A $55B Pentagon swarm push, a Dutch-Chinese chip lawsuit, and Huawei's 1.4nm flex — sovereignty is being priced this week.
The pattern this week is unmistakable: every great power is buying more AI capability than its institutions can govern. The Pentagon asked Congress for a near 24,000% jump in autonomous-warfare funding while senators noted that the underlying doctrine hasn't been updated since 2012. Beijing answered Washington's chip-export thaw by quietly telling customs to block the very Nvidia parts the White House just approved — and Huawei used the gap to announce a 1.4nm-equivalent process roadmap.
Watch & Listen First
- AI & Geopolitics, 17–24 May 2026 — Full Analysis · Weekly audio rundown built around the Palantir Pentagon fight, the Nexperia-Wingtech escalation, and Germany picking a French AI vendor over Palantir.
- CSIS AI Policy Podcast — A New Vision for Advancing AI Governance, w/ Andrew Freedman · Greg Allen sits down with Future of Privacy Forum's Freedman on the U.S. state-by-state regulatory patchwork now filling the federal vacuum.
- Foreign Policy Live — The Geopolitics of AI · FP's Ravi Agrawal takes audience questions on great-power AI competition.
Key Takeaways
- Doctrine lag is the new chokepoint. The Pentagon's autonomous-weapons rulebook (DoDD 3000.09) was written for a $225M program, not a $54.6B one.
- Export controls have flipped polarity. It's now Beijing — not Washington — restricting H200 sales, weaponizing demand to nurture domestic substitutes.
- The "sovereign AI" club has a price tag. Between the UK's £500M unit, France's €109B package, and India's Bharat-stack push, middle powers are committing real capital, not communiques.
- AI is leaking into industrial-policy disputes. Wingtech-Nexperia turned chip governance into a sanctions law fight; expect more EU-China test cases.
- Geneva will reset the diplomatic frame. The UN's first Global Dialogue on AI Governance opens May 26 — with the U.S. opposing it and China backing it.
The Big Story
Senate to Pentagon: $54.6B for autonomous weapons, no doctrine to govern them · May 20 · Military Times
→ Sen. Joni Ernst's subcommittee made under-secretary Emil Michael concede that policy "absolutely needs updating" — an unusual public admission given DAWG's funding is going from $225M to $54.6B in a single year. Strategically, this is the moment the U.S. crosses from experimental autonomy into industrial-scale procurement without a rules-of-engagement framework that allies (or Geneva) can scrutinize; the second-order effect is that adversaries gain political cover to skip their own restraint debates, and the Anthropic-Pentagon split over autonomous-use guardrails becomes the template, not an outlier.
Also This Week
China sues over Nexperia: Wingtech demands €6.8B from the Netherlands · May 22 · DutchNews
→ Beijing is using its Anti-Foreign Sanctions Law as an offensive tool for the first time in a chip dispute — meaning every European screening regime now carries direct retaliatory exposure for its own multinationals.
Huawei unveils 1.4nm-equivalent roadmap via "LogicFolding" · May 25 · BNN Bloomberg
→ If verified at scale, this collapses the timeline analysts had penciled in for Chinese node parity and undermines the strategic premise — node leadership equals leverage — that the entire export-control regime rests on.
DoD wants $30B for FY27 AI supercomputing — bulk tied to reconciliation · May 22 · DefenseScoop
→ The Pentagon is now competing with hyperscalers for the same TSMC and SK Hynix supply, which means national-security demand will price civilian capex out of the queue.
China formally blocks H200 imports despite U.S. green light · May 22 · Semafor
→ Beijing chose industrial policy over short-term compute access — the most expensive vote of confidence Huawei has ever received from its own government.
UN Global Dialogue on AI Governance launches in Geneva · May 26 · UN
→ With Washington opposing multilateral rules and Beijing championing them, the U.S. risks ceding the standard-setting microphone to a Chinese-friendly forum — a long-game loss disguised as a procedural skirmish.
From the Lab
Legal and Policy Approaches to Mitigate Catastrophic Harms from AI · RAND, May 7
→ Romanosky, Treyger, and Alhajjar argue that comprehensive federal action on catastrophic AI risk is unlikely this term, so meaningful guardrails will have to come from state legislatures and voluntary industry measures. For geopolitics, that means U.S. AI rules will be 50 different things to international counterparts — a negotiating disadvantage in any forum where the EU and China show up with one position.
Worth Reading
- How A.I. Was the Elephant in the Room at the Trump-Xi Summit — Why both presidents structured the summit to avoid the one issue that defines their rivalry.
- The AI Cold War: How US-China Tech Rivalry Is Reshaping Global Power — A useful synthesis of how chip controls, sovereign-AI funds, and military procurement now constitute a single system.
- Pentagon's $54.6B Drone Swarm Bet Lacks Doctrine, Senators Warn — The numbers behind the policy gap, with the FY27 line items broken out.
Power is being denominated in compute and codified in customs forms — and this week, the slowest-moving variable was the rules.