73% of UK CFOs Back AI. So Why Are Only 1,376 New Finance Holding Companies Bothering to Incorporate?

The Number That Doesn't Add Up

1. That's the percentage of UK CFOs who told Deloitte this month that their view of AI's impact on business performance has materially improved over the past twelve months. Up from 59% at the end of 2025. Up from 39% in Q3 2024. A near-doubling in two years. The survey ran 1–13 July 2026, covering 58 finance directors including the CFOs of 10 FTSE 100 companies and 21 FTSE 250 companies. These are not interns filling out a Google Form. These are the people who sign the cheques.

Now hold that number against this one: 1,376.

That is the count of new SIC 64.20 companies, the activity code for holding companies, registered at Companies House in Q3 2026. Down 78.8% on the prior period. If AI is the greatest productivity revolution since electricity, as the CFO class now apparently believes, you'd expect the holding-company formation pipeline to be booming. Shell vehicles, SPVs, acquisition structures: these are the scaffolding of enterprise investment. Instead, the scaffolding is collapsing.

Something does not square.

Let's Do the Maths

Deloitte's Q2 2026 survey, the 76th edition of what the firm calls "the only survey of major corporate users of capital", found that 96% of respondents expect UK business investment in digital technology to rise over the next five years, with 93% expecting a rise inside 12 months. Half already expect productivity gains from AI deployment within the year.

HM Treasury published its Financial Services AI Adoption Plan on 14 July 2026, drafted by two government-appointed AI Champions: Harriet Rees, Group CIO of Starling Bank, and Dr Rohit Dhawan, Head of AI and Advanced Analytics at Lloyds Banking Group. Ten recommendations, spanning regulatory clarity, agentic payments and AI-powered financial advice. The government is not fiddling around the edges here.

The ONS, publishing its 2023–2026 AI business survey on 20 July, confirmed that self-reported AI use among UK businesses with 10-plus employees has risen from around 12% to around 35% since late 2023. The financial services sector, Parliament's Treasury Committee noted, "substantially outpaces" other sectors in adoption.

But here's what the ONS also found: close to 60% of businesses use AI only to improve existing operations. Fewer than one in five use it to develop new products, services, or markets. A mile wide, an inch deep. That is not a revolution. That is a Microsoft Office upgrade with better branding.

The Brand Gap Nobody Wants to Talk About

There's a second data point that deserves its own column but will have to settle for this one. Our analysis of UK Intellectual Property Office filings shows just 615 Class 36 trademark applications in Q3 2026, Nice Class 36 covering financial, insurance, and real estate services, down 74.2% on the prior period.

A firm that genuinely believes AI is restructuring its competitive position protects that position. You file trademarks. You build brand moats. You defend revenue lines. A 74.2% drop in financial-sector trademark activity signals that the army marching into AI is advancing without ammunition.

Worse: AIBD analysis of Companies House and IPO data finds that 98.6% of active SIC 64.20 companies hold no Class 36 trademark at all. Not one. Ninety-eight point six percent. These are financial holding entities with no registered brand identity in the very class that defines their activity. Buffett famously obsessed over economic moats. These companies are building castles with no drawbridge and no wall.

Paying restaurant prices for a microwave meal.

The Regulatory Picture: Finally Getting Serious

The Bank of England and FCA's own AI survey, published in late 2024, showed 75% of UK financial services firms already using AI, with the largest uptake among insurers and international banks. Cambridge's Centre for Alternative Finance put the global figure at 81% of financial services firms adopting AI at some level by 2026. Fintechs are ahead of incumbents; incumbents are ahead of regulators.

That last gap matters. The Treasury Committee warned last year that the FCA's, PRA's, and Bank of England's "wait and see" approach risks consumer harm and financial instability. The regulators pushed back in April 2026, insisting their principles-based, outcomes-focused framework is both intentional and proportionate. The Financial Services AI Adoption Plan calls for broader use of the FCA's AI Lab, including its Live Testing function and the so-called Supercharged Sandbox.

"Supercharged Sandbox" is, incidentally, four words that would have been pure satire in 2019.

HM Treasury also made its first designations under the Critical Third Parties Regime in July 2026, a direct response to concentration risk as financial institutions pile onto a small number of cloud and foundation-model providers. The plan explicitly flags that this concentration creates operational, security, and resilience risks extending well beyond any single organisation. Systemic, in other words.

Why the Mismatch?

Here is one reading. CFO optimism is real but it is cheap. Sentiment surveys measure intent, not deployment. Saying 73% of CFOs are AI bulls tells you what they believe, or what they want believed; it does not tell you what they are actually building, incorporating, or filing. The 14-point jump in optimism between Q4 2025 and Q2 2026 coincides with a 78.8% collapse in new holding-company formations in the same cohort classification. Optimism does not automatically generate capital structures.

The ONS framing supports this. AI adoption is nearly tripling in headline terms while depth of application has barely moved. Large language models lead actual deployment at just 18% of businesses. Robotics: 2%. The big numbers are driven by firms using AI to automate existing workflows, expense processing, contract review, invoice matching, not to build new revenue architectures.

That is perfectly rational behaviour. But it is not what a 73% optimism reading implies.

The Prediction

CFO optimism surveys are coincident indicators, not leading ones. The market learned this with cloud computing circa 2014 and again with blockchain circa 2018. In both cycles, boardroom sentiment raced six to nine quarters ahead of actual capital commitment.

Expect a trademark and incorporation surge in this sector in Q1–Q2 2027 as current pilot programmes reach board sign-off, the HM Treasury adoption plan crystallises into firm-level action, and the FCA publishes the guidance firms are explicitly demanding. The current trough in SIC 64.20 formations and Class 36 filings is not a structural retreat. It is a pause. The scaffolding is being ordered. It just has not arrived on site yet.

But if it doesn't arrive by mid-2027, then 73% was just a number in a survey. And those, as Taleb would note, are precisely the numbers you should distrust most.