AI Now Drives 89% of Global Trading Volume*

Research Note

Abstract

Automation has become the default operating mode of modern financial markets. A widely circulated industry estimate claims that "AI-driven systems" account for roughly 89% of global trading volume. While this headline is directionally consistent with the reality that most orders are routed and executed electronically, the exact percentage depends heavily on definitions (AI vs. algorithmic execution vs. electronic workflows), asset class, and what is being measured (orders, trades, volume, or notional).

This research note clarifies the terminology, reviews the strongest available public evidence, and explains how AI and algorithms reshape market structure, liquidity, and price formation. It also provides practical implications for retail traders, experienced discretionary traders, and quantitative teams—with emphasis on what has changed, what remains stable, and what can be measured reliably.

Note: "89%" is treated here as a headline estimate rather than a regulator-verified statistic. We explain why measurement is difficult and how to interpret such figures.

1. Introduction

Over the last two decades, markets have shifted from human-centered trading (phone orders, manual routing, single-venue execution) to machine-mediated trading (electronic venues, smart order routing, automated market making, and systematic execution). Regulatory and industry studies consistently show that a large share of activity is now driven by automated workflows—especially at the execution layer. For example, Coalition Greenwich reported that ~37% of U.S. equity volume in 2023 was executed through algorithms and/or smart order routers (in the context of buy-side execution practices).

At the same time, public commentary often compresses multiple ideas into "AI trading." In practice, markets are dominated by automation, while "AI" (machine learning models making or adapting decisions) is unevenly deployed and is most visible in execution optimization, surveillance, and signal research.

2. Definitions (What People Mean When They Say "AI Trading")

To make the discussion useful—and measurable—we separate three layers:

1. Electronic trading (e-trading)

Orders are sent, matched, and reported electronically (this is now standard in most major markets).

2. Algorithmic execution

Rules or optimization procedures split orders (VWAP/TWAP), route them across venues (smart order routing), and manage market impact. Coalition Greenwich's "algorithms and/or smart order routers" statistic fits here.

3. AI-driven decision-making

Models that learn from data (machine learning) or adapt policies over time; used for signal discovery, forecasting, and dynamic execution/risk controls.

Key point: A market can be "mostly algorithmic" without being "mostly AI" in the strict technical sense.

A single global percentage is difficult to validate because:

No single consolidated global tape: Market structure differs by asset class and jurisdiction; fixed income transparency is still evolving, and definitions vary by venue and participant type.
Orders vs. trades vs. volume: High-speed participants generate large message traffic and cancellations; "algorithmic dominance" looks different depending on the metric. Research and regulatory discussions frequently emphasize that modern markets involve extremely high cancellation activity (orders that never become trades), which complicates interpretation.
Different "algorithmic" baselines:
- In equities, some estimates focus on HFT share of volume (often discussed separately from all algorithmic execution). A U.S. Congressional Research Service overview (2016) cited HFT at roughly ~55% of U.S. equity volume and ~40% in European equity volume (noting that estimates vary).
- In buy-side execution surveys, the focus is often on institutional execution methods (algorithms/SOR) rather than market-wide automated market making. Coalition Greenwich's reported ~37% figure for U.S. equities (2023) is an example of this framing.
"AI" is often inferred, not observed: Public market data rarely labels "AI-generated" decisions. Even regulators and academic researchers often infer algorithmic behavior from observable patterns rather than direct system attribution.

Practical interpretation: "89%" can be understood as a broad claim that machines dominate end-to-end trading workflows (generation, routing, execution) across asset classes. But readers should not treat it as a universally measured, regulator-certified statistic.

4. Evidence Snapshot (What We Can Cite with Confidence)

4.1 Equities: Execution Automation Is Large and Growing

~37%

of 2023 U.S. equity volume executed through algorithms and/or smart order routers

Coalition Greenwich reported (survey-based, execution-method framing).

4.2 High-Frequency Trading: A Major Component of Equity Volume

~55%

of U.S. equity volume

~40%

of European equity volume

A U.S. Congressional Research Service overview (2016) reported that HFT "by most accounts" represented these percentages (acknowledging uncertainty).

4.3 Regulation and Oversight: Algorithms Are Now a Core Policy Topic

Regulators have produced staff reports and rulemaking efforts focused on algorithmic trading and market structure (fragmentation, order types, and market events).

4.4 Derivatives and Emerging Markets: Algo Oversight Is Expanding

India's regulator (SEBI) introduced "track and trace" requirements for retail algorithmic orders, reflecting the growing scale and regulatory priority of algorithms.

5. What Actually Changed in Market Structure

5.1 Liquidity Provision Became More Automated

In many liquid instruments, market making is now substantially automated. Algorithms continuously update quotes and manage inventory based on volatility, order flow, and hedging conditions. This often tightens spreads in normal conditions, but liquidity can thin rapidly under stress.

