AI and Creative Technologies (Communications and Digital Committee Report)

Lord Tarassenko Excerpts

Friday 13th June 2025 Lords Chamber

[Lord Tarassenko (CB)]

My Lords, I draw the House’s attention to my registered interests as the founder director of Oxehealth, a University of Oxford spin-out which uses AI for healthcare applications and which will be looking for scale-up capital in the near future. I congratulate the noble Baroness, Lady Stowell, on her excellent report, which identifies the key issues with clinical precision and puts forward some possible solutions. I also congratulate the noble Lords, Lord Massey and Lord Evans, on their excellent yet contrasting maiden speeches.

I entirely agree with a sentence in the very first page of the report:

“It is homegrown AI companies, not big tech incumbents, that will drive the innovation needed to realise the UK’s AI potential”.


So I want to focus on the following question: do we now have the right conditions for the UK to develop its own sovereign AI capability, which I will define as having at least two homegrown AI companies, each worth £100 billion?

DeepMind has already been mentioned by the noble Lord, Lord Willetts. If DeepMind—generally reckoned to be worth $100 billion today—had not been sold to Google for somewhere between $400 million and $650 million in 2014, and if ARM, which is currently valued at around $150 billion, had not been sold to SoftBank in 2016, that sovereign AI capability would exist today. So the question can be reframed as follows: how can we, today, help successful UK AI companies scale up rather than be acquired by internationally owned companies such as Google and SoftBank?

All the evidence in the report points to the fact that there is no problem with the early stages of the pipeline. Nearly two-thirds of Europe’s unicorns are headquartered in the UK. As the report informs us, the UK has produced 20 AI unicorns to date, including four in 2023-24 alone. Just this week, however, as the noble Baroness, Lady Stowell, mentioned, three new deep-tech unicorns have been sold to US investors or companies, including Oxford University’s quantum computing spin-out, Oxford Ionics: three more to add to the list of those companies which will not scale up as UK companies.

As the report highlights, without significant scale-up capital from domestic sources, the UK risks being just an “incubator economy” for other nations. At the start of the scale-up journey—series B and C—the British Business Bank should be able to invest, and we heard from the Chancellor this week that its funding capacity is to rise by two-thirds to £2.5 billion per year. This is welcome news, but, as others have noted, the report highlighted that companies found it difficult to keep track of the bank’s 21 different programmes, and the restructuring of its offering cannot come soon enough.

At the other end of the scale-up journey—series D and beyond—raising rounds of £100 million or more cannot be done without institutional capital. UK pension funds manage more than £3 trillion yet invest barely 1% of this in growing domestic companies. US endowments typically allocate 5% to 15% to venture and growth equity and Canadian models between 10% and 20%. Given this evidence, I cannot see why the Government’s proposed reserve power to enable them to force pension funds to increase their investments in British companies and infrastructure should be seen as controversial.

Finance is not the only issue which impacts the ability of UK AI companies to grow; talent, regulation, data and compute also matter. I shall speak only very briefly about talent and focus on data and compute. I am all for expanding the global talent visa to attract highly skilled AI researchers to our shores, but I worry about the decreasing pool of home postgraduate talent. Further evidence emerged this week: UK students accounted for just 43% of the 25,000 enrolments for full-time postgraduate research degrees at British universities this academic year, compared to 51% in 2017-18.

The AI Opportunities Action Plan also called on the Government to

“identify at least 5 high-impact public datasets”

to be made available to AI researchers and innovators. However, the Government committed only to explore how to take forward this recommendation as part of DSIT’s work to develop the national data library. We have heard nothing about the national data library—apart from in the intervention of the noble Baroness, Lady Kidron, earlier—from the Government for the past few months, although the announcement of £600 million funding for the health data research service in April was very welcome. Delivering on the aims of that investment—a single, secure entry point to access aggregated, anonymised patient data—will not be a trivial matter.

The stakeholder map drawn up by Health Data Research UK shows 40—I stress 40—stakeholders on the map, and it says that it is a non-exclusive attempt to draw this map. Nevertheless, building on the existing guidelines for sharing patient data in secondary and primary care, a minimum viable product should be deliverable within a matter of months, not years. Does the Minister intend to discuss with her DHSC colleagues giving preferential access to this sovereign data asset to UK companies, with higher rates charged to any international competitors?

Turning to sovereign compute now, recent news is positive: investment into data centres and into exascale compute in Edinburgh, and the twentyfold expansion of the UK’s high-performance compute capacity by 2030. The report of the noble Baroness, Lady Stowell, recommended that UK AI scale-ups should be granted access to these facilities to catalyse commercial opportunities. Will the Minister confirm that it is indeed the Government’s intention to do so?

However, there is also a key lesson from DeepSeek: it has demonstrated the power of distillation. Training high-performance distilled large language models is now possible with just one or two GPUs. The ingenuity of DeepSeek has paradoxically led to several data-centre buildings in China lying empty. We do not have to copy the data-centre and hardware compute frenzy generated by US big tech companies. They are only part of the compute solution.

I chose my definition of sovereign AI capability arbitrarily, based on where we would be today if both DeepMind and Arm were still UK companies. Today, the UK AI ecosystem is thriving, even at the unicorn level. For example, we are world-leaders in the use of machine learning to navigate the complexities of British streets: not the straight and perpendicular roads of US cities navigated by Waymo and others. We are also world-leaders in low-cost hardware for machine learning, and in AI-generated video for enterprise use. If we implement the right measures on scale-up finance, enable privileged access to sovereign data and deliver the promised sovereign compute on time and on budget, I believe that at least two UK AI companies will reach £100 billion valuation within the next five years.

In our relationship with the big tech companies from the US, we will only ever be AI-takers. If we want to be AI-makers, the development of a sovereign AI capability should be the UK’s top priority.