On a Sunday evening in early 2026, a quant team at a US Tier 1 bank kicks off the weekly recalibration of its retail credit model and pushes 3.6 petabytes through a Spark cluster before the New York open. According to the IDC Worldwide Big Data and Analytics Spending Guide, US financial-services spending on big data platforms reached roughly 75 billion dollars in 2025, growing in the high teens, with banks, asset managers, and insurers now treating the category as core operational infrastructure rather than a science-project line item. Big data analytics in finance has quietly become one of the largest enterprise software categories in the country, and the dollar figure undersells how deeply it now shapes pricing, risk, and customer experience. The technology stack inside a US Tier 1 bank in 2026 looks more like a software company than a financial institution, with thousands of data engineers, dozens of streaming pipelines, and an increasingly tight integration between the data platform and every product line that touches a customer.
From data warehouses to streaming pipelines
US banks have been collecting transaction data at scale since the 1990s, but the early architecture was a nightly batch job feeding a relational data warehouse. That worked for accounting reporting and regulatory submissions, but it could not handle the volume or latency the modern bank needs. The shift to columnar databases, distributed file systems, and stream processing began in the early 2010s and accelerated as cloud-native platforms made the engineering tractable. By 2024, the median US bank ran a hybrid stack of Hadoop or Snowflake for batch analytics, Kafka or Kinesis for streaming events, and Spark or Flink for transformation. The numbers behind the platforms became hard to ignore: Bank of America publicly disclosed that it processes more than 4 petabytes of data daily, and JPMorgan Chase has said its data lake has grown past 450 petabytes total across its corporate, retail, and asset management businesses. The same architectural shift is on display in recent coverage of machine learning in finance, which now sits as the primary consumer of those pipelines rather than a separate workflow living in a different team.
Where the spend actually goes
The 75 billion figure breaks into five major categories. Risk and analytics, including stress testing, value-at-risk, and capital modelling, takes the largest share at roughly 32 percent. Customer 360 platforms, which unify retail and commercial customer data into a single profile, account for 24 percent. Fraud detection and anti-money-laundering screening absorb 19 percent. Treasury and liquidity analytics, which became increasingly important after the 2023 regional bank crisis, sit at 15 percent. Compliance and regulatory reporting account for the remaining 10 percent. The mix has shifted noticeably over the last three years, with risk and analytics gaining at the expense of compliance reporting as banks reuse the same data infrastructure for multiple workloads.
Where US bank big data spend actually goes (2026)The mix has tilted away from compliance reporting and toward customer-facing use cases over the last three years. Compliance was the entry point for big data in banking; it is no longer the dominant use case. Banks have learned that the same pipeline running their AML screening can power a personalised customer offer, a real-time pricing decision on an FX trade, or an instant credit-limit increase, and they have begun to design the data platform with all of those use cases in mind. The same multi-purpose design appears across other parts of the financial sector, including real-time payment infrastructure, where the same data feeds power both fraud and customer experience.
The cloud question and where the data lives
The hardest architectural question for US banks in 2026 is no longer whether to use cloud data platforms; it is how much of the regulated data can live there. Snowflake, Databricks, BigQuery, and Microsoft Fabric have all built financial-services compliance offerings, with HIPAA-style controls extended to bank data classifications. The OCC and Federal Reserve have signaled comfort with cloud data platforms as long as encryption, access controls, and data lineage meet supervisory expectations. The remaining concern is concentration risk. A handful of cloud data platforms handle a meaningful share of US bank analytics workloads, and the failure or pricing power of any one of them would be felt across the system. Banks have responded with multi-platform strategies that allow them to move workloads between providers, but the operational lift of true platform portability is significant and continues to consume a sizable share of the engineering teams at the largest banks. The same architectural debate is shaping digital transformation in finance, where data platform choices now drive most of the downstream technology decisions a bank makes.
Privacy, lineage, and the regulatory bar
The legal infrastructure around bank data has tightened. State-level privacy laws, including California’s CPRA and a growing wave of similar statutes in Texas, Virginia, Colorado, and other states, place explicit requirements on how customer data is stored, accessed, and deleted. The Consumer Financial Protection Bureau finalised its 1033 open-banking rule in 2024, which gave consumers the right to share or revoke access to their financial data with third parties, and that flow now needs to be tracked at the row level. Data lineage tools, which were a niche product five years ago, are now standard at every Tier 1 bank, with platforms like Collibra and Alation embedded in the data engineering stack at most large US institutions. The regulatory bar on what counts as auditable data lineage continues to rise, with the OCC publishing guidance in 2025 that requires banks to show end-to-end traceability of any data point used in a customer-facing decision. That requirement alone has driven a multi-billion-dollar wave of investment in metadata tooling and data observability platforms, with newer vendors like Monte Carlo, BigEye, and Acceldata winning seats inside bank technology stacks that were dominated by older incumbents only a few years ago.
What 2026 is likely to settle
Three things are likely to change through the rest of the year. First, generative AI workloads will move from the experimentation budget into the production data platform, with banks running fine-tuning and inference next to their core analytics rather than as a separate environment, and the resulting workload sharing is already changing how data engineering and machine learning teams are organised inside US banks. Second, the cloud platform consolidation will continue, with Tier 1 banks narrowing their preferred providers and pushing harder pricing terms as the volume gets large enough to matter. Third, regulators will publish more granular expectations on data governance, particularly around AI-driven decisions, which will force a wave of investment in model documentation and decision logging that is already partially under way. The investment level required to keep pace with these expectations is high enough that mid-tier banks are increasingly buying managed services rather than building their own data platforms, with vendor revenue from that segment growing faster than the overall market. The competitive picture for vendors has also tightened, as the largest US banks now run procurement processes that span months and routinely include multiple proof-of-concept rounds before a contract is signed.
For most of the last decade, big data analytics in finance was the kind of category that lived inside a CIO presentation and stayed there. In 2026 the same category sits at the centre of how US banks make money, manage risk, serve customers, and report to their regulators, and the budget line that funds it now rivals the spend on traditional core banking software, with several institutions reporting more than parity inside their 2026 technology plans. The next argument is no longer about whether to invest. It is about which platforms to pick, how to keep the data portable, how to keep the regulators comfortable with the answer, and how to maintain a sustainable cost base in an environment where data volumes keep doubling roughly every two years.
