The Landgrab Has No Territory: Public Relations as the New Infrastructure of AI Value
Why Narrative Position Is the Only Moat That Matters When Anyone Can Build Anything in an Afternoon
The geopolitical hostility now has a technical correlate: adversaries who know more about a model’s agentic failure modes than most of the enterprises currently deploying it.
The Map Precedes the Territory
In 1494, the Treaty of Tordesillas divided the undiscovered world between Spain and Portugal along a meridian neither empire could locate precisely. Both crowns signed a document carving up continents they had never seen, using coordinates that corresponded to ambition rather than geography. The map held for decades — not because it was accurate, but because it was believed, and because belief, backed by ships and money, has a way of making itself true.
Frontier AI labs are drawing meridians right now. When Anthropic acquired Coefficient Bio — nine people, six months old — for $400 million in April 2026, the transaction purchased a position: Anthropic is a serious actor in biotech. The flag planted on that beach will move enterprise biopharma sales conversations, recruit the next generation of life sciences AI talent, and signal vertical seriousness to every regulated industry Anthropic wants inside. Against a $380 billion post-money valuation, $400 million is the cost of a credible map line, not a technology investment.
Frontier lab acquirers are operating by a different calculus entirely, paying asymmetric prices to own a narrative position rather than race to develop one from the inside.
Code Is Free. Position Costs Everything.
A competent developer can scaffold a working application in an afternoon using commodity foundation models. The software has become nearly free to produce. What a prompt cannot generate is occupying a conceptual territory before a better-capitalized competitor decides it belongs to them. This is the structural shift that VC orthodoxy has not metabolized. The framework still demands validated MVPs, repeatable customer acquisition, and proof of manufacturing efficiency — a coherent set of criteria for a world where code has cost. Frontier lab acquirers are operating by a different calculus entirely, paying asymmetric prices to own a narrative position rather than race to develop one from the inside.
The build strategy and the narrative strategy are no longer sequentially related. The story is the product — particularly for the class of company whose exit is a strategic acquisition by an incumbent with capital anxiety. A founder who builds first and narrates second is optimizing for a buyer who may have already purchased the zip code.
When PR Becomes a State-Level Risk Vector
The distillation controversy — Anthropic documenting 24,000 fraudulent accounts and 16 million interaction campaigns by Chinese labs targeting Claude’s agentic reasoning capabilities — lands in public discourse as an IP dispute. The actual audience for Anthropic’s complaint is Washington, enterprise CISOs in regulated procurement cycles, and the EU AI Act enforcement apparatus deciding which vendors qualify as trusted infrastructure. The legal filing is a positioning document addressed to geopolitical stakeholders, and its effectiveness is measured in procurement decisions, not court outcomes.
The operational danger inside that PR posture is prompt injection at scale. An enterprise deploying a frontier model in an agentic loop — processing external documents, executing tool calls, operating with persistent memory — faces an attack surface that state-sponsored interaction campaigns have spent millions of API calls probing and mapping. The geopolitical hostility now has a technical correlate: adversaries who know more about a model’s agentic failure modes than most of the enterprises currently deploying it. PR-as-international-relations and IP ambiguity have produced a vulnerability that lives in the gap between a marketing claim about model safety and the actual threat model an enterprise security team should be running.
The Meridians Are Imaginary. The Ships Are Real.
Chinese frontier models are compressing the capability gap at a pace that makes the training-run economic moat of American labs subject to a depreciation schedule nobody has priced into current valuations. The distillation irony — that models trained on scraped human writing are complaining about their outputs being scraped — matters less than the strategic reality it points toward: capability advantage at the frontier is a time-bounded asset, not a permanent structural defensibility. The labs that survive are the ones that convert model capability into narrative infrastructure — distribution relationships, regulatory legitimacy, and vertical positioning — before the capability differential closes.
That is the treaty being signed in real time. Whoever draws the map first names the continent, controls the trade route, and collects the tariff — at least until the sailors find the actual coast and the meridians have to move again.