In the 20th century nations fought over oil. In the 21st century they’re fighting over compute. Countries that own the stack capture growth and leverage — see Nvidia (US), TSMC (Taiwan), Huawei (China). The widely-read AI 2027 scenario sketches why: when compute is scarce, it becomes the binding constraint on progress, so the countries that control it set the pace.
The UK’s world-class research gives it a genuine ace. Four of the top ten universities on the QS ranking are British: Imperial, Oxford, Cambridge, UCL. That matters because compute breakthroughs still begin in the lab. Our universities are idea factories, and when they work, they spin out companies that harden IP into products.
Yet we are still renters of other nations’ compute infrastructure. Today, Britain accounts for just under 3% of the world’s total compute, much of which isn’t optimised for AI. Very little of it is also based on UK IP.
Hosting strategic compute companies changes a country’s trajectory. TSMC underpins around 25% of Taiwan’s economy and acts as a ‘silicon shield’. Nvidia has turned US AI capex into a macro growth engine and export leverage. Huawei helped China project standards and influence across emerging markets. The lesson is simple: when you own the stack, you export value and import leverage.
And this has suddenly become very urgent. The supply of chips has become a nation-state battleground. In just the last few weeks, the White House has floated 100% tariffs on imported semiconductors (with carve-outs for companies manufacturing in the US) and signalled further hikes are coming. Export controls that started under one administration and tightened under the next remain a fact of life, and policy changes with the political weather. The UK relying on someone else’s chips increasingly means relying on someone else’s politics.
At London Tech Week earlier this year UK prime minister Keir Starmer said he wants the UK to become an “AI maker, not an AI taker”.
For that to actually happen we should build and buy a sovereign slice of compute on UK IP.
The technological innovation is there, and the companies to develop it are blossoming. The UK boasts amazing companies across much of the compute stack, be it the likes of Graphcore and Fractile in energy efficient and high-speed architectures, Oriole Networks and Salience Labs in optical interconnects/switching, Intrinsic in next-gen memory or ARM in CPUs. These can deliver order-of-magnitude gains on specific workloads.
At AlbionVC we’ve been investing across this sector in the UK for more than a decade, and in that time ‘future of compute’ companies in the UK went from a rounding error to a real sector. Equity investment rose from £3m in 2015 to £284m in 2024; across 2015-2024 the category raised £1.53bn. Nearly 40% of 2024 deals were bigger than £10m — the sort of rounds deeptech needs to translate science into silicon.
The government should build a UK-IP data centre and buy from it. It also has a direct interest in ensuring there are successful UK compute hardware companies. Not because we can (or should) onshore every step of fabrication tomorrow; fabs are incredibly expensive and geography-sensitive. But we can assemble a competitive stack where the critical IP is British: accelerators and novel architectures; photonics and interconnect; memory and storage; orchestration, cooling and power systems. Manufacture where it’s most efficient and retain sovereign control over the designs.
When geopolitics turns again, IP can travel. Capacity can be duplicated. Sovereignty is the ability to re-manufacture and re-deploy without asking permission.
The government should also act as a first anchor customer, for two main reasons.
First, revenue beats rhetoric. European chip companies still find it harder to raise growth capital than their US peers. The US has an alphabet soup (DoD, DARPA, CHIPS Act) supporting the development of its ecosystem. In early markets, the difference between a great demo and shipping a product is a reference customer with real workloads. A government-backed, UK-IP data centre saying, “we’ll run inference and training here, at scale,” would bring forward revenues a full funding cycle. That compounds as talent joins, boards deepen and supply chains firm up.
Second, it changes our bargaining power. Today, if a foreign supplier hikes prices, diverts allocation or falls under a new export regime, we’re powerless. With a domestic stack, we can call the shots. Even moving a small slice — say 10-20% — of national AI work onto UK-controlled kit gives us a real fallback in a crisis and bargaining power in peacetime.
The government should set a clear procurement bar on cost-per-token, latency, energy per query and security-by-design, fund multiple approaches and rotate traffic based on service level objectives (SLOs). The point is to create a market that rewards UK IP for being better, not just British.
What would I do tomorrow?
- Commission a 50MW pilot 'sovereign data centre’” with a rolling five-year purchase commitment tied to measurable SLOs (cost, energy, latency, security).
- Hard-wire it into the AI Research Resource (AIRR) so startups and SMEs can burst onto sovereign compute for credits, not just hyperscalers. Make procurement frictionless and innovation-led.
- Mandate open interfaces (for example using PCIe based cards to allow maximum flexibility) so we don’t trade one lock-in for another.
- Co-invest with universities on packaging, photonics and memory and pre-commit to buying the output that passes performance gates.
Will we still buy from Nvidia, AMD and others? Of course. We should. Best-of-breed wins workloads. But today we’re almost entirely renters. If tariffs, export controls or ‘friends-only’ allocations tighten further, we’re price-takers. Building and buying a UK-IP data centre won’t make us completely self-sufficient, but it will make us credible and in control of our destiny.
The prize isn’t just resilience. It’s a new growth engine: defensible IP, exportable hardware and tools, high-wage clusters and global political leverage. A UK-IP data centre wouldn’t be a museum piece. On targeted workloads it can be world-class on energy, latency and cost — and it would make Britain a maker, not just a taker, of the future.
