Chokepoints of the Digital Supply Chain

The first American-born pope took the name Leo in part as a deliberate echo of Leo XIII, whose 1891 encyclical Rerum Novarum framed the labour question at the dawn of industrialism. Leo XIV has named artificial intelligence one of the defining moral challenges of the century, explicitly drawing the analogy between the displacement faced by 19th-century industrial workers and the displacement now confronting knowledge workers.

The Pope's first Encyclical Letter, Magnifica Humanitas, names artificial intelligence as a defining moral challenge of the century and frames three areas as requiring protection: truth against manipulation, work against dehumanization, and freedom against digital dependency. The Encyclical's organizing line, a "Civilization of Love" built on justice, brotherhood, and dialogue, is positioned against what the document calls a "culture of power."

At the Synod Hall on 25 May 2026, the Vatican Dicastery for Promoting Integral Human Development hosted Chris Olah, Anthropic co-founder, for a presentation on AI alignment and interpretability research. The conversation is part of a longer Vatican dialogue with frontier AI labs that includes the Pontifical Academy for Life's working group on AI and human integrity, and the Rome Call for AI Ethics, originally signed in 2020 and now counting Microsoft, IBM, Cisco, the UN's FAO, and the Italian government as signatories.

The institutional Church's posture is not anti-technology. It is anti-displacement-without-recourse. AI is a tool that must be made to serve persons rather than displace them; the Vatican is willing to use its convening power to put that demand into the room with the labs.

On 23 January 2025, three days into his second term, Trump signed Executive Order 14179, "Removing Barriers to American Leadership in Artificial Intelligence", revoking the prior administration's October 2023 AI executive order in its entirety.

The accompanying AI Action Plan, released in July 2025, frames the contest as a race the United States must win. Federal procurement preferences for U.S.-developed models, expanded chip-export controls against the People's Republic of China, a permitting fast-track for AI data-centre construction, and a National AI Research Resource expansion are its four operative levers.

Trump's Project Freedom escort operation in the Strait of Hormuz (Issue 3, Chapter Four) and the Project Stargate compute build-out, anchored by an OpenAI-Oracle-SoftBank consortium, share an animating thesis: chokepoints are won by holding them, not by negotiating around them. The administration's view of the AI chokepoint runs through Taiwan, the Netherlands, and Phoenix, Arizona, and is being secured by export-control, permitting, and capital flow in parallel.

Zuckerberg has bet Meta's next decade on the proposition that open-weight frontier models can dominate the closed labs. The bet is structural: Meta's Superintelligence Labs consolidate the company's frontier-model research under a single executive, anchored by roughly $14 billion of new investment in Scale AI and an aggressive recruiting line against OpenAI and Anthropic.

The Llama model family is the flag. Public weights, permissive licensing, and a deliberate strategy of commoditising the model layer have moved Llama from research curiosity in 2023 to production deployment across cloud providers and downstream startups by 2026. The release of Llama 4 and the subsequent open-source ecosystem are explicit attempts to redraw the competitive map.

The strategic logic: if the commoditised tier of the model market belongs to whoever ships open weights first, the chokepoint moves up the stack, to compute, distribution, and consumer surface. Three things Meta already owns. Zuckerberg is not betting Meta will build the best model. He is betting the best model will not be a sufficient moat for anyone else.

Thiel's recent lectures have framed AI in deliberately theological terms, invoking the figure of the Antichrist as his chosen warning about technocratic governance done in the name of safety. The Cambridge Union and Manhattan Institute speeches make the same argument: the danger is not that AI is built, but that a transnational bureaucracy claiming to govern AI for the common good ends up as a one-world regulator.

The position is paradoxical: Palantir Technologies, which Thiel co-founded in 2003, is now one of the most important defence-AI vendors in the West. The company holds U.S. Army TITAN contracts, anchors the UK NHS Federated Data Platform, supplies Ukraine's wartime targeting infrastructure under disclosed contracts, and has become a top-twenty S&P 500 company by market capitalisation since 2023.

The throughline is consistent across Thiel's writing; regulation by transnational bureaucracy is the threat, not AI itself. EU rule-making and global governance frameworks draw the bulk of his criticism; the U.S. national-security state, by contrast, is where his portfolio companies have grown. The position has become the political doctrine of a substantial faction of the American right.

In September 2017, addressing a Russian classroom in Yaroslavl, Putin said "whoever leads in AI will rule the world." Eight years on, the line still defines Russian state framing of the technology; Beijing and Washington both treat it as a real strategic claim, not as rhetoric.

Russia's National Strategy for the Development of Artificial Intelligence, codified by Decree 490 in 2019 and updated in 2024, sets a domestic-model and sovereign-compute mandate. In practice, the visible Russian AI output is military: Lancet loitering munitions with autonomous terminal guidance, Iranian-derivative Shahed drone variants produced at Yelabuga, electronic-warfare countermeasures aimed at Ukrainian battlefield AI, and, increasingly, AI-assisted disinformation operations across multiple language theatres.

