U.S. Artificial Intelligence in Fintech Market

Upstart originated 558,014 loans worth USD 4.2 billion in the quarter to June 2026, up 50%, and 91% of them were approved with no human involvement, the company said in its second-quarter results of 4 August 2026; Pagaya’s network volume rose 33% to USD 3.5 billion in the same quarter. American lending decisions are moving to machines faster than any other part of finance. The U.S. artificial intelligence (AI) in fintech market covers the machine-learning and generative AI software, platforms and decision-network fees that U.S. banks, credit unions, insurers, broker-dealers, payment firms and fintech lenders pay for to score credit, stop fraud and money laundering, serve customers, monitor risk and compliance, advise investors and automate back offices. Douglas Insights sizes it at USD 6.85 billion in 2025, from about 15,400 buying institutions at an average USD 444,800 of AI spend each, and at USD 28.5 billion in 2035, a revenue CAGR (compound annual growth rate) of 15.30%. The United States is 90.6% of the North America row of our worldwide Artificial Intelligence in Fintech Market report, the study sits in our information technology and semiconductors coverage, and every figure follows the Douglas Insights research methodology.

Which AI uses earn the most at U.S. banks and lenders: underwriting, fraud or assistants?

Credit scoring and underwriting earns the most U.S. AI in fintech spend, 31.0% of 2025, USD 2.12 billion, because AI networks charge per approved loan. Conversational AI grows fastest, 18.2% a year, to USD 5.10 billion.

Underwriting leads at USD 2.12 billion, above its 24.1% share in our worldwide model, because the United States is where AI lending networks were born: Upstart and Pagaya sell decisions to banks and credit unions for a fee on each funded loan, and U.S. consumer credit files are deep enough to train on. It grows 16.0% a year to USD 9.37 billion. Fraud and anti-money laundering (AML) models, USD 1.64 billion, grow 13.6% a year because every card payment and instant transfer already passes through a score, so growth comes from new rails rather than new adopters.

Customer service and conversational AI, USD 959 million, grows fastest at 18.2% because generative assistants now handle work that call centres did; Bank of America’s Erica alone is used more than 2 million times a day, the bank’s fact sheet of 24 February 2026 says. Risk, compliance and regtech, USD 856 million, grows 14.2% on sanctions screening and model monitoring; trading and wealth advice, USD 719 million, 13.0%, the slowest, because large brokers build in-house; back-office automation, USD 548 million, 16.2% as document-reading agents replace manual loan processing. Application values for 2035 sum to USD 28.5 billion, 0.12% above the headline. How payment rails themselves are priced is covered in our Payment as a Service Market report.

What drives U.S. AI in fintech spend per institution up 16.0% a year to 2035?

U.S. AI in fintech spend per buying institution rises 16.0% a year, from USD 444,800 in 2025 to about USD 1.96 million in 2035, because automated lending, generative assistants and fraud scoring add 19.2 points before restraints take 3.2.

The number of buyers falls 0.60% a year, from about 15,400 to 14,500, because banks and credit unions keep merging. Four drivers make up the gross spend leg.

Automated lending: 7.0 points

AI underwriting adds 7.0 points. Upstart’s originations rose 50% in the June 2026 quarter while the share of fully automated loans held at 91%, and the company kept its 2026 guidance of about USD 1.4 billion of revenue. Pagaya, whose network links 36 lending partners, raised its full-year 2026 net income guidance to USD 155 to 180 million in its second-quarter release of 30 July 2026. Every bank that joins such a network converts lending staff cost into a per-loan AI fee.

Generative assistants and agents: 5.6 points

Assistants add 5.6 points. Bank of America spends about USD 13.5 billion a year on technology, more than USD 4 billion of it on new initiatives including AI, and says 94% of its employees use Erica for Employees, its fact sheet records. Mid-sized banks now license the same capability from vendors rather than building it.

Fraud scoring on instant payments: 3.8 points

Fraud models add 3.8 points. Instant payments leave no time for manual review, so U.S. banks score every transfer in milliseconds; Visa bought Featurespace and Mastercard bought Recorded Future in December 2024 to sell those scores across their networks, the Visa announcement notes.

Compliance and model monitoring: 2.8 points

Compliance tools add 2.8 points as state AI laws require notices and explanations for automated decisions. The four contributions, 7.0 + 5.6 + 3.8 + 2.8, give 19.2 points.

What restrains U.S. AI in fintech: bank consolidation, state AI laws or vendor squeeze?

Three restraints remove 3.2 points from U.S. AI in fintech spend per institution, leaving 16.0% a year: legal uncertainty over state AI laws takes 1.4 points, credit-cycle risk 1.0 point and vendor price competition 0.8 points.

Legal uncertainty is the largest brake. Colorado replaced its 2024 AI Act with S.B. 26-189, signed on 14 May 2026, which from 1 January 2027 requires lenders to notify consumers before an automated decision and explain an adverse one within 30 days, Ropes & Gray summarised; at the same time a federal order seeks to challenge such laws. Banks wait for clarity before buying new decision models, so Douglas Insights removes 1.4 points.

