Large tokenisation numbers arrive in the inbox most weeks, and readers have learned to discount them. Oliver Wyman‘s Global Financial Market Infrastructure 2026 report carries several: $45 billion of existing market infrastructure revenue that could move onto tokenised rails, $25 billion of new revenue from tokenised markets, a quarter of sector revenue exposed to artificial intelligence, and a nine point margin uplift by 2030. The Fintech Times asked the firm to defend the figures and to name who wins and who loses, rather than restate the direction of travel.
Hiten Patel, partner and global head of financial infrastructure, technology and services at Oliver Wyman, answered in writing. His team advises market infrastructure, data, technology and software providers across financial services, and he also leads the firm’s corporate and institutional banking practice in Europe. His answers separate the revenue that moves from the revenue that disappears, identify the category of firm most exposed to AI, and set out what would have to be true in two years for Oliver Wyman to conclude its tokenisation thesis was wrong.
Two numbers that measure different things
The $45 billion and the $25 billion are often read as one story. Patel says they are not. The first is existing financial market infrastructure (FMI) revenue that could migrate onto tokenised infrastructure. The second is the gross new revenue that tokenisation could create.
“The $45 billion represents around 35 per cent of today’s FMI revenue pool. Importantly, we do not expect all of that revenue to disappear. We estimate that around $7 billion could be structurally eliminated as tokenisation removes legacy frictions, manual processes and some float-related economics. The remaining $38 billion is effectively up for grabs: the question is who captures issuance, registry, clearing, custody, settlement, servicing, data and workflows as those functions move onto new infrastructure.”
The $25 billion is a different measure. It represents gross new revenue under a full-penetration scenario, as institutional market infrastructure expands into crypto post-trade and derivatives, syndicated loans, money market funds and private markets. “After accounting for the roughly $7 billion of structurally eliminated revenue, that would imply around $18 billion of net sector growth, taking the revenue pool from approximately $131 billion to around $150 billion.”
The caveat, in his account, is timing. “The $25 billion is a full-penetration scenario, not a forecast that $25 billion of new revenue will materialise by 2030.” He watches both figures: “one tells you how much of today’s economics could move; the other tells you how much bigger the market could ultimately become.”
Exposure is not erosion
The report’s second headline figure, that AI could erode a quarter of sector revenue, draws a similar correction. “We are not saying that 25 per cent of FMI revenue will be eroded. We are saying that 25 per cent is exposed to AI, and the degree of exposure varies significantly.”
Oliver Wyman assessed FMI revenues against six sources of defensibility: network effects, workflow embeddedness, mission criticality, regulatory and audit relevance, proprietary intellectual property, and trust and authority. Around 9 per cent of sector revenue, approximately $12 billion, falls into what the firm calls the ‘Contest’ category and faces the most direct exposure. A further 16 per cent, approximately $21 billion, sits in the ‘Augment’ category. “Those revenues remain defensible, but providers will need to embed AI into their products and workflows to sustain differentiation and pricing power.”
Asked to name the firms that lose, Patel points to a type rather than a list. “Businesses whose value proposition rests primarily on generic analytics, non-proprietary datasets, non-embedded reporting or shallow workflow functionality. AI makes those capabilities easier to replicate. By contrast, an authoritative benchmark, proprietary reference dataset, participant network or deeply embedded risk workflow is much harder for an AI model alone to recreate.”
The nine point margin figure comes from the cost side. Oliver Wyman sees scope for around a 20 per cent reduction in costs across software engineering, operations, risk, compliance, surveillance and corporate functions, which in its modelling takes sector EBITDA margins from 54 per cent to 63 per cent by 2030. “But that uplift will not come from isolated AI pilots. It requires FMIs to industrialise AI through reusable technology, shared data, measurable productivity targets and robust governance.”
The $100 billion test
Between 2020 and 2025, FMIs deployed more than $100 billion on acquisitions, the great majority directed at data and technology. The report frames new infrastructure as a stress test of that spending. Asked how much of it was wasted, Patel declines to give a figure.
