ALLOCATOR SERIES FOR FUNDS OF FUNDS · No. 4 of 5
Why AI Will Be Good for Fintech
This is our fourth article in a series of 5 aimed at fund of funds and asset allocators in general. You can find the first three articles here, here and here respectively.
Five software moats are eroding and five are holding — and the five that hold are fintech’s. A framework for underwriting durability in the AI era.
Pascal Bouvier, MiddleGame Ventures · August 2026
In 2025, AI companies captured roughly half of all venture capital deployed globally — about $211 billion. That is the crowded trade, and its price already assumes much of its promise. The more interesting question for an allocator is not whether to buy intelligence at any valuation; it is where AI changes the value of everything else. Our answer, argued publicly in our April letter and tested weekly in our own portfolio, is that nowhere does the AI era redraw the map more favorably than in fintech.
The public markets show why the question matters. Over the past year, the so-called SaaSpocalypse erased nearly $300 billion of market capitalization from listed B2B software companies; a single product announcement from Anthropic wiped over $15 billion off cybersecurity stocks in a day. Markets are repricing software because AI attacks the reasons customers historically stayed. The repricing is real — but it is indiscriminate, and indiscriminate repricing is where specialists earn their keep. The right instrument is not a sector view. It is a moat-by-moat view.
Ten moats: five falling, five holding
In our April 2026 GP letter — The AI Follies — we adopted a framework from Nicolas Bustamante that identifies the ten economic moats of vertical software and asked which survive capable AI agents. Our conclusion: five are being weakened. Learned interfaces — years of UI familiarity — dissolve when natural language becomes the interface. Custom workflows and business logic can now be described by the practitioner who holds the knowledge, without an engineering bottleneck. Public data access is fully commoditized. Talent scarcity inverts when subject-matter experts can build software without deep engineering benches. Bundling weakens as building new features becomes cheap.
Five moats are holding, and some are strengthening. Private and proprietary data becomes more valuable as models make data the scarce input. Regulatory and compliance positions cannot be prompted into existence — licenses, supervision and enforcement are not code. Network effects remain structural. Transaction embedding — sitting inside or adjacent to the money flow — keeps its switching costs. And the system of record, where the authoritative state of an asset or transaction lives, remains the stickiest asset in software: core banking, core payments, custody.
Now the observation that matters for this series, quoted directly from the letter: the five moats left standing are found, in totality or in combination, in fintech. You cannot vibe-code a banking license. A compliance perimeter, an embedded position in a payment flow, a golden record of who owns what — these are precisely what AI cannot commoditize, and precisely what defines the businesses we underwrite. Fintech is structurally more AI-resilient than horizontal software, and the market’s indiscriminate repricing of everything with a subscription model is mispricing that difference.
Resilience is the defensive half. Demand is the offensive half
If the moat analysis explains why fintech survives AI, the spending data explains why it profits from it. Financial institutions are not debating the technology; they are buying it. AI spending in the financial sector exceeded $75 billion in 2026 and is forecast to reach roughly $125 billion by 2028 — a 29% compound growth rate. As of early 2026, 52% of financial institutions were piloting agentic AI or had moved beyond piloting into scaling. And the deepest pool of demand is the one regulation itself creates: DORA compliance, AML, onboarding, reporting — obligations written into law, with deadlines, whose cost of delivery AI resets. Regulation writes the purchase order; AI collapses the cost of serving it; the vendors in the middle are fintechs. For a category the market is repricing as a victim of AI, that is a remarkable amount of AI-driven revenue arriving on schedule.
New categories, not just cheaper old ones
The third effect is the one we find most underwritten: AI is creating financial infrastructure categories that did not exist. Agentic commerce is the clearest example. When software agents transact with each other — procuring, subscribing, settling — the billing and payment rails built for humans with credit cards stop fitting. Bloomberg Intelligence projects generative AI driving roughly $1.3 trillion in revenue by 2032; every dollar of it needs to be priced, billed, settled and reconciled by something. We led the pre-seed round of Paygentic, which is rebuilding exactly that payments stack for AI-native businesses, on this thesis. The same logic carried us into Theia Insights, whose AI-driven classification platform replaces static market taxonomies for institutional investors — we led its $8 million Series A in March. These are not AI features bolted onto fintech; they are fintech businesses that could not exist before AI.
And history argues the demand effect compounds. As we wrote in April: the advent of Excel did not doom accountants — it exploded the demand for accounting. When the cost of building financial software collapses, the number of financial software businesses multiplies, and every one of them needs the regulated primitives — payments, custody, compliance, identity, data. In a gold rush of company formation, fintech owns the shovels.
What an allocator should do with this
Three practical implications. First, underwrite moat by moat, not sector by sector — “fintech” contains both AI’s victims and its beneficiaries, and the ten-moat lens separates them. Second, interrogate your managers with it: ask which of the ten moats each portfolio company owns, which are eroding, and whether the manager can defend the answer in writing. A GP still pattern-matching from the SaaS decade — ARR multiples, land-and-expand, UI stickiness — is underwriting moats that are dissolving. Third, treat the market’s fear as your entry price: repricing driven by a sector-level narrative, applied to businesses whose moats are strengthening, is the textbook definition of a buyer’s market.
None of this is AI maximalism. As we argued in the letter, the truth will be more measured than the pundits on either side allow — every S-curve meets friction, and timing remains genuinely uncertain. But the direction is clear enough to invest against: AI compresses the value of everything undifferentiated and raises the value of everything regulated, embedded, and record-keeping. That is fintech’s side of the ledger.
This is No. 4 in a five-part series for allocators. Next: European fintech as a distinct allocation — and the sovereign tailwind behind it.
Sources
MiddleGame Ventures, GP Letter – April 2026, “The AI Follies” (middlegamevc.com/articles/gp-letter-april-2026) — Ten Moats analysis, SaaSpocalypse (~$300B), Anthropic/cybersecurity selloff (>$15B), Excel argument.
Nicolas Bustamante — original ten-moats framework for vertical software (cited and adapted in our April letter).
AI share of global venture funding 2025 (~$211B, ~half of total) — industry data as cited in MGV thematic letters.
Statista — financial-sector AI spending >$75B (2026), forecast ~$125B by 2028 (~29% CAGR); agentic AI adoption in financial services: 52% piloting or beyond, early 2026.
Cambridge Centre for Alternative Finance, “2026 Global AI in Financial Services Report” (jbs.cam.ac.uk).
Bloomberg Intelligence — generative AI toward ~$1.3T revenue by 2032 (via MGV Paygentic announcement).
MiddleGame Ventures — “Our Investment in Paygentic” (Oct 2025) and “Our Investment in Theia Insights” (Mar 2026), middlegamevc.com/updates.