. It stood up an AI Litigation Task Force inside the DOJ to sue states in federal court, and threatened to withhold federal broadband funds from states whose rules are deemed "onerous." Light-touch regulation is the polite name for developer immunity.
In Europe: open war on the EU's Digital Services Act, framed as a defence of free speech. In December 2025 Secretary of State Rubio sanctioned Thierry Breton, the act's former architect — froze his US assets, barred him from the country. The US Trade Representative threatened European firms, including SAP, Spotify and Mistral, with "fees or restrictions" unless Brussels backed off. The point was never speech. It was to break any foreign rule that could slow the expansion of American platforms.
The landing: proliferation
Put these together and the result is mechanical.
When the largest power runs flat-out and bundles surveillance, compute and lethality into products that ship over commercial cloud, no other state can opt out. India has committed over $200 billion. Gulf funds are building "AI factories." Global sovereign-AI spending will pass $100 billion in 2026. Everyone must build their own stack or be run by someone else's.
The standard comparison is nuclear, and it is exactly wrong. Nuclear weapons were controllable because they were physically scarce — rare materials, few facilities, a containment regime. Militarised AI is the opposite: distributed by design, sold by subscription, delivered through the same data centres that host your email.
You cannot contain what is being sold as a service. That was always the business model. We just didn't notice it was also the weapon.