Financial Firms Replatform AI for Speed and Compliance
Major financial institutions like Citi are replatforming their AI infrastructure to meet dual demands for ultra-low latency and stringent regulatory compliance. Firms are using transformer models for tasks like sub-50 millisecond fraud detection, while also adapting to new SEC and CFTC rules that extend model risk management frameworks to generative AI. This requires engineering leaders to build systems that are not only fast but also auditable and explainable.
- In October 2024, Citi announced a multi-year partnership with Google Cloud to migrate various applications to Google's infrastructure and utilize its Vertex AI platform. This initiative aims to enhance digital products and streamline internal workflows, including using high-performance computing for Citi's Markets business to execute millions of daily computations. - The push for AI adoption is a significant financial investment for firms like Citi, which spent $11.8 billion on technology in 2024 and has retired 2,000 legacy applications over the past three years as part of its modernization efforts. This investment is driven by the need to overcome historical underinvestment in infrastructure and address regulatory actions related to data management. - While the SEC and CFTC have not issued new AI-specific regulations, they have provided guidance emphasizing that existing financial regulations on risk management, supervision, and transparency apply to AI systems. Regulators are particularly focused on the "explainability" of AI models to ensure they can be audited. - Transformer models are being adopted for more than just fraud detection; their ability to analyze sequences of data makes them suitable for modeling the complex relationships in financial markets, including transaction streams and the connections between cards, devices, and merchants. However, their adoption requires a cost-benefit analysis due to increased latency and infrastructure complexity compared to traditional models like XGBoost. - The broader financial industry is significantly increasing its investment in AI, with one survey showing 98% of management planning to boost infrastructure spending in 2025. This is leading to the creation of centralized "AI factories" to process large datasets for developing new models. - AI agents are emerging as a key technology for transforming SRE and DevOps workflows within the financial sector. These agents can autonomously handle tasks like monitoring, incident response, and infrastructure provisioning, which helps to reduce manual errors and improve the efficiency of engineering teams. - The adoption of generative AI in financial services is accelerating, with one survey indicating that 72% of firms are making moderate to large investments in the technology in 2025, a significant increase from 40% the previous year. JPMorgan Chase, for example, has over 300 AI use cases in production and projects $2 billion in business value from AI. - The move to AI-native trading systems is seen as the next industry standard, where AI is not just a feature but the core governance layer of the trading system, influencing everything from execution logic to systemic risk management. This shift is driven by the structural necessity to manage the complexity and speed of modern financial markets, which have surpassed human cognitive limits.
