AI trading market size: eight estimates that cannot be reconciled

Published estimates of the AI trading market range from USD 2.53 billion to USD 57.65 billion for overlapping periods — a spread of roughly 22 times. That is not a measurement problem to be resolved by picking the most credible firm. It is a definitional problem: each estimate draws the market boundary somewhere different, so each is measuring a different thing. This page publishes all eight side by side, explains why the spread exists, and sets out what about AI trading adoption is actually measurable.

Published 27 August 2026 · Updated 28 August 2026 · AI Trading Book Editorial · Reading time about 11 minutes

The eight published estimates

Each row below comes from a named research firm. We have not adjusted, harmonised or reconciled them, because doing so would require assumptions about boundaries the firms do not publish.

Why the spread is definitional, not statistical

If seven firms measured the same quantity with imperfect methods, you would expect their estimates to cluster with some dispersion. A 22-fold range is not dispersion. It means the firms are counting different things, and the differences are systematic rather than random.

The boundary decisions that produce most of the gap are these. Does the market include only software licences for AI trading tools, or also the data feeds, cloud compute and execution infrastructure those tools depend on? Does it cover retail platforms, institutional systems, or both? Does it count algorithmic trading generally — the large majority of which is execution automation, not machine learning — or only AI-driven systems? Does it include robo-advice platforms, which are automated but not usually trading systems in this sense? And is it a global figure or one region extrapolated?

Each of those choices is defensible, and each moves the total by a multiple. A report scoped to retail AI trading software licences and one scoped to AI across capital-markets infrastructure will differ by an order of magnitude while both being competently produced.

Why averaging produces a fake number

Averaging estimates is appropriate when they measure the same quantity and their errors are independent. Neither condition holds here. The mean of these figures is approximately USD 18 billion, and that number describes nothing: no firm produced it, no definition supports it, and it cannot be traced to a source.

The failure mode is worse than imprecision. An averaged figure loses its boundary, so a reader cannot tell what was counted — and once published without attribution it circulates as an apparent fact. This is the mechanism by which a commercial estimate becomes an industry consensus that no one actually produced.

The same reasoning applies to picking the estimate closest to the middle, or to quoting a range as though the extremes were error bars. USD 2.53bn is not a low estimate of the same thing that USD 57.65bn is a high estimate of.

What is actually measurable

Market size is not measurable in this category. Adoption is, because regulators survey defined populations and publish their methods. These figures are less exciting and considerably more useful.

Note what these do and do not say. The 85% figure measures automation of any kind, not machine learning. The 75% adoption figure covers AI anywhere in a firm, with process optimisation, cybersecurity and fraud detection as the leading use cases rather than trading. Neither is a market size, and neither can be converted into one.

Figures we excluded

Several widely repeated numbers did not survive source checking and are therefore absent from this page. We name the pattern rather than the individual claims, because repeating an unverifiable figure to debunk it still circulates it.

• Three incompatible series attributed to the same firm. During research, three different base-year values for 2024 were each attributed to Grand View Research in different secondary sources. Only one could be matched to a named report, and it is the one in Table 1. Multiple conflicting figures under one firm's name is a reliable signal of second-hand fabrication.
• Figures with no named report. Several estimates circulate attributed only to a firm, with no report title, publication date or scope statement. Without those, the figure cannot be checked and does not meet the standard set in our editorial policy.
• Adoption percentages from self-selected surveys. Retail trading surveys with small, self-selected samples circulate as adoption statistics. They measure who answered a survey, not who uses AI.

How to cite these numbers responsibly

1. Name the firm, the figure, the base year and the scope, or do not use the number.
2. Never average, and never present the range as error bars. The extremes are different questions, not bounds on one answer.
3. Match the estimate to your question. Writing about retail tools? The narrow estimates are the relevant ones. Writing about institutional infrastructure? The broad ones.
4. Prefer adoption measures where they answer the question. "75% of 118 surveyed UK firms use AI" is verifiable in a way that no market size figure is.
5. Do not treat growth as evidence of performance. Market size measures spending on tools, not returns from them.

What we could not establish

• Any official statistical series for AI trading. No statistical agency in the UK, Australia, New Zealand, the EU or the US publishes this category. Data not found.
• Published methodologies behind the eight estimates. Scope statements are generally behind paywalls or absent, which is why the boundary analysis in Table 2 is reasoning about plausible causes rather than a reconciliation. Data not found.
• Any market size figure specific to Australia, New Zealand or the United Kingdom. The published estimates are global. Data not found.
• Retail versus institutional split within any estimate. Data not found.

