Summary

• LSEG is combining Nvidia cuDF with Polars to accelerate analysis of historical market information, reporting substantial gains in selected benchmarks.
• Tick History contains approximately 110 trillion rows across more than 100 million instruments and over 30 years of market activity.
• The approach avoids creating another complete copy of the dataset while retaining the permissions and licences governing financial information.

London Stock Exchange Group is working with Nvidia to accelerate analysis of historical financial market information, using graphics processors to reduce selected computational workloads from hours to seconds without requiring another complete copy of the underlying data.

The collaboration combines LSEG’s Tick History service with Nvidia’s cuDF technology and the Polars analytical framework, allowing financial institutions to process large collections of historical trading information using GPU infrastructure.

Benchmark tests have shown years of Tick History information being analysed in seconds rather than hours. The results apply to the workloads tested and do not establish that LSEG’s entire historical database, or every customer query, can be processed within the same period.

Tick History contains approximately 110 trillion rows covering more than 100 million financial instruments and over 30 years of market activity. Its sources include more than 580 exchanges, trading venues and data contributors, providing detailed records used in investment research, market analysis and trading strategy development.

The collaboration forms part of LSEG’s Everywhere AI strategy, which seeks to make licensed financial information available within the computing and analytical environments used by customers while retaining the commercial controls governing that information.

More detailed analysis during the trading day

Financial institutions have traditionally processed some large analytical workloads after markets close, allowing time to load, transform and examine substantial datasets before the following trading session.

That operating model becomes more restrictive as markets move towards extended trading hours and firms seek to evaluate information continuously.

Historical trading records contain large quantities of individual transactions, quotations and changes in market conditions. Reconstructing activity across many instruments and venues requires substantial computing resources, especially when analysts retain the detail of individual market events.

Researchers may examine how a trading strategy would have performed during periods of volatility, investigate the prices achieved when orders were executed or compare liquidity across different market conditions.

Reducing the dataset to broad averages can make calculations faster, but it can also remove details important to understanding how prices moved or how orders interacted with available liquidity.

GPU acceleration provides another approach by distributing suitable calculations across large numbers of processing units. Nvidia’s cuDF technology supports data operations on GPUs, while Polars provides a framework for filtering, transforming, joining and aggregating information.

Using the technologies together can allow analysts to accelerate supported operations without rebuilding every part of an existing data processing workflow.

Performance improvements will nevertheless vary. Some operations benefit substantially from parallel processing, while others remain constrained by storage performance, data transfer, memory capacity or the structure of the calculation.

GPU systems also introduce costs for hardware, cloud capacity and specialist operations. A faster analysis may be commercially attractive when it enables valuable additional work, but speed alone does not establish a lower cost per completed task.

Data permissions remain part of the architecture

Financial market information is governed by licensing agreements specifying which organisations and users can access particular datasets and how the information may be used.

Those restrictions remain relevant when institutions move from conventional analytical systems to AI and accelerated computing platforms.

Creating multiple complete copies of licensed data can complicate governance. Additional datasets may require separate storage, synchronisation, access controls and procedures for ensuring that outdated or unauthorised information is not used.

LSEG says its new approach avoids generating another full copy of Tick History while preserving the licences, permissions and controls already governing access.

Keeping the original data under established controls could simplify the use of GPU processing across different analytical environments, particularly for financial institutions managing several teams and applications.

However, avoiding an additional complete dataset does not mean that information never moves between systems or enters temporary processing memory. The technical configuration still needs to account for data movement, derived results and appropriate protection of information accessed by individual workloads.

Customer organisations must also govern any analytical outputs subsequently used for trading, investment decisions or AI applications.

Models trained or evaluated using historical financial data may reproduce errors, incorporate unsuitable assumptions or identify relationships that do not persist in live market conditions.

Improving the speed of calculation does not automatically improve the quality of the decisions made from those calculations. Reliable testing still requires appropriate methods, comparable datasets and controls over how results are interpreted.

Higher data volumes reshape infrastructure demand

LSEG reports that traffic across its real-time network increased by 29% in the first half of 2026 compared with the equivalent period in 2025, reaching a peak of 26 million messages per second in June.

Customer API requests for Tick History rose by 48% over the same period, indicating growing use of the company’s historical data services.

These measures describe different activities. Real-time network traffic reflects the flow of current market information, while requests for historical data indicate demand for records of previous trading activity.

Neither figure establishes that the entire Tick History archive is processed simultaneously, nor that every additional request requires the same amount of computing capacity.

Nevertheless, increasing use of market data creates pressure to improve processing efficiency and reduce the delays associated with large analytical workloads.

The development also coincides with LSEG’s plans for extended trading. Its proposed LSE 24 venue is intended to support near-continuous trading from Monday to Friday, with client testing targeted for the end of 2026 and an initial launch of exchange-traded products planned for the first half of 2027, subject to regulatory approval.

Longer trading hours reduce the periods during which firms can rely exclusively on overnight processing. Analytical systems may increasingly need to examine historical information while trading operations continue.

Faster access to detailed data could support more extensive strategy testing, execution analysis and market surveillance without requiring every calculation to be scheduled outside normal trading activity.

Financial institutions will still need to examine whether GPU acceleration provides an acceptable return compared with existing processing systems. The relevant measures include infrastructure cost, execution time, data governance, operational reliability and the additional analytical work that faster processing makes possible.

LSEG has not disclosed customer pricing for the accelerated approach or independently verified benchmarks covering every supported workload and deployment environment.

The announced results demonstrate the potential for substantial acceleration in selected analyses of historical information. Wider commercial adoption will depend on whether customers can reproduce useful gains while maintaining the licensing, data protection and operational controls required in financial markets.