Trade didn't retreat in 2025 — it hit record highs and rewired around politics and compute. "Deglobalisation" mislabeled the shift. The guarded-globalization thesis holds, and sharpens: the moat is regional IP, trusted collaboration interfaces, and — now — trust you can verify.
TL;DR: The data is in, and it refutes the headline. Despite the highest U.S. tariffs since the Second World War, global goods trade grew about 6.5% in 2025 to record volumes, with AI-related goods alone driving roughly a third of the increase. The world isn't deglobalising; it's reglobalising — fragmenting some links while forging new ones along lines of politics and compute. That vindicates the guarded-globalization thesis and clarifies its core lesson: advantage comes not from scale or self-sufficiency, but from differentiated regional IP, trusted collaboration interfaces, and trust that can be verified. Europe's 2026 moves — the Technological Sovereignty Package and a live AI Act — turn trust from a value into infrastructure.
The headline was wrong
Nine months ago, in Part 1, we argued that deglobalisation was pushing technology from scale-at-all-costs toward "guarded globalization." The direction was right. The word dominating the discourse was not. "Deglobalisation" implies retreat — less trade, shorter chains, a smaller map. That is not what happened.
- Global goods trade grew ~6.5% in 2025, outpacing the world economy — even as U.S. tariff rates reached their highest level since WWII.
- U.S. imports and Chinese exports both hit all-time highs.
- The World Economic Forum, UNCTAD and McKinsey have converged on a replacement term: reglobalisation — a strategic realignment in which some links fragment (decoupling, export controls, reshoring) while new forms of integration appear elsewhere.
The map didn't shrink. It was redrawn.
What actually rewired it: politics and compute
Two forces did the redrawing.
Politics. Tariffs moved from negotiating tactic to durable policy, rerouting trade through non-aligned "connector" economies rather than shutting it down. Fragmentation changed the path of trade far more than its volume.
Compute. AI-related goods — semiconductors, GPUs, servers, networking gear — accounted for about one-third of all 2025 trade growth, expanding ~37% globally (U.S. +66%, EU +22%, China +16%). The U.S. added roughly half of the world's new data-centre capacity in 2025, pulling hardware through supply chains that still run through Taiwan, South Korea and ASEAN.
The insight worth sitting with: the single biggest engine of "globalisation" in 2025 was the buildout of the most strategically sensitive, most export-controlled category on earth. Trade grew because of the AI race, not despite the tariff war. That is reglobalisation in one sentence — integration and rivalry riding the same wire.
The friend-shoring mirage
The comforting story is that supply chains are quietly moving to trusted partners. The data says: partly — and not the way the slides suggest.
"Connector" economies — Bloomberg Economics names Vietnam, Indonesia, Mexico, Poland and Morocco — have absorbed most of the rerouting. In Q1 2026 the U.S. ran its largest goods deficits with Taiwan (~$59B), Vietnam (~$54B) and Mexico (~$43B), with China (~$30B) now sitting behind them.
But imports from China into those same connectors rose even faster: Vietnam's China-import share climbed from roughly 28% to 33%, Mexico's from 18% to 20% over 2017–2022. A large share of "friend-shored" product is Chinese input wearing a new passport. Genuine reshoring and friend-shoring are real at the margin — India's share of U.S. outbound FDI rose from 7.6% to 11.6% between 2019 and 2023 — but "trusted" is too often assumed, not established.
This is the crux of Part 2: in a reglobalised economy, trust is not a jurisdiction — it is a measurement. A friend-shored input is not trustworthy because it cleared a friendly border; it is trustworthy when its provenance, labour, security and IP exposure are known. Which is precisely where Europe's model stops being a slogan and becomes an edge.
Europe: trust becomes infrastructure, then law
Between Part 1 and now, Europe stopped talking about "open strategic autonomy" and started building it.
- The Technological Sovereignty Package (June 2026) — the Cloud and AI Development Act, Chips Act 2.0 and an EU Open Source Strategy — introduces sovereign-cloud assurance tiers (Levels 1–4) and explicit "no kill switch" guarantees over the software supply chain.
- The AI Act reaches its main application milestone in August 2026, converting "human-centric, trustworthy AI" from a principle into an enforceable, dated requirement (with the highest-risk obligations phased further out).
Europe's bet is coherent: if trust is the scarce asset in a reglobalised economy, make it measurable, certifiable and native to the stack. Not the loudest player, nor the largest — the most verifiable.
What changes for leaders
- Stop reporting "reshoring %." Report verified provenance. A Friend-Shored Dependency Ratio only means something if the inputs behind your trusted suppliers are traced. Otherwise you are counting passports, not trust.
- Treat AI-hardware exposure as a first-class supply-chain risk. When a third of trade growth is the most export-controlled category on the planet, your compute supply chain is now a geopolitical position, not a procurement line.
- Build collaboration interfaces, not walls. The winners of 2025 traded through connectors and standards, not through autarky. Invest in data-safe APIs and provenance rails that let you collaborate without leaking crown-jewel IP.
A Stockholm note
Reglobalisation rewards neutral, high-trust nodes that can verify. The Nordics — high institutional trust, strong public-private coordination, EU-aligned governance — are unusually well placed to host the emerging "verification layer": the provenance, assurance and standards work that lets trusted cross-border trade actually be trusted. Stockholm is a natural pilot site for turning trust from a claim into a credential.
Key metrics (updated for reglobalisation)
- Verified-Provenance Ratio — share of "trusted" inputs whose origin, labour and security are actually traced, not merely sourced from a friendly jurisdiction.
- AI-Hardware Exposure — share of your critical compute supply chain subject to export controls or single-region chokepoints.
- Friend-Shored Dependency Ratio (from Part 1) — now measured net of upstream China content.
- Time-to-Assurance — lead time to meet safety, ethics and IP requirements for launch; increasingly set by the AI Act clock.
- IP Retention Rate — share of high-value inventions kept and commercialised in-region.
Call to action
Deglobalisation was never the story. Reglobalisation is: a smaller number of trusted, verifiable links carrying a larger, more strategic set of flows. Anchor your IP locally, standardise how you share, and — the 2026 upgrade — make your trust auditable.
Interested in piloting a human-centric, friend-shored, verifiably-trusted program out of Stockholm? Get in touch.
Sources
- McKinsey Global Institute — Geopolitics and the geometry of global trade: 2026 update (trade +6.5%, AI-goods ≈⅓ of growth, all-time highs, AI-goods growth by region, data-centre capacity)
- McKinsey — AI-related goods lead global trade growth
- World Economic Forum — Reglobalization: rewiring the world economy for a new growth era
- UNCTAD — Global Trade Update: top trends redefining global trade in 2026
- U.S. Bureau of Economic Analysis — U.S. International Trade in Goods and Services, April 2026 (Q1 2026 bilateral deficits)
- CEPR / VoxEU — 'Connector' countries in a geoeconomically fragmented world (China-import shares of Vietnam and Mexico)
- Atlantic Council — 'Connector economies' and the fractured state of foreign direct investment
- Bank of America Institute — Reshoring vs. friendshoring (US FDI share shifts, incl. India)
- European Commission — Strengthening Europe's tech sovereignty (June 3, 2026) (Tech Sovereignty Package, CADA, Chips Act 2.0, assurance tiers)
- European Commission — AI Act regulatory framework
Part 2 of the Guarded Globalization series. Part 1 established the thesis (October 2025); this instalment updates it against 2026 trade data. Part 3 — the sovereignty stack and trust-as-law — follows.