# "AI-Enabled" Is Accurate. But It's the Wrong Pitch.

*Close to three-quarters of the UK's biggest venture haul in four years carried the same AI label, and the label turns out to describe two entirely different kinds of company.*

**Faraz Rizvi × [Foundry](https://www.spinupforge.com/foundry/)**

*Faraz Rizvi is a UK operator-practitioner writing about the work between a research breakthrough and a fundable company. He runs SpinUp Forge, a single-operator practice that does a UK university spinout's company-building while the founders stay on the science. [Foundry](https://www.spinupforge.com/foundry/) is SpinUp Forge's custom agentic harness.*

---

![Isometric illustration on a dark ink ground, with a wide band of empty sky around the scene. Two buildings of exactly the same square footprint and the same roof height stand well apart, side by side on one level. Each carries an identical pale signboard proud of its front wall, and both signboards read "AI-ENABLED". What stands under the two signs is not the same thing. On the left is a pale stone workshop with a pitched roof, a door and two windows; the only machine on its plot is a small dark equipment shed standing clear on the ground beside it, banded with a single lit ember shelf and joined to the building by a slung ember cable — a thing that could be unplugged and carted away, leaving the workshop standing. The line beneath it reads "AI is the Method". On the right there is no stonework at all: the building is itself one tall machine, the same dark chassis and the same lit ember banding as the shed, repeated floor to roof, with a single door cut into its base. Take the ember out of it and nothing is left standing. The line beneath it reads "AI is the Product". A plain grey walkway runs out of that right-hand door toward the viewer, and a founder in an ember dress and a bone beanie stands on it in the clear air between the two buildings, part-way along the walk to the door it leads to. She is small against the two buildings and she is the only person in the picture; the workshop is behind her to the left. The headline reads "“AI-Enabled” Is Accurate. But It’s the Wrong Pitch." The line under it reads "A founder who has not classified their mechanism first is knocking on the wrong one’s door." The SpinUp Forge mark and wordmark sit top left and the footer reads spinupforge.com. No quantity is encoded: there is no axis, no scale and no number anywhere in the image.](figures/mechanism-readiness/mechanism-readiness-hero.png)

A biotech founder spent six months building a pitch for investors who had just made enterprise AI software the biggest bet in AI. The pitch was good. Everything the deck said about the company's AI was true. It was still the wrong room.

That distance, between being true and being useful, is what this piece is about. Nobody sorts their own pitch into the right room on purpose. The sorting happens earlier than that, and almost nobody notices the moment it happens, usually months before the first deck goes out.

## The money did not spread

In early July, HSBC Innovation Banking and its data partner Dealroom published their count for the first half of 2026, and the headline was a genuinely strong one: the best opening to a year for UK startups since 2022. Read past the headline, though, and the strength did not spread evenly across it. It concentrated. AI startups took $12.6 billion of it, "almost three quarters of all venture capital invested in the UK," in the two firms' own words. The same pattern held into the second quarter, when Beauhurst's count of Companies House filings found a short list of very large AI raises, not a broad field of smaller ones, doing most of the work. Two data providers, two cuts of the year, the same shape: growth this year did not lift every AI company's odds evenly. A founder reading "record year for AI" and adjusting their own expectations upward is, more often than not, reading a story about somebody else's cheque. The story is real. It is just not about most of the founders reading it.

![A dot plot of UK venture capital in the first half of 2026: six rows on one shared horizontal scale that starts at zero, each dot placed at the money that part of the market took, with no gridlines and no axis ticks — every dot carries its own value. Largest first, the rows are "AI: enterprise software" at $5.2bn, drawn in ember and furthest right, with a faint ember rule dropped down the chart at its position so that nothing else reaches it; "Everything not AI (computed)" at $4.4bn, which is every part of the UK market outside AI taken together; "AI: health" at $2.6bn; "AI: hosting" at $2.1bn; "AI: robotics" at $1.5bn; and "AI: other sectors (computed)" at $1.2bn. The six rows partition the whole $17bn between them, so enterprise-software AI alone took more than everything outside AI combined. The two rows marked computed are this figure's own subtraction from totals reported by HSBC Innovation Banking and Dealroom for the half — $17bn of UK venture capital in all, of which AI startups took $12.6bn — and are not figures that report prints; its $17bn is a rounded total, so the non-AI remainder carries that rounding.](figures/mechanism-readiness/uk-h1-2026-split.svg)