Sanctions and brain drain have hardened Moscow's posture rather than softened it. AI is no longer pitched as a globalised commodity but as a sovereignty asset; the chokepoint analysis runs through Russian-language LLM development, sovereign cloud infrastructure, and the export of military AI to states excluded from Western supply chains.

Ukraine has built the most operationally tested wartime AI stack on record. Zelensky's government, through the Ministry of Digital Transformation, runs Brave1, a defence-tech accelerator, procurement platform, and grant pipeline that has channelled small-batch capital into more than a thousand domestic and foreign-partnered defence-tech firms since 2023.

The AI Coalition of supporting states, anchored by the United Kingdom, Estonia, Latvia, Denmark, the Netherlands, and Canada, fields autonomous-targeting research, computer-vision drone swarms, and signal-intercept models trained on intercepted Russian radio traffic. Ukrainian-built strike systems, the Bober, the Lyutyi, the new Palianytsia jet drone, already incorporate machine-vision terminal guidance to defeat Russian jamming.

Kyiv's posture inverts the deterrence script that Washington and Brussels have spent five years debating. AI is not a future capability to govern but a present capability already shaping the front line. Zelensky's framing in addresses to Western parliaments and at the Munich Security Conference has been consistent: the question is not whether autonomous weapons will be used, but who writes the doctrine after Ukraine has already used them.

Iran's Permanent Representative to the United Nations, Ambassador Amir-Saeid Iravani, has been the country's most consistent AI voice in multilateral fora. Tehran's domestic AI policy has been delivered largely through Supreme National Security Council statements; the diplomatic articulation belongs to Iravani.

At UN Security Council briefings on AI and international peace and security, Iran has framed the technology as another vector of unilateral coercion, citing U.S. export controls on advanced chips, restrictions on academic collaboration with Iranian universities, and what Tehran calls discriminatory sanctions on AI research. Iravani has called for a treaty-based regime that constrains autonomous weapons and prohibits chip-export controls as a sanctions instrument.

The 2026 Hormuz crisis (Issue 3, Chapter Four) sharpened that argument. Following the 28 February strikes on Iranian nuclear and military infrastructure, Iranian statements at the UN explicitly tied the chip-export-control regime to what Tehran characterised as a coordinated technological siege. The position is internally consistent: from Tehran, the chokepoint is not Hormuz; it is the chip.

Xi's 2017 New Generation Artificial Intelligence Development Plan remains the architecture: AI as a core technology for socialist modernisation, sovereign compute as priority infrastructure, and a stated 2030 target of global leadership across theoretical research, industrial application, and AI talent development.

The Cyberspace Administration of China's Interim Measures for Generative AI Services, in force since 15 August 2023, made China the first major economy to issue binding rules on consumer LLMs: model registration with the regulator, mandatory content alignment with "core socialist values," and security review before public deployment. Domestic frontier labs, Zhipu AI, Moonshot, MiniMax, Baichuan, DeepSeek, have built within that perimeter, and DeepSeek's December 2024 R1 release demonstrated that sub-frontier U.S. chip restrictions did not prevent training-cost-competitive frontier-class models from emerging.

At the diplomatic layer, Beijing's Global AI Governance Initiative, announced in 2023 and expanded through the China–Africa AI cooperation framework, positions sovereign equality and developing-country access as alternatives to Western chip-export controls. The chokepoint, on Xi's reading, is American. The strategic response is sovereign compute, domestic models, and a parallel governance forum that treats Beijing as a peer rule-setter.

Pyongyang's AI footprint is the smallest of the nine treated here, and the most procurement-driven. North Korean state media has profiled AI-aided military training simulators and surveillance systems for the domestic population; published academic output is concentrated at Kim Il Sung University and Kim Chaek University of Technology, with applied work in image recognition, machine translation, and reinforcement-learning game agents.

The strategic interest is not domestic research. Open-source threat reporting from Mandiant, Microsoft Threat Intelligence, and the U.S. Treasury Department documents DPRK threat actors using commercial frontier models via proxy infrastructure: AI-assisted social engineering, generative-CV fraud for IT contracting through cut-out identities, deepfake voice for executive impersonation in cryptocurrency theft, and laundering tradecraft that has become measurably more efficient since 2023.

The chokepoint for Pyongyang is not building AI but accessing it; the regime has demonstrated it can route around export controls and platform terms-of-service one stolen identity at a time. The Treasury OFAC advisories of 2024 and 2025, the CISA/FBI joint advisories on DPRK IT-worker schemes, and the Mandiant disclosures on UNC4899 and adjacent threat groups document the playbook in detail.