5.2 Trading Became Faster—and More Fragmented

Automation pushed trading into a multi-venue environment with smart routing, co-location, and latency competition. Recent reporting shows that even venue infrastructure (e.g., ultra-low latency connectivity) can become a regulatory concern when it creates perceived fairness issues.

5.3 Risk Transmission and "Crowding"

When many participants use similar signals and execution logic, markets can become more correlated and more fragile during regime shifts. A BIS working paper (2025) links HFT-related changes in trading speed to broader market effects, underscoring that "faster" is not always "safer."

6. What "AI" Adds Beyond Traditional Algorithms (Plain-English View)

Many trading systems are still rule-based at their core. Where AI most often adds value:

Execution optimization

Choosing when and where to execute based on live liquidity and predicted market impact.

Detection and classification

Identifying regimes, anomalies, and microstructure states.

Research acceleration

Feature discovery, signal testing, and adaptive parameter selection (with strict controls to avoid overfitting).

Surveillance and controls

Monitoring behavior, preventing runaways, and supporting compliance tooling.

In other words, AI is frequently an upgrade to how trading is executed and managed—not a guarantee of alpha.

7. Implications for Retail Traders (Actionable, Non-Hype)

Retail traders do not need microsecond speed to compete, but they must trade in a market shaped by machine behavior.

Use limit orders more intentionally: In fast markets, market orders can suffer slippage, especially around news and session opens/closes.
Avoid low-liquidity windows: Spreads and volatility can widen sharply when automated liquidity pulls back.
Trade timeframes that reduce microstructure noise: Many retail strategies improve when they avoid the shortest horizons dominated by queue position and routing effects.
Expect "clean" levels to be probed: Stops at obvious locations are easier for automated flows to detect and interact with. Plan invalidation levels with volatility in mind.
Treat indicators as context, not triggers: The edge is increasingly in process (risk control, execution discipline, regime awareness), not in a single indicator.

8. Implications for Experienced Traders and Portfolio Managers

Execution is alpha-adjacent: Poor execution can erase a good signal. Algorithms and routing decisions often matter more than small improvements in entry logic.
Liquidity is conditional: Liquidity is plentiful until it is not. Build playbooks for volatility spikes and correlation shocks.
Understand venue behavior: Fragmentation means the "best" venue varies by time, volatility, and instrument.
Measure what you do: Track slippage, fill quality, and post-trade outcomes. Even discretionary trading benefits from basic transaction cost analysis (TCA).

9. Implications for Quant Teams (Useful Without Over-Engineering)

Separate signal research from execution research: A signal that works on close-to-close data may fail in live trading due to costs and impact.
Use realistic backtests: Include spreads, fees, and conservative slippage; stress test at 2× costs.
Control overfitting: Walk-forward validation and robust parameter ranges matter more than complex model architecture.
Capacity matters: Many short-horizon edges collapse as size increases; measure impact and queue dynamics where possible.
Be explicit about definitions: "AI share" and "algorithmic share" must be defined in your reporting (trades, volume, orders, notional, or message traffic).

10. Risks and Policy Considerations

Market stability: Feedback loops and synchronized de-risking can amplify moves (flash events).
Fair access: Co-location and latency advantages raise fairness debates.
Transparency: Complex systems are hard to audit; regulators increasingly require traceability and controls, as seen in SEBI's retail algo framework.

11. Outlook (What to Expect Next)

More automation in less-electronic markets: Fixed income and fragmented markets continue moving toward greater electronic execution.
AI concentrates in execution and controls: The near-term "AI edge" is more likely to be better routing, sizing, and risk management than magical prediction.
Regulation will keep tightening: Audit trails, algo registration, and controls will expand as retail access grows.

12. Conclusion

The core reality behind the "89%" headline is that machines dominate modern trading workflows—especially the execution layer. The exact percentage is not a universal constant and varies by definition and market, but the strategic implication is stable: market participants operate in an environment where liquidity, price discovery, and execution quality are heavily shaped by automated systems.

Retail traders can still succeed by choosing appropriate horizons, respecting volatility, and improving execution discipline. Experienced traders and quants should treat market microstructure and transaction costs as first-class components of strategy design. The next phase of "AI trading" is less about replacing humans and more about expanding automation in execution, monitoring, and decision support—while regulators push for stronger traceability and risk controls.

References

Liquidity Finder (2025). "AI for Trading: The 2025 Complete Guide" (source of the widely circulated "~89%" estimate). LiquidityFinder
Coalition Greenwich / Coalition Greenwich press reporting (2024). U.S. equity execution statistics citing ~37% volume executed through algorithms and/or smart order routers in 2023. Coalition Greenwich
U.S. Congressional Research Service (2016). High-Frequency Trading: Overview of Recent Developments (HFT share estimates). Every CRS Report
U.S. SEC (2020). Staff Report on Algorithmic Trading in U.S. Capital Markets (market structure and oversight context). SEC
Bank for International Settlements (2025). Working paper on HFT and market outcomes (selected evidence on broader effects). BIS
Reuters (2025). SEBI retail algorithmic trading rules and traceability requirements. Reuters