*Where the UK's $17bn went in the first half of 2026, per HSBC Innovation Banking and Dealroom. Enterprise-software AI alone took more than everything outside AI combined.*

| Where the money went | H1 2026 (US$bn) | Basis |
|---|---:|---|
| AI: enterprise software | 5.2 | Sourced |
| Everything not AI | 4.4 | Computed: $17bn total less $12.6bn AI, per HSBC/Dealroom |
| AI: health | 2.6 | Sourced |
| AI: hosting | 2.1 | Sourced |
| AI: robotics | 1.5 | Sourced |
| AI: other sectors | 1.2 | Computed: $12.6bn AI less the four named sectors ($11.4bn), per HSBC/Dealroom |
| **Total UK venture capital, H1 2026** | **17.0** | Sourced |

Source: HSBC Innovation Banking / Dealroom, ["UK Innovation update H1 2026"](https://www.hsbcinnovationbanking.com/gb/en/resources/uk-innovation-update-h1-2026), published approximately 4 July 2026. The four named AI sectors and both totals are the report's own figures; the two rows marked *Computed* are this figure's subtraction from them and appear in no HSBC publication. Per HSBC Innovation Banking and Dealroom the $17bn is a rounded total, so the non-AI remainder is $4.4bn ±0.5bn; the comparison in the caption holds across that whole band. Figure: [`uk-h1-2026-split.svg`](figures/mechanism-readiness/uk-h1-2026-split.svg) · provenance: [`uk-h1-2026-split.provenance.md`](figures/mechanism-readiness/uk-h1-2026-split.provenance.md)

None of that money was spreading out to look for companies. It already knew what kind it wanted. What, exactly, is an investor buying when they buy "AI"?

## What the cheque is actually buying

I have spent years on the institutional side of exactly this decision, advising academics through commercialisation at the University of Surrey, and I have watched companies land on both sides of a line most of them never see drawn. Two fictional spinouts out of the same engineering faculty make it visible. One has trained a model that predicts equipment failure for industrial buyers: the model, sold as a subscription, is the whole of what the customer pays for. The other runs machine learning to search faster for a coating that resists corrosion: strip the AI out and the search slows down, but there is still, eventually, a coating to sell. Ask either whether it is "AI-enabled" and the honest answer is yes, for both. Where they differ is what happens once an investor asks the next question: what, precisely, are we buying.

Ask what an investor's cheque is actually a bet on, in each case, and the two spinouts stop looking like members of the same category. With the equipment-failure spinout, the bet is on the model itself: does it keep getting sharper, harder to copy. With the coatings spinout, the bet is on an engineering outcome: does the company keep delivering formulations that customers can act on. A cheque can only be answering one of those questions at a time.

There is a plainer way to ask it, and it needs no more machinery than the question itself: would the customer still buy the outcome if you swapped the AI out for a different method entirely? Take the AI away from the equipment-failure spinout and there is nothing left to sell. The model is the product, full stop, and a customer buying it is buying that capability directly. Take the AI away from the coatings spinout and there is still a company, slower perhaps, but still selling the same thing its customers actually wanted: a coating that resists corrosion, not a method for reaching it.

Two classes, then, and the distinction did not need inventing. The investment data has been sorting on it all half-year without ever saying so. Call the first class A: the AI is the product. Call the second class B: the AI is the method. Isomorphic Labs makes the scale of it concrete. Its £1.47 billion fundraise drove most of Q2's life-sciences AI investment on its own, on Beauhurst's count, and it sits in class A by the same test: what Isomorphic sells is the AI platform for drug discovery, not any single therapeutic candidate that comes out of it.

Not every company sits neatly on one side. A drug-discovery outfit that also licenses its underlying model to other biotechs is selling both things at once, and the honest answer to the substitution question can be "some of both." The equipment-failure and coatings spinouts above are useful precisely because they do not have that problem, invented clean on purpose. Most companies are not so clean, which is exactly why the test, not the label, has to do the sorting.

Get the class wrong and the error sits silently in the room. A misclassified pitch can sit through an entire meeting without either side naming what actually went wrong: the founder feels persuasive or unpersuasive, and rarely feels misclassified, which is exactly why the mistake can survive six months of otherwise careful work.

Same "AI-Enabled" label on the outside of all three. Two different bets underneath it.

## The label is not the test

The biotech founder from the opening was not lying about anything. The company genuinely used AI somewhere in its actual science, and "AI-enabled" was true of it in every sense a careful reader would check the phrase against. None of that made it the right word for the room the founder was sitting in.

"AI-enabled" is accurate for almost every spinout working in life sciences, quantum, or advanced materials today. The tools are in use nearly everywhere, and saying so is not a stretch for anybody. I would argue that AI now sits in three places inside a company like that: the product, the method, and the operations. Operations means simply how the company runs day to day: the board pack, the payroll, the customer notes, the small procedures nobody outside the company ever reads. The product and the method are what the test above actually sorts, and what an investor's cheque is betting on.

