No Adversary Required

by Patrick Jaritz / VICTOR

Gray-zone coercion needs an adversary calibrating against a response it wants to avoid. The tax provisions letting the AI industry pay a fraction of the statutory corporate rate need nothing of the kind: no probing, no concealment, no one to name. That absence makes the second problem harder to mobilize a fix for.

Russia's drone campaign over Europe and Iran's tanker coercion in the Strait of Hormuz share a method: stay below the threshold that forces a collective response, and learn exactly where that threshold sits by testing it, incident after incident. Each drone flight and each tanker seizure is a data point the aggressor reads and adjusts against. The method requires an adversary calibrating in real time, and a target that has never said in advance what crosses the line.

The way AI companies pay tax looks like the same story at first glance: a fast-moving technology, a slow state, a correction that keeps not arriving. ITEP's 2025 analysis found Amazon, Alphabet, Meta and Tesla paid a combined 4.9 percent effective federal rate on 315 billion dollars of US profit. NVIDIA paid 15.5 percent against a 21 percent statutory rate. The mechanisms are net operating loss carryforwards, accelerated depreciation and R&D credits, each older than the AI boom by decades, each written into law in full view of Congress. Nobody is testing anything here. The rate a company pays sits wherever the statute already places it, and has for years.

That difference matters more than the resemblance. Gray-zone coercion has a target: an adversary whose calculus can, in principle, be changed by a credible commitment to respond. Deterrence works because there is someone to deter. A company using an NOL carryforward exactly as Congress wrote it has no reaction to probe for and no incentive to stop until the law itself changes. There is nothing to name and nothing to threaten, because nothing is being tested.

The distinction sharpens against the case usually cited for both: the British East India Company. In 1765 the Mughal Emperor granted the Company the Diwani, the right to collect tax revenue across Bengal, Bihar and Orissa. The Company received that authority once and exploited it openly for decades, building a private army eventually twice the size of Britain's own. Correction came only when a 1770 Bengal famine, worsened by the Company's own tax policy, forced it into bankruptcy and a Bank of England bailout, and even then the Regulating Act of 1773 was a partial fix. Full correction waited until 1858, after an armed rebellion. Ninety years passed between an unrevisited grant and the crisis that finally forced Parliament to revisit it.

That is the shape the tax case fits: a grant, once made, exploited exactly as written, correctable only by an external shock large enough to force a legislature to look again. Harvard economist Jason Furman found data center investment accounted for 92 percent of US GDP growth in the first half of 2025. That number is precisely why nobody currently in office wants to be the one who looks.

Gray-zone coercion gets fixed by a state naming a threshold in advance and committing to defend it, which gives a government something to rally around. The tax base's problem has no adversary to rally against, only a legislature that would have to decide, on its own initiative, that a footnote in an economist's growth analysis matters more than this quarter's headline number. That is a harder ask, not because the statute is harder to amend than a military posture is to commit to, but because there is no enemy to make the amending feel urgent.