“I would not put a number on how much of that $100 billion was ‘wasted’. Many of those acquisitions have delivered real value through recurring revenues, broader customer reach, new technology capabilities and cross-sell. But AI is changing the test of which of those assets will remain differentiated.”
He offers three questions instead. “First, does the asset strengthen network effects, trusted infrastructure or regulatory relevance? Second, is it embedded deeply enough in clients’ day-to-day workflows that it is difficult to substitute? Third, does it own or create proprietary, authoritative data or intellectual property that is difficult to replicate?” Assets built around generic analytics, aggregated non-proprietary content or shallow workflow functionality are the exposed ones, and around $12 billion of current FMI revenue sits in that contestable category. “That does not mean $12 billion simply disappears. It means providers will have to work much harder to defend it.”
“The lesson for the next M&A cycle is not ‘do not diversify’. It is that diversification needs to be defensible and genuinely connected to the core franchise.”
Compute as an asset class
The report offers prediction markets and compute as new frontiers for market infrastructure. Compute as a tradable asset needs a forward curve, standardisation and something like a clearing model, so The Fintech Times asked what exists today and what is still a slide.
“Compute is already a real commercial market. What does not yet exist at scale is a mature financial market around it.” AI developers, enterprises and model providers already buy capacity from hyperscalers, neoclouds and specialist providers, often priced per GPU-hour, and procurement is moving from on-demand usage towards combinations of spot access, reserved capacity and forward commitments. Providers carry utilisation and revenue risk; users carry future cost and capacity risk. The rationale for hedging, Patel says, is real.
What is missing is the standardised, transparent and liquid forward curve, with the benchmarks, clearing and market infrastructure that a mature commodity market would have. “Standardisation is the hardest problem. One GPU-hour is not necessarily fungible with another. Its value depends on the chip, memory, networking, latency, location, software environment and service quality.” A deeper financial market therefore needs credible contract specifications, quality tiers, delivery conventions, reference pricing and independent performance verification. Benchmark providers are beginning to address this, but he places the market at an early stage, and because compute is heterogeneous and often regional he doubts the end state is a single global listed contract. A mix of bespoke over-the-counter forwards and request-for-quote markets, alongside listed and centrally cleared contracts where standardisation and liquidity emerge, is in his view more plausible.
“So compute is not ‘just a slide’. The physical market exists and the hedging need is real. What remains nascent is the financial infrastructure around it, and that is precisely where FMIs and data providers have an opportunity to help shape the market.”
What would prove the thesis wrong
Money has been about to move onto tokenised rails for several years. The Fintech Times asked what would have to be true in two years for Patel to conclude the thesis was simply wrong, and asked for a specific answer rather than a diplomatic one.
“If, two years from now, tokenisation is still primarily a collection of pilots with fragmented activity, limited liquidity and narrow institutional participation, we would have to revisit our thesis.” The real test, he says, is whether tokenisation moves beyond technical feasibility to meaningful institutional scale: recurring activity, sufficient liquidity and broad participation.
The benefits also have to reach issuers and investors. “If tokenisation simply changes the infrastructure underneath a market, or shifts economics between intermediaries, without improving outcomes for end participants, the investment is difficult to justify. The test is whether it makes the market better, not just whether it changes the plumbing.”
Oliver Wyman does not expect wholesale adoption across every asset class or every part of the value chain. Patel describes a more realistic path of selective scaling where tokenisation solves a genuine economic problem, and names money market funds, tokenised securities, collateral and repo as examples. In two years he would expect the leading use cases to show repeatable institutional activity and tangible benefits: more efficient collateral mobilisation, lower market friction, improved liquidity or better access.
“If those outcomes are not emerging, then we should not simply push the adoption curve further into the future. We would need to question whether the opportunity is narrower than we currently believe.”
The report’s revenue and margin scenarios run to 2030. Patel’s own test for the tokenisation thesis falls due earlier, two years from now.
The post Oliver Wyman’s Hiten Patel on the $45 billion Tokenisation Shift appeared first on The Fintech Times.