It is worth a beat to clarify that the operations are not background hum behind that bet. They are the floor it gets placed on. [This series has already made that case](operator-substrate-first-18-months.html): AI-native operations let the same founding team carry a company that used to need more hands to run, and once that is true of one spinout in the room, an investor starts expecting it of the next. A cheque is meant to fund the bet on the product or the method, not to backfill headcount the operations should already have made unnecessary, and money doing exactly that is money an investor will not commit, or will commit at a worse price. None of this is a published finding, any more than the routing claim below is; it draws on the same vantage, one step further along: the founders I advise already build this before anyone asks for it, the same unremarkable way they keep a spreadsheet, and its absence would be a concern, not its presence. That investors are starting to price it the same way is my own inference from that vantage. If it is right, what follows needs no survey of its own: a spinout that has not built this is behind the curve before the test above ever applies, whichever side of it the company lands on. The floor does not run along that line; it sits underneath both.

Leading a pitch on the label now fails twice over: it is not only accurate, it is expected, and a claim every founder in the room can make with a straight face stops telling an investor which of two very different companies they are looking at. On the operations third, "expected" has already hardened into "assumed." Nobody credits a spinout for having a bank account, and by now nobody credits one for running on AI either. In H1 2026 the investors with the biggest cheques were, by the product-method test above, mostly looking for one of them specifically.

The Royal Academy of Engineering's Spotlight on Spinouts report, published in June, put a number on how much of the UK's spinout base now sits in that world at all. Caroline Hargrove, who chairs the Academy's Enterprise Committee, said deep tech now "defines the UK spinout landscape." The same report found UK university spinouts have nearly tripled in value to £49 billion since 2020. Most of that, as far as this piece's own test can tell, sits on the class-B side of the line: the life sciences, quantum, and advanced materials companies where a licensed domain outcome is the thesis and AI is how they get there. Isomorphic Labs is in the same cohort and is not one of them. This is my own classification of the data, not the Academy's: its report does not split spinout value by mechanism class, and I cannot prove the split more precisely than that.

The strongest objection to all this is a fair one. The H1 2026 concentration numbers show where the biggest cheques went that half-year. That does not necessarily imply the door closed on everyone else. Specialist funds in quantum, in life sciences, in advanced materials, are still active; the Academy's own valuation figures are proof they have been writing cheques for years. Fair enough, as far as it goes. What it does not do is help the founder who has already spent six months in the wrong room, because the cost was the time spent looking for it in the wrong place, not the absence of class-B money.

## Two playbooks, not one

A class-A (AI *product*) founder can walk into an accelerator built around enterprise software and find people who have underwritten this exact pitch many times before: the model's performance, their data's competitive advantage, the sales motion into a large buyer. It is a well-worn path, and it is well-worn for a reason. Investors who spent H1 2026 writing nine-figure cheques into AI infrastructure and enterprise software know exactly what they are looking at when they see it again, and the diligence questions they ask are the ones the whole ecosystem has already standardised.

A class-B (AI *method*) founder walking into the same room is speaking a language nobody there was hired to evaluate. What a specialist deep-tech investor wants to know is different in kind, not just in emphasis: what a peer-reviewed proof of concept looks like, where the regulatory pathway sits and when it starts to bind, what the licensing terms mean for whether the company can actually own its own outcome once it is proven. A class-B founder's route runs through technology transfer offices at institutions that have closed deep-tech rounds before, through corporate venture arms for whom the domain outcome is already a strategic priority, and through programme managers who have placed founders in front of the small number of funds actually built for this.

Send a class-B document into a class-A room and it reads as unambitious: too many caveats, not enough velocity. Send a class-A document into a class-B room and it reads as thin: all momentum, no proof. Neither audience is wrong to react that way. They are reading a document built for somebody else's company. Under both buildings runs the same floor. Neither room tests for it.

Neither path is harder than the other. They are different buildings, and a founder who has not classified their mechanism first is knocking on the wrong one's door.

This is not the only sorting decision that happens before a founder thinks to look for it. A public funder runs one of its own: UKRI's plan for spinout support [sorts companies into two tracks](ukri-selection-gate.html) months before its own documentation deadline in November.

"AI-enabled" will still be true of the biotech founder's company on the day they finally close whichever round they are actually in the market for. The question that mattered was what survives if you take the AI away, and a label cannot answer that on a founder's behalf. Before the next deck goes anywhere, it is worth writing one sentence in two halves: what the company actually sells, and whether the AI is that thing or only the method for reaching it. Everything an investor asks afterwards is really a question about the first half.

## Paired prompt kit

**[Mechanism Class Classifier](/toolkit/mechanism-class-classifier/):** three prompts that classify a research mechanism as class A or class B, map the classification to the matching investor pool, and draft the one-sentence descriptor this piece asks for, run entirely against a founder's own description of the mechanism, with no external inputs required.

