Hard to Compute

Why Startups Need More Power

The Compute Roadmap published earlier this year is hugely ambitious. From AI Growth Zones to a 20x expansion of public compute capacity, it has the power to be transformative - nurturing and bolstering a flourishing AI ecosystem.

Government will not be able to execute on this roadmap alone. It will take close partnership with industry, and not just the usual suspects.

We are used to talking about startups as consumers of compute, and of course they are, more so now than ever before. However, startups can also support the Government to bring this roadmap to life. British founders are solving the very bottlenecks that threaten to derail our compute ambitions. Our national mission should be to support them and create a pro-innovation environment for them to thrive within.

That is why today we are publishing our new report, Hard to Compute: Why Startups Need More Power.

The report investigates and explores both sides of the coin. How we can ensure startups can get access to the compute they need to scale and how they can also support the build out of the very infrastructure ecosystem itself.

To write this report we looked at key aspects of the Roadmap and applied a startup lens.

First in our line of sight - procurement. The Compute Roadmap commits to “improving the eligibility of novel compute technologies for procurements” so that British innovations “are not discounted from procurement exercises because they do not meet criteria set to evaluate performance against incumbent solutions.”

This is the right instinct. But, as founders know, vague commitments to “improve eligibility” won’t cut it when chip supply chains require 1-2 year lead times and product generations turn over every 18 months. By the time a traditional procurement cycle concludes, the technology it’s buying is already obsolete.

The solution: Advanced Market Commitments (AMCs). These are binding commitments to purchase products that don’t yet exist commercially. They create guaranteed demand for innovations that would otherwise take years to reach the market.

Think of it this way: a startup developing breakthrough chip architecture doesn’t need a grant, they need a customer. AMCs provide that customer, which startups can then leverage to unlock significant private investment. The guaranteed contract becomes the runway to build.

Second comes the AIRR problem: access without accessibility. The AI Research Resource (AIRR) is the centre of public compute in the UK. The Government plans to increase its capacity by 20x by 2030. Brilliant. Transformative. Essential.

But AIRR, as many founders have told us, as currently structured, is not designed for startups. That is why we recommend creating a dedicated startup access programme with mission-focused AIRR programme directors who have the autonomy to rapidly allocate compute to strategic, high-growth projects.

To do this we’ll need to tackle issues of subsidy control head-on and make the selection process transparent and merit-based. AIRR should operate like a competitive grant programme with clear criteria, regular application cycles, and published success metrics.

Researchers have the scaffolding of their institutions around them. Startups, the engines of economic growth, don’t. We must bridge that gap.

Finally, we cannot and should not skirt the energy problem. Data centres are power-hungry beasts. The UK already has approximately 500 data centres, making it the world’s second-largest market. By 2030, data centres could consume 6% of UK electricity. By 2050, potentially 10% under heavy AI adoption.

The Government gets this. AI Growth Zones will receive priority grid connections, discounted energy (up to £24/MWh in Scotland), and access to clean power. That’s positive. But the brutal reality remains, British electricity prices are roughly four times those in the United States. Running a data centre in the UK therefore costs four times as much as operating an identical facility in the US purely because of energy costs.

We are clear on the action needed: we need to ensure data centres are a catalyst for Clean Power 2030. View them as sources of reliable, sustained demand for green electrons that can justify and accelerate renewable deployment. To do that:

• NESO’s Centralised Strategic Network Plan needs to adopt a “demand” principle. Right now, the plan doesn’t explicitly focus on demand and therefore economic growth. That undermines grid connectivity outside AI Growth Zones.

• The government should consult on reforming energy bill levies. The current piecemeal approach, exempting some energy-intensive businesses while spreading costs to households and small businesses, doesn’t solve underlying issues. Energy bill levies disincentivise local constraint markets, which could deliver cheaper energy during times of excess supply.

• And we must unlock data centre flexibility. Current regulations explicitly bar backup generators from providing balancing services or participating in demand-response schemes, even though the Government’s own Flexibility Roadmap identifies data centres as key participants in flexibility markets. Reform these regulations to turn data centres into grid-stabilising assets.

That is a very sweeping look at the report - have a read of it in all its glory and get in touch with your feedback, comments and suggestions. We’ll be doing a lot more on this issue - so watch this space. And if you want to get involved, we’d love to hear from you.