Credit-cycle risk removes 1.0 point. AI lending fees depend on loan volumes; Upstart’s conversion rate slipped to 19.7% from 21.0% a year earlier, a reminder that funding and credit appetite can tighten. Vendor competition removes 0.8 points: nCino’s subscription revenue grew 10% to USD 143.5 million in the quarter to July 2026, its results of 25 August 2026 show, slower than AI spend overall, as cloud and model providers bundle AI into tools banks already pay for. Consolidation sits in the buyer leg: FDIC-insured institutions fell by 41 to 4,238 in the second quarter of 2026 alone. Financing needs of smaller firms that AI lenders serve are sized in our Unsecured Business Loan Market report.

Where in the United States is AI in fintech bought, and why do Texas and Florida grow fastest?

The Midwest buys the most U.S. AI in fintech value, USD 1.16 billion or 17.0% of 2025, and New York leads the single states at USD 1.10 billion. Texas grows fastest, 16.6% a year, with Florida at 16.4%.

AI spend follows where financial firms are headquartered, not where customers live. New York, USD 1.10 billion, hosts the largest banks, brokers and insurers and grows 15.0% a year. California, USD 891 million, grows 16.2% because San Mateo-based Upstart and most AI-native lenders and payment start-ups sit there. Texas, USD 548 million, grows fastest at 16.6% as banks and fintechs move operations to Dallas and Austin. North Carolina, USD 411 million, is anchored by Charlotte, home of Bank of America and Truist, and grows 15.6%; Florida, USD 343 million, grows 16.4% on payment firms in Miami and Jacksonville.

The Midwest, USD 1.16 billion, holds the largest regional share because it has the most community banks and credit unions, but grows only 14.4% as those institutions merge fastest. The rest of the Northeast, USD 822 million, grows 14.6% on Boston asset managers and Connecticut insurers; the rest of the South, USD 925 million, 15.6%; the rest of the West, USD 651 million, 15.4% on Utah’s industrial banks. State values for 2035 sum to USD 28.6 billion, 0.66% above the headline; in the data table the row labelled Global holds the U.S. total, USD 6.85 billion in 2025.

Who leads U.S. AI in fintech: Upstart, Pagaya, the card networks or FICO?

Upstart holds the largest share of U.S. AI in fintech spend, about 10.5% of 2025, USD 719 million, Douglas Insights estimates. With Pagaya, Visa and Mastercard the top four hold about 26.5%; the rest is fragmented across hundreds of vendors.

Shares are Douglas Insights estimates because no vendor reports U.S. AI revenue on its own. For Upstart we count about 76% of its USD 950 million of 2025 fee revenue, excluding servicing fees; for Pagaya about 45% of USD 1.30 billion of 2025 revenue, excluding interest and capital-markets income. FICO’s Scores revenue rose 41% to USD 458.9 million in the June 2026 quarter while its software revenue rose only 2% to USD 215.3 million, its filing of 29 July 2026 shows; we count only the AI decisioning share of that software, not scores.

The card networks are the swing. Visa and Mastercard bundle AI fraud scores into network fees, which makes them hard to displace and lets them grow as payments grow. Upstart and Pagaya compete on approval rates; nCino and FICO sell into banks’ existing workflows. The databases these models read from are sized in our U.S. Relational Databases Software Market report.

How much does a U.S. financial institution spend on AI, and what does an AI-approved loan cost?

A U.S. financial institution spent an average USD 444,800 on AI in fintech tools in 2025, Douglas Insights estimates, from USD 100,000 for a credit union to USD 1.02 million for a fintech lender.

The bank average hides a wide range: a community bank may pay tens of thousands of dollars for a fraud or chatbot module, while Bank of America’s new-initiative budget alone exceeds USD 4 billion. Per-decision pricing is clearer in lending. Upstart’s USD 348 million of fee revenue on 558,014 loans in the June 2026 quarter works out at USD 623.60 a loan, or 8.29% of the USD 4.2 billion originated; Pagaya’s USD 387.0 million of revenue and other income equalled 10.9% of its USD 3.54 billion network volume. Average AI spend per U.S. institution reaches about USD 516,000 in 2026 and USD 1.96 million in 2035.

Which U.S. rules shape AI in fintech: the Colorado AI Act, the December 2025 executive order or Section 1033?

Three policies shape U.S. AI in fintech: Colorado’s rewritten AI Act effective 1 January 2027, the 11 December 2025 executive order that targets state AI laws, and the enjoined Consumer Financial Protection Bureau (CFPB) Section 1033 open-banking rule.

Colorado’s S.B. 26-189 lists financial and lending services as a covered domain: deployers must notify consumers before automated decision technology materially influences a lending decision, explain an adverse outcome within 30 days and, where commercially reasonable, offer human review; the state Attorney General enforces it and there is no private right of action. The executive order “Ensuring a National Policy Framework for Artificial Intelligence” created a Department of Justice AI Litigation Task Force from 10 January 2026, asked Commerce to list burdensome state laws by 11 March 2026 and named the Colorado law directly, Paul Hastings summarised.

Data access is the third lever. The CFPB’s 2024 rule giving consumers the right to share bank data with third parties was enjoined by the U.S. District Court for the Eastern District of Kentucky, so its first compliance date of 1 April 2026 passed without enforcement while the Bureau reconsiders it after an August 2025 advance notice, Cozen O’Connor reported. AI underwriters that rely on cash-flow data from bank accounts face slower access until that is settled.

What if credit tightens or AI