Full Disclaimer: I am a contractor working in AI at the Department for Business and Trade, all opinions expressed here are my own and are not reflective of any knowledge of trade policy or strategy other than that which is publicly available.

—————————————————————————————————————————————————————————————————-

The last time Britain rewired the world it started with a loom in Lancashire and ended with the telegraph humming across oceans. Today the warp and weft are silicon and energy, not cotton and steam, but the prize is the same: set the technical standard and you set the terms of trade for a generation.

A tale of two action plans

Washington’s playbook: America’s AI Action Plan is unapologetically muscular—deregulate at speed, pour concrete for data‑centres, and weaponise export controls to keep allies close and rivals thirsty for GPUs .
Whitehall’s counterpoint: The UK Compute Roadmap promises a twenty‑fold jump in public compute, new National Supercomputing Centres and “AI Growth Zones” to crowd‑in private capital, while the AI Opportunities Action Plan sets out 50 levers—from sovereign GPU pools to an AI Energy Council—to make Britain an “AI maker, not taker”.
Shared anxiety: Both plans acknowledge systemic risk. The US folds security into export controls; the UK publishes an International AI Safety Report and a series of Frontier‑AI risk papers to keep the debate evidence‑led.

Scale is not destiny

Starmer’s £1 billion compute boost is necessary but in my view, not sufficient. The UK cannot out‑spend the US or China, so advantage must come from orchestration:

Targeted openness
- America’s plan leans into open‑weight models “founded on American values” .
- The UK could pilot licensed open‑weight releases tied to our safety‑case framework, allowing SMEs to innovate without signing their data sovereignty to hyperscalers.
Compute as critical infrastructure
- Treasury might treat GPU clusters like we once did railways—long‑life assets with regulated access. Issue 30‑year “compute bonds” against predictable departmental demand, then syndicate spare capacity to universities and startups.
Trade policy as a force‑multiplier
- DBT might insert compute reciprocity clauses into free‑trade agreements: preferential market access in exchange for co‑located GPU fabs or power‑purchase agreements that de‑risk our grid upgrades.
Demand‑side pull
- Unlike the US plan, our published strategies underplay government procurement. Why not mandate that every major service transformation in the Roadmap for Digital and Data sources at least one model/service from the AIRR pool or an AI Growth Zone startup? That guarantees a home market before we chase export wins.
Safety markets, not safety museums
- Frontier‑AI risk work is world‑class, but it will ossify unless tied to cashflow. How do we start to act not talk? We could create a regulated marketplace for certified model evaluations where insurers, auditors and government departments buy and sell assurance artifacts. Let safety expertise compound like fintech did a decade ago.

Guardrails, not handcuffs

America’s narrative is “build, baby, build.” Ours risks becoming “assess, consult, delay.” The sweet spot is conditional acceleration: release compute and data only if the applicant publishes an auditable safety‑case and commits to independent red‑teaming. That turns regulation into a passport, not a barricade.

What we owe ourselves

Britain’s industrial story has always been about making global systems more trustworthy—think Lloyd’s for shipping or GCHQ for cryptography. AI needs an equivalent institution. We have the academics, the regulators and perhaps soon even the silicon. The missing move is to make trust a tradable export.

If we weave that fabric now—compute, safety, and trade interlaced—we won’t just keep up with the superpowers; we’ll sell them the standards they need to keep their own machines honest. That is a target worth warping the looms for.

——————————————————————————————————————————————————————————————————