Key takeaways

• USD 2.53bn to USD 57.65bn is the honest answer, together with the reason the range is that wide.
• The spread is definitional. Seven firms answering seven questions, not seven attempts at one.
• Averaging manufactures a sourceless number that fails the moment anyone checks it.
• No official statistic exists, so every figure in circulation is commercial.
• Adoption is measurable where market size is not — regulator surveys with stated populations.
• Spending is not performance. A growing market says nothing about whether the tools work.

Frequently asked questions

How big is the AI trading market?

There is no single defensible answer. Published estimates for overlapping periods range from USD 2.53 billion to USD 57.65 billion — a spread of roughly 22 times. The correct response to the question is the range and the reason for it, not a number.

Why do market size estimates differ so much?

Because each firm draws the market boundary differently. Some count AI trading software licences only; others include the whole algorithmic trading market, AI across financial services, robo-advice platforms or institutional infrastructure. Different boundaries produce different totals from the same underlying economy.

Can I average the estimates to get a reasonable figure?

No. Averaging assumes the estimates measure the same quantity with independent errors. They measure different quantities, so the mean of a 22-fold spread describes nothing that exists. It produces a number with no source and no definition.

Is there an official AI trading market statistic?

No. No statistical agency in the UK, Australia, New Zealand, the EU or the US publishes AI trading as a category. Every figure in circulation comes from a commercial research firm using its own definition.

Which estimate should I cite?

Cite a specific firm, its figure, its base year and its definition — for example, Grand View Research at USD 21.06 billion for 2024 — rather than a bare number. Any citation without the source and boundary is unverifiable and will not survive scrutiny.

What growth rates are being forecast?

Compound annual growth rates in the published estimates range from 6.0% to 16.7%. The lowest and highest come from firms whose base-year figures differ by more than an order of magnitude, so the growth rates are not comparable either.

What can actually be measured about AI trading adoption?

Regulator surveys, which have defined populations and stated methods. The Bank of England and FCA found 75% of 118 surveyed firms using AI. ASIC reviewed 624 use cases across 23 licensees. ASIC also estimates algorithmic trading at roughly 85% of Australian listed equities turnover.

Are algorithmic trading volume figures more reliable than market size figures?

Somewhat, because volume can be observed rather than modelled. Select USA reports 60 to 75% of turnover in US equities and other developed markets as algorithmic. But these measure automation of any kind, not AI, and definitions still vary between sources.

Why would a research firm publish an inflated boundary?

Not necessarily inflation — different reports serve different buyers. A report aimed at infrastructure vendors reasonably counts hardware and data feeds; one aimed at software vendors counts licences. The problem arises when a figure is quoted without the boundary that produced it.

Does the market size tell me anything about whether AI trading works?

No. Market size measures spending on tools, not returns produced by them. A growing market is evidence of adoption and marketing, not of performance. The performance evidence sits in fund index data and peer-reviewed research, covered on our evidence page.

How should a journalist or analyst handle these numbers?

Quote one firm with its definition and date, or present the range and say why it is wide. The one thing to avoid is a bare figure with no attribution, which is how a single commercial estimate becomes an apparent industry fact through repetition.

What figures on this topic could not be verified?

Several widely circulated figures could not be traced to a named report — including three mutually incompatible series all attributed to the same research firm. Where attribution failed, we exclude the figure rather than repeat it.

Sources

• Fortune Business Insights — USD 2.53bn (2025) to USD 4.33bn (2034), CAGR 6.0%.
• Coherent Market Insights — USD 3.59bn (2026) to USD 6.68bn (2033), CAGR 9.3%.
• Dataintelo — USD 4.2bn to USD 9.8bn, CAGR 9.8%.
• IMARC Group — USD 18.8bn to USD 43.2bn, CAGR 9.39%.
• ReAnIn — USD 20.58bn to USD 43.27bn (2032), CAGR 11.2%.
• Grand View Research — USD 21.06bn (2024) to USD 42.99bn (2030), CAGR 12.9%.
• Straits Research — USD 57.65bn to USD 150.36bn (2033), CAGR 12.73%.
• Technavio — incremental growth of USD 23.94bn (2025–30), CAGR 16.7%.
• Bank of England and FCA — "Artificial intelligence in UK financial services – 2024", 21 November 2024, n=118 — adoption 75%, planned 10%, automated decisioning 55%, fully autonomous 2%.
• ASIC — REP 798, 29 October 2024 — 624 AI use cases across 23 licensees; 61% intending to increase use.
• ASIC — "ASIC moves to modernise trading system rules", 27 August 2025 — algorithmic share of approximately 85% of listed equities and about 94% of SPI 200 futures.
• Select USA — 60–75% of turnover in US equities, EU and major Asian markets as algorithmic; plateau around 70–80%.

Informational research only. Nothing on this page is personal financial, legal, tax or investment advice. Commercial market-research figures are reproduced as published; we have not audited the methodologies behind them and in most cases those methodologies are not public.