I don't think £80m is going to buy much AI, but even if it does then how will the government keep it "sovereign"? They've stupidly foregone any hold on the IP (what sort of early stage investor would countenance such a daft move?), so these companies are just waiting to either fail or be bought by cash-rich foreign tech companies.

DeepMind would have qualified as sovereign AI...until Google bought it. And that's the ghost of Christmas Future on this.

As the retrenched script writer sighed: " You can't make this shit up. "

My pitch might named " Haversine † " - a veiled "Have us on." Well, I might as well and join the throng.

Odyssey is probably the most honest. After 20 years faffing around with dodgy characters, Odysseus ended up exactly where he started from - namely a barren island in a forsaken backwater. (You might be forgiven thinking Britain, but Ithaca.)

† half versed sine.

So another 500 million flushed straight down the toilet. Nice to see that 'Gov procurement' remains an ozymoron par excellence.

Strip the sovereignty-speak off this procurement notice and it's the cleanest extraction vehicle a British civil servant has published since PFI.

Here's the playbook:

Step 1: Incorporate. Two founders, Ltd, Companies House, done in 48 hours. WeWork hot desk optional.

Step 2: Pitch the bingo card. The DSIT challenge list is "scientific discovery, health and social care, national security and defence, cybersecurity, transport, energy and net zero, and public service delivery." Ask an LLM to generate twenty plausible deck headlines in any of those lanes and you've got your shortlist entry by Monday.

Step 3: Staff the company. No UK-hiring requirement exists, nor could it - the WTO Government Procurement Agreement forbids explicit domestic-labour clauses, and Parliament has twice declined to invent one. Crucially, the Sovereign AI Unit itself subsidises overseas hiring: ten cost-free Skilled Worker visas per backed company, with next-working-day Home Office decisions. Route through the new entrant band (£33,400) or the STEM PhD band (also £33,400) and two developers - on-visa, cost-free to the employer - cost you about £77k a year all-in after employer NI. Add a WeWork, a laptop, an accountant and a brass plaque: call it £125k/year. Two-year project run cost: ~£250-400k. Use Claude/Cursor and the headcount stays at two.

Step 4: Cash the £5M. Milestone delivery on a "prototype capability" whose success criteria you largely wrote yourself in the bid. Retain - per the actual notice - all background and foreground IP. DSIT explicitly "will not seek to capture further economic value." You are now the 100%-equity owner of productised R&D that the taxpayer financed at retail.

Step 5: The sums. £5M revenue minus £400k costs = £4.6M gross profit. At 25% corp tax, £3.45M retained in company. Distributed as dividend across a two-founder cap table, each founder nets ~£1.06M personally after dividend tax - combined household take of ~£2.1M.

Step 6: Enjoy.

House: £600k buys you a four-bed detached in most of the home counties outside the M25.

Moderate yacht: £500k lands you a tidy 50ft motor cruiser - used Princess, Sunseeker Manhattan, Fairline Squadron.

Used supercar: £150k puts a 992-gen 911 Turbo or an Aston Vantage on the drive, with change for the insurance.

Total: £1.25M. You still have change for a Rolex and a tax advisor.

Bonus round: raise a Series A six months later at a VC-friendly markup, because you've got a government contract on the deck and a "de-risked" product. Your £5M procurement cheque just became a £50M valuation.

Now - the sovereign bit. What does the UK actually own of the AI stack?

Chips: None. No leading-edge fab on British soil. Newport Wafer Fab was partially divested to Vishay (American) on national security grounds after the Nexperia/Wingtech (Chinese-parent) saga. Nothing below mature-node compound semi gets made here.

AI silicon IP: One company ever built credible training-class accelerators in the UK - Graphcore, Bristol. It was sold to SoftBank (Japan) in July 2024 for about $600m. The then-Secretary of State for Science, Innovation and Technology, Peter Kyle, publicly blessed the deal as "a welcome end to the uncertainty." The same department is now launching a £500M fund to rebuild the capability it waved out the door two years ago.

ARM: Yes, British-designed. Owned by SoftBank since 2016. Listed on NASDAQ since 2023. Generates its growth licensing to US hyperscalers. Not a UK asset in any meaningful economic sense.

Sovereign supercomputers: Isambard-AI at Bristol - the flagship "AIRR" resource Sovereign AI companies get a million GPU-hours on - is built from 5,448 Nvidia GH200 Grace Hopper Superchips supplied by HPE, on HPE Cray EX architecture with HPE Slingshot interconnect. Dawn at Cambridge runs 1,024 Intel Max 1550 GPUs on Dell hardware. The UK contribution to the "sovereign" stack is the building, the electricity bill, and a plaque with Brunel's name on it.

Flagship AI lab: DeepMind. London. A division of Google since 2014.

Flagship AI scaleup: Wayve. London. Funded by Nvidia, SoftBank, and Microsoft.

So £500M is being directed at a "sovereign capability" in which the UK manufactures no silicon, owns no frontier-class chip IP (having sold both ARM and Graphcore to SoftBank), operates no fabs at relevant nodes, and runs its "sovereign" compute entirely on US silicon integrated by US vendors. The only ingredient actually domiciled in Britain is people writing software - and the visa desk is being used to import more of those from overseas at minimum threshold salaries on the taxpayer's tab.

This is Subsidised AI Consulting, with IP Giveaway, for Founders Clever Enough to Read the Notice. The capability it builds is the capability of a small cohort of London-based founders to afford Sunseekers. Call it what it is and price the tickets accordingly.

The Thick of It - I call app Britain:

https://www.youtube.com/watch?v=Ei9iM_zzzQk