## Sources

- HSBC Innovation Banking UK / Dealroom. "UK Innovation update H1 2026." Published approximately 4 July 2026. https://www.hsbcinnovationbanking.com/gb/en/resources/uk-innovation-update-h1-2026
- Beauhurst / Kirstie Pickering, UKtech.news. "UK's AI investment reaches record £4.56bn in Q2." Published 16 July 2026. https://www.uktech.news/ai/uks-ai-investment-reaches-record-4-56bn-in-q2-20260716
- Royal Academy of Engineering. "Spotlight on Spinouts 2026." Published 3 June 2026. https://raeng.org.uk/news/spotlight-on-spinouts-2026/

---

## Evidence note

- **The H1 2026 totals.** "UK startups raised $17bn in H1 2026 (+102% YoY)," the strongest first half since 2022, per HSBC Innovation Banking and Dealroom. The body carries the AI concentration figure, $12.6bn and "almost three quarters"; the overall total and the growth rate sit here.
- **The megaround counts.** HSBC's report states that AI companies "accounted for all four $1billion+ funding rounds completed during the period." The same page carries two different megaround counts in different sections: "17 of 26" in its opening key facts and "19 of the UK's 28 megarounds" in its main prose. That is an inconsistency on the publisher's own page, confirmed against a live re-read of it, and neither figure is used in the body.
- **Enterprise software AI investment.** "Enterprise software attracted the highest levels of AI investment ($5.2 billion)" in H1 2026, the largest single AI sub-sector on HSBC and Dealroom's count. The opening scenario carries no figure of its own: the founder there is a composite, not a sourced case.
- **Late-stage funding share.** Per HSBC Innovation Banking and Dealroom: "late-stage rounds accounting for 68% of all capital raised, up from 42% a year earlier and above the European average of 59%". Not carried in the body.
- **The Beauhurst Q2 concentration.** Per Beauhurst's Q2 analysis: "eight £100m+ AI equity raises accounted for almost 80% of disclosed AI investment in the quarter, with fundraising rounds by Isomorphic Labs, Ineffable Intelligence, NScale, PhysicsX and Fractile driving a significant share of the total." The body describes this qualitatively, as a short list of very large raises doing most of the work, without the count or the percentage.
- **The Hargrove quote in full.** Dr Caroline Hargrove CBE FREng, Chair of the RAEng Enterprise Committee, in the Royal Academy of Engineering's Spotlight on Spinouts report: "Deep tech now defines the UK spinout landscape, accounting for 96% of total value and underpinning much of the country's innovation strength". The body quotes only the opening clause; the 96% figure is preserved here.
- **The £49 billion and what it does not say.** "UK university spinouts have nearly tripled in value to £49 billion since 2020" is the Royal Academy of Engineering's figure. That most of that value sits in class-B companies is this piece's own reading of it, and the body says so. The report does not split spinout value by mechanism class, and Isomorphic Labs, class A on this piece's test, sits in the same reported cohort. Do not read "most" as the Academy's own finding.
- **Isomorphic Labs.** Per Beauhurst's Q2 2026 count: "life sciences and health securing the most investment by value, with 36 deals raising £1.52bn – driven primarily by Isomorphic Labs' £1.47bn fundraise". The body's "drove most of Q2's life-sciences AI investment on its own" paraphrases "driven primarily by"; the £1.47bn figure is as the source states it.
- **"AI-enabled" routing toward class A.** Not a published finding. The body marks it as the author's own read from advising founders on these pitches.
- **Operations as the floor.** Not a published finding either. The claim is that AI sits in three places in a spinout, the product, the method and the operations; that the first two are what the test sorts and what a cheque bets on; and that the third has become something an investor expects to find already built, so that a cheque backfilling headcount those operations should have made unnecessary is a cheque an investor will not write, or will write at a worse price. A spinout that has not built it is behind before the test applies, on either side of the line. This is the author's practitioner read, one step on from the vantage declared below, and the inference that investors are pricing it that way is his own. Its other basis is this series' earlier argument, linked in the body rather than restated. No external source or statistic stands behind these paragraphs.
- **The author's advisory vantage (first-person).** The author has been Impact Acceleration Manager in the Faculty of Engineering and Physical Sciences at the University of Surrey since January 2020, advising academics from proof of concept through licensing and spinout. The body carries the role and the vantage only, with no company named. First-person attestation from lived experience, not a public-sourced claim for this piece. The two spinouts illustrating class A and class B are composites, invented for the comparison and drawn from no real company; the opening founder scene carries the same licence. Isomorphic Labs is the section's one real example, cited to Beauhurst.