£4.56bn Quarter, 800 Trademarks: UK AI Investment Is Breaking Records. Its Brand Pipeline Is Not.

£14.4 billion. That is what UK businesses pulled in from equity investors across the first half of 2026. And AI ate most of it.

The Headline Number That Needs Unpacking

Beauhurst's data, cited in a Barclays analysis published yesterday, shows UK businesses attracted £14.4 billion in equity investment during H1 2026, up 26% on the prior half-year. AI was the engine. In Q2 alone, UK AI companies raised a record £4.56 billion: 57% of all equity capital deployed across every sector in the country. That is not a sector outperforming. That is a sector consuming.

The Q2 figure was itself 4.1% above Q1's then-record £4.38bn, and nearly three times the £1.59bn raised by AI companies in the same quarter a year earlier, a 186.8% year-on-year jump. Henry Whorwood, managing director of Beauhurst Insights, called it "a landmark quarter," noting the breadth of companies raising, not just the scale.

So. The money is real. But let's do the maths on what it's actually buying.

Eight Deals, 80% of the Pot

Here is the uncomfortable concentration stat buried in the Beauhurst data: eight rounds of over £100 million each accounted for nearly 80% of disclosed AI investment in Q2. The named beneficiaries include Isomorphic Labs, NScale, Fractile, Ineffable Intelligence, and PhysicsX. That is five names doing the heavy lifting for an entire quarter that the press is calling transformational.

To borrow from Nassim Taleb's framework: this is not a broad distribution. This is a power-law payout to a handful of outliers, dressed up as sector-wide momentum. The other 261 deals split the remaining 20%. Michelin-star prices for a meal where eight tables get the tasting menu and everyone else gets the bread basket.

And those 261 smaller deals still need something the eight mega-rounds have: brand identity. Which brings us to the number nobody is talking about.

800. Down 66.5%.

According to AIBD analysis of Intellectual Property Office data compiled by TrademarkDashboard, Class 36 trademark filings, the Nice Classification covering financial services, insurance, and investment brands, stood at just 800 in Q3 2026 to date. That is a 66.5% collapse versus the prior period.

Class 36 is where fintech brands live. Payment platforms. AI-driven insurance products. Robo-advisory services. Wealth management tools. If you are building a financial services AI company in the UK right now, Class 36 is the trademark class that protects your commercial name in the market you are operating in. And filings have fallen off a cliff.

This is not a rounding error. A 66.5% drop is a structural signal. Either founders are skipping brand protection entirely (plausible, given how many are moving fast) or the pipeline of genuinely new financial AI brands is far thinner than the formation data implies.

The Formation Surge Looks Different Up Close

Your Company Formations' inaugural Business Formation Barometer, published 28 July, showed AI and software development registrations up 86% in Q1 2026. Business and domestic software development grew from 4,035 to 7,506 registrations in a single quarter. Impressive on a chart. But there is a structural caveat the data providers themselves acknowledge: there is no dedicated SIC code for artificial intelligence companies in the UK's classification system. Businesses working in AI register primarily under software development, making it the closest available proxy, not an exact count.

So the 86% formation surge is real, but it is a broad proxy. It captures everything from genuine AI infrastructure plays to solo developers registering shell entities under the software SIC. The trademark data is a sharper filter: it tells you how many of those formations intend to build a defensible brand in financial services. Right now, the answer is: dramatically fewer than a year ago.

Charlie Munger used to say that all you have to do is find a few great companies and sit on your ass. What he didn't say is that the companies need to know what they're called first.

The Regulatory Backdrop: Good News for Insurance, Awkward Timing for AI

The PRA and FCA dropped Consultation Papers CP11/26 and CP26/29 on 14 July, proposing a tailored UK regime for captive insurers: a bespoke, proportionate framework separate from Solvency UK, designed to make Britain a competitive domicile for captive insurance structures. Currently there are no captive insurers established in the UK, with businesses typically setting up captives offshore in Bermuda, Guernsey, or Vermont.

The consultation closes 14 October 2026 with the regime set to launch in summer 2027. The proposals include a four-to-six-week application target, simplified governance, and reduced regulatory reporting. The FCA would also disapply retail-focused Consumer Duty requirements, recognising captives' distinct risk profile.

This matters for AI finance specifically. Captive insurance is an instrument that sophisticated corporate groups use to self-insure operational risks, including, increasingly, AI model failure, data liability, and algorithmic error. A UK-domiciled captive regime, if it delivers on the competitive promise, could make London a structurally more attractive base for AI-forward financial groups that need bespoke risk management. Sidley Austin notes the UK domicile could become attractive particularly where proximity to the London market matters.

The tension is this: the PRA itself acknowledges that tax treatment of UK captives versus Bermuda and Cayman falls outside its remit. Regulatory architecture without tax architecture is a half-built bridge. The City has been promised this bridge since the Chancellor's Mansion House speech in July 2025. Twelve months later, it has a consultation paper.

The Verdict

Three data points. £4.56bn in AI equity in one quarter. 7,506 AI and software incorporations in another. And 800 Class 36 trademark filings, down two-thirds, for the sector where most of that capital claims to be heading.

The 75% of UK financial services firms already using AI, per Bank of England and FCA survey data, are not all registering new brands. They are mostly bolting AI into existing products, using existing names. That is fine. But the formation surge is supposed to represent new entrants. New entrants without protected brand identities in their core operating class are not building companies. They are building experiments.

Record investment quarter. Collapsing brand pipeline. The money is moving faster than the infrastructure that turns a startup into a franchise.

My concrete prediction: by Q1 2027, a meaningful proportion of the 7,506 Q1 2026 software incorporations will have dissolved without ever filing a single trademark. The ones that survive will look back at 2026 as the year everyone raised money and nobody thought about what they were going to be called.