Is Wall Street ready for AI on-chain loans?

Figure Technology Solutions has spent most of this decade building a blockchain-based infrastructure for consumer credit, positioning it as a faster, cheaper alternative to the fragmented loan systems the mortgage industry relies on. According to the company, it has originated more than US$ 18 billion in loans on its Provenance Blockchain and executed tens of billions more in blockchain-based transactions.

That footprint, paired with the firm’s recent securitization pilots, has drawn growing interest from institutional investors experimenting with tokenized real-world assets. On-chain loans can move between counterparties with fewer intermediaries and lower operational costs. Standardized loan objects promise easier risk assessment and packaging.

Meanwhile, institutional desks are beginning to eye AI-driven credit models to run on loan-level blockchain data, potentially enabling real-time loan valuation.

But the infrastructure underpinning the U.S. mortgage market was never built for this level of transparency or automation. The data problems that undermined earlier modernization efforts remain unresolved. For now, demand for AI-priced, on-chain credit appears to be growing faster than the industry’s ability to produce clean, standardized data.

The AI-Liquidity Thesis

In June 2025, a Figure affiliate closed a US$ 355 million securitization, FIGRE 2025‑HE3, backed by home-equity loans recorded on Provenance. The issuance received ratings from major agencies, marking one of the first rated, blockchain-native securitizations.

For investors, such pilots suggest the possibility that tokenized collateral can meet disclosure, credit-quality, and performance standards required in structured finance. This has helped fuel broader institutional interest. Large asset managers such as BlackRock have publicly projected that tokenized real-world assets could emerge as a multi-trillion-dollar category, if operational and liquidity benefits materialize.

Similarly, industry firms like State Street foresee long-term efficiency gains from blockchain settlement architecture.

The size of the U.S. mortgage market underscores the potential upside. According to the Mortgage Bankers Association (MBA), total single-family mortgage originations are forecast to exceed US$ 2.3 trillion in 2025.

Even a small fraction of that pool moving on-chain, and becoming subject to AI-based loan-level pricing, could fundamentally transform how credit is traded, valued, and risk-managed.

AI underwriting and analytics models are already widely used today. Lenders employ machine-learning systems to classify documents, detect fraud or anomalies, and score borrower risk attributes. Investors and credit desks deploy ML models to estimate prepayment risk, default probabilities, and expected cash-flow variability. For these models, clean, standardized loan-level data matters; tokenized loans on a common blockchain rail could, in theory, provide precisely that, improving model accuracy and enabling continuous valuation.

This is the central thesis driving institutional interest: with standardized, on-chain collateral, loans could begin to behave more like continuously priced financial instruments, not static assets tied to quarterly or annual performance cycles.

The Data Problem No One Wants to Talk About

But the mortgage industry’s data foundation was never built for real-time, high-fidelity analysis. Academic research has documented persistent inconsistencies across servicing systems, origination platforms, and loan transfers, producing gaps, misaligned fields, and outright errors in income data, employment history, collateral valuation, lien status, and more. In the 2000s, faulty loan tapes and documentation irregularities contributed to model failures and widespread mispricing in mortgage-backed securities, and created legal disputes over chain-of-title and ownership transfers.

Tokenization (including blockchain recordation) does not inherently fix these problems. It can only encode and preserve what is provided at intake. If lenders submit incomplete, erroneous, or nonstandard data, the “standardized loan objects” that platforms like Figure promote still reflect those deficiencies. Once embedded on an immutable ledger, those mistakes become hardened and propagated to investors and servicing entities alike.

Industry groups are aware of these structural limitations and have published data standards intended to address them. The non-profit MISMO has developed eMortgage and eNote specifications, plus a comprehensive dataset format (the Mortgage Compliance Dataset, MCD) to harmonize data capture and transfer across originators and servicers. But adoption is uneven. Legacy loan-origination systems, servicing platforms, and mortgage servicers vary widely in whether, and how, they map internal data to MISMO standards.

Even more challenging is the role of AI. Regulatory bodies such as the Consumer Financial Protection Bureau (CFPB) require lenders that use algorithmic decisioning to provide borrowers with clear adverse-action notices and maintain auditability of model outputs, especially in credit decisions. For institutional buyers relying on AI to value on-chain loans, inconsistent data inputs and opaque model logic raise both compliance and investment risks.

Moreover, global watchdogs are sounding alarms. The International Organization of Securities Commissions (IOSCO) recently flagged data quality, disclosure, and transparency as primary obstacles to large-scale adoption of tokenized markets. Their analysis underscores a sobering reality: blockchain alone does not guarantee data integrity or liquidity; those depend on uniform data standards, robust reporting frameworks, and model governance that the mortgage industry is only beginning to adopt.

Figure’s infrastructure appears capable of supporting efficient loan transfer, recordation, and even securitization under current rules. Yet widespread adoption of AI-priced, tokenized credit depends on a far deeper prerequisite: clean, consistent, loan-level data at origination. Until the industry delivers on that front, investor interest will continue to outpace the system’s ability to supply analyzable, high-fidelity inputs.