Keeping Up with the Dow Jones: How Financial Services Firms Can Select the Right Hardware to Power AI Use Cases
A blog by Ben Langstreth, Senior Account Manager at Vespertec
—
Release date: 29 October 2024
Since its foundation, the financial services (FS) industry has made high-tech innovation part of its DNA. From the transatlantic cable in the late 19th century and the Fedwire in the early 20th, to the NASDAQ in the ‘70s. Now, AI is joining this long lineage of tech advancements used to improve the flow of commerce and currency.
We recently ran a survey of FS firms to understand their readiness for AI adoption, and we discovered 84% of respondents are already using AI in their day-to-day operations, with a further 14% planning to do so soon.
Over 55% of respondents acknowledged that their own AI initiatives are spurred by worries that competitors may otherwise gain an edge, with 69% saying they are certain that their competitors are using AI.
Last month, Markus Pelger, Assistant Professor of Management Science and Engineering at Stanford, said in a webinar with NASDAQ leaders that “Machine learning can generate large investment gains if used right.” However, Doug Hamilton, Nasdaq’s Head of AI Research and Engineering, underlined that one-size-fits-all approaches will struggle, with tailored solutions, and hardware, being the key to success.
When asked what was prompting them to allocate budget to AI projects, nearly 40% of firms we surveyed simply said: ‘The board/leadership team have announced a focus on AI’. With this mandate set out by the executive team, infrastructure and compute heads now have the task of choosing exactly which hardware will best achieve these aims.
With this in mind, we’re going to lay out some key use cases, and our top tips on types of hardware best suited to meet the industry’s ever evolving challenges.
Data Analytics and Reporting
A closer look at how FS firms are applying AI reveals a familiar trend: the ability to rapidly process and analyse vast, unstructured datasets is foundational to their competitiveness.
This is hardly shocking, but few other industries deal with the same torrent of raw, multi-source data: ranging from transactional records and real-time market feeds to regulatory compliance datasets and customer behaviour analytics.
In our discussions with industry leaders, the overwhelming majority cited ‘leveraging data for better decision-making’ as the primary motivation behind their AI investments.
No surprises here, then – but it’s the second part of that answer that’s interesting. It’s not just about number crunching. Today’s models have more value to add than a well-organised spreadsheet. Recent projections from KPMG suggest that as soon as 2027, 99% of banks will be using AI to generate automated business reports, crunching large amounts of data and turning it into actionable information.
The real challenge is scaling these processes without introducing latency or bottlenecks. As data sources grow in complexity, firms require hardware capable of handling enormous datasets with real-time analysis. This is where NVIDIA’s H100 Tensor Core GPU, built on the Hopper architecture, shines. Offering up to 80GB of HBM3 memory, the H100 provides a dramatic boost in processing power over its predecessor, the A100. It ensures that firms can manage increasingly complex data loads efficiently, minimising lag and generating faster, more accurate insights.
Fraud Detection / Risk Management
With the Payment Systems Regulator (PSR)’s new rules, designed to combat the rise in Authorised Push Payment (APP) fraud, as of October 2024, UK consumers have new regulatory protections up to the value of £85,000. This means payment service providers (PSPs) have a much greater vested interest in combatting fraudulent transactions.
AI’s data analytics capabilities make it perfect for this task. False positives—where legitimate transactions are flagged as suspicious—can lead to customer attrition and operational inefficiencies, while false negatives—where fraudulent transactions slip through, can lead to PSPs falling foul of the new regulations. That makes this is a real area for technology-assisted growth.
To meet these new demands, financial institutions are doubling down on scalable, high-performance AI infrastructures. In this space, NVIDIA’s GPUs, particularly those optimised for machine learning workloads, have become the backbone of real-time fraud detection systems.
Engines commonly used in fraud detection and risk calculation applications, such as Apache Spark, have been optimised through NVIDIA RAPIDS, enabling seamless GPU-accelerated processing. This allows PSPs to execute fraud detection algorithms in near real-time, maintaining both speed and accuracy at scale, while minimising the risk of AI hallucinations—erroneous outputs that can disrupt fraud detection pipelines.
Algorithmic Trading
Probably the most exciting use case for AI in financial services is algorithmic trading. While ‘algo-trading’ platforms—driven by historical data and pre-set parameters—have long been a cornerstone of high-frequency trading, the integration of more sophisticated AI models has significantly evolved their capabilities. The rise of machine learning (ML) and deep learning techniques has enabled robo-traders to go beyond basic pattern recognition, incorporating a wider array of unstructured and alternative data into their decision-making processes.
Algorithms can now analyse satellite imagery to predict crop yields, observe oil reserves, or track infrastructure developments. This access to revolutionary new sources of alternative information – and crucially, the ability to assess impact on the markets – means algorithmic trading is more attractive and business critical than ever before. According to Bloomberg, algorithms are already handling 75% of trading in spot markets for currencies.
Once again though, FS firms are beginning to run up against a processing barrier. Given the speed at which these systems need to run, experiencing as little latency as possible, firms can’t tolerate inadequate or limited architecture. NVIDIA’s GH200 Grace Hopper superchip, delivering up to 10x higher performance than its nearest rivals, is ideal for running vast datasets in algorithmic trading. It ensures faster data processing and more accurate insights for future trades.
Customer Experience
For years, firms have deployed chatbots to handle customer service queries, with varying degrees of success. At some point in the last decade, I’m sure we’ve all sat head in hands whilst engaging with a bank’s unhelpful customer service chatbot, wishing they’d put you (a) through to a human, or (b) out of your misery.
With modern AI, it doesn’t need to be that way. FS firms are ahead of the game on this, with many already incorporating the technology to improve customer experiences. Ever-improving LLMs can give responses better tailored to customers’ needs – provided they’re run on the right hardware.
NVIDIA ACE, run on the company’s advanced GPUs, provides a digital human service. The suite of technologies can be used in everything from chatbots to gaming, and its Tokkio offering is optimised to quickly learn everything it needs to know about a business and its products and services.
This transition—from rule-based chatbots to AI-driven, digital customer service experts—depends on both advanced AI models and the high-performance hardware required to run them. NVIDIA’s GPU architecture, with its unparalleled parallel processing capabilities, ensures these models operate at scale without latency, delivering a seamless, efficient customer service experience.
The Big Shortcut: How can firms keep up with competitors?
The use cases for AI in financial services are so diverse that it can be difficult to keep up, let alone know what’s right for your firm. 93% of those we surveyed are driven by competitive considerations, saying that they think their competitors are using AI, so they should be too.
They’re right to do so, as we’ve seen with the boom in use cases. However, not all decision makers are convinced they’ve got the infrastructure to get the best out of AI applications. If FS firms are going to commit to boosting their outcomes with AI, they cannot afford to put their heads in the sand about the importance of using cutting edge hardware.
With this in mind, if you’re interested in how NVIDIA’s HGX solutions, including the B200, will operate on new Arm-based cores, check out our Q&A with NVIDIA and Arm about Grace, the new CPU design in that architecture.
