By Stuart Tarmy, Global Director, Financial Services Industry Solutions, Aerospike
Capital markets are entering a new journey in the adoption of AI. The global agentic AI market is expected to be approximately $7.55 billion in 2025 and could reach $200 billion by 2034. Financial services firms are a large portion of this spending. Since budgets are private, it’s difficult to know how much capital markets firms spend on AI. However, industry estimates suggest they allocate billions annually, with 10–20% of R&D directed toward AI projects and a growing portion going to agentic systems.
Adoption at capital markets firms is already happening. According to a recent survey, 15% of buy-side traders use AI to some extent in their execution workflows, and another 25% expect to have it in use within the year. Many companies are starting to transition from decision-making models to agent-driven, autonomous workflows.
From Predictive AI to Agentic AI
AI adoption typically moves through three stages at most firms, including capital markets firms (also think asset management and brokerage): 1) predictive AI, 2) generative AI (GenAI), and 3) agentic AI. Each has increasing complexity on the technical infrastructure. Below is a quick summary of each stage.
Predictive AI uses statistical analysis and machine learning to analyze historical data, and has been around for 50+ years. Early use cases included fraud detection and credit scoring (e.g., FICO scores). Predictive AI models in capital markets forecast price movements, create execution plans, assess market and credit risks, mitigate opportunities for fraud risk, and provide customized analysis of clients to provide personalized portfolio strategies. They’re typically well-scoped, trained using historical data, and yield results quickly.
Generative AI, which is neural net-based, large language models (LLMs), gained prominence with the introduction of ChatGPT-3 in 2020. It’s ideal for creating new content such as reports, white papers, or answering questions. In capital markets, Gen AI is used to develop research reports and client pitch books, summarize complex information, extract trading signals from unstructured data, and answer customer inquiries.
The new frontier is agentic AI, which uses autonomous software agents to implement AI capabilities in systems without human interaction. Its uses range from automating routine, low-level work to complex, multi-step, human-like activities.
Each phase provides increasing capabilities that add more demands on the enterprise technology infrastructure to perform in real time and with ever-increasing amounts of data. For leaders in capital markets, the question is no longer whether or not to deploy these technologies, but how to best design their architectures to maximize performance at the least cost.
Front Office: Trading and Alpha Generation
Capital markets firms have long relied on complex algorithms to identify opportunities and execute strategies. Agentic AI is taking this even further. For example, Goldman Sachs is deploying an agentic AI system called Devin for its 12,000 developers. By autonomously generating, testing, and refining code, these agents can accelerate the creation of custom trading and risk systems, allowing Goldman to update and optimize algorithms at unprecedented speed. BlackRock has also advanced in this area by implementing AI agents into its Aladdin platform to perform portfolio analytics, risk management, and regulatory oversight, and by installing guardrails to protect against AI hallucinations. Aladdin is Blackrock’s internal trading, risk, and operations platform, used by over 200 leading financial services firms and governments to manage over $21 trillion in assets.
Alpha generation, or the ability to generate investment returns in excess of a benchmark without incurring additional risk, is the holy grail for investment firms. It often involves using advanced quantitative models and AI to analyze large amounts of data to identify new investment opportunities.
With agentic AI, the capabilities once concentrated in top quantitative firms are becoming more accessible. Renaissance Technologies’ Medallion Fund, often cited as the most successful quant hedge fund, has delivered returns exceeding 50% annually before fees. Compare this to mutual funds, where 90% of funds picked more losing stocks than winners. Adopting agentic AI can help a broader set of firms level the playing field by enabling them to develop and run advanced models much more easily and without the same concentration of Ph.D.-level quant talent on their teams.
The potential goes beyond analyzing data to find new trading opportunities. Agentic AI can enable new approaches to securitizing derivative assets (from plain vanilla mortgage-backed securities (MBSs) to more complex fiber asset-backed securities), optimize trade execution pricing by understanding liquidity and order-book dynamics to minimize price slippage, and deliver hyper-personalized investment strategies for corporate and retail clients. These shifts will influence the pace of trading and the products firms bring to market.
Middle Office: Compliance and Risk Management
The financial services industry is the most regulated in the U.S., with numerous regulatory agencies at the federal and state levels (e.g., FINA, SEC, FDIC, and Federal Reserve) and thousands of pages of rules. For example, FINRA’s Annual Regulatory Oversight Report for 2025 is 80 pages long and only provides a summary of its programs. Risk and compliance functions depend on understanding the reams of compliance rules in financial services, coupled with timely, accurate insight across millions of transactions.
Agentic AI can be a game-changer here to help compile and understand the numerous regulations to automate compliance. Agentic AI can operate in real time, identifying anomalies, monitoring compliance, initiating remediation steps, and calling for human intervention if needed. Citigroup has committed to deploying agentic AI across the bank to automate compliance checks, streamline onboarding, and strengthen transaction monitoring. JPMorgan’s NeuroShield pilot demonstrates the impact: In early testing, its agentic fraud detection system reduced fraudulent transactions by 40%. UBS is investing in agentic AI for risk analytics and client advisory, giving financial advisors real-time recommendations on client opportunities, held-away assets, risk exposures, and portfolio adjustments.
Using agentic AI in middle-office use cases will require attention to explainability and auditability. Regulators and clients expect transparency into how decisions are made, even as the systems evolve.
Back Office: Clearing, Settlement, and Reconciliation
Clearing and settlement processes remain costly and inefficient. Firebrand Research reports that the industry has incurred at least $914.7 billion over the past decade in penalties and resolution costs for settlement failures. During 2021 alone, the peak of market volatility, those costs reached $161 billion in equities and fixed-income markets. Agentic AI can analyze pre-trade data to flag at-risk transactions early and intervene before failures occur, which reduces exceptions and resolution overhead.
Reconciliation presents another area for impact. Corporate actions such as stock splits, M&A activity, or special dividends can cause mismatches between order books and cleared trades, requiring large teams to reconcile these discrepancies overnight. It’s estimated that one large broker-dealer has a team of over 100 people doing manual overnight reconciliations before trading begins the following day. Agentic AI systems can automate reconciliations in near real time (due to pre- and post-trade monitoring), surface discrepancies, and trigger corrections as they occur. The result is lower operational risk, faster cycle times, and a better client experience.
Building the Right Technology Foundation
This is an incredibly exciting time in capital markets, but implementing agentic AI doesn’t just happen. It requires firms to modernize their technology architectures. These systems need immediate access to large volumes of real-time data to perform. At the same time, the underlying architecture must handle high concurrency with low latency, enabling hundreds (thousands?) of agents to operate simultaneously without performance degradation. Meeting these requirements requires an ultra-low latency, real-time data platform to ingest, process, and serve information at scale without sacrificing consistency or reliability.
The Competitive Window
Capital markets firms are quickly moving into agentic AI across front, middle, and back offices to improve performance and gain a competitive advantage. To maximize agentic AI’s impact, it’s crucial for capital markets firms to pick the right, ultra-low-latency data platform to architect their systems. The leading firms are modernizing their data infrastructure and operating models today to capture the advantages of agentic AI and leapfrog the competition, while those who delay risk being left behind.
Stuart Tarmy leads global partnerships and financial services industry solutions at Aerospike. He has more than 25 years of experience as a general manager and head of sales, partnerships and product management for leading global financial services technology, capital markets, electronic payments, artificial intelligence, data privacy and regulatory compliance companies. He has held executive roles with Fiserv, Mastercard, Deutsche Bank and McKinsey & Company. Stuart began his career as a computer design engineer at Texas Instruments developing artificial intelligence based computing systems. Stuart holds an MBA from the Yale School of Management, an M.S. in electrical and computer engineering from Duke University, and an Sc.B. with honors in electrical and computer engineering from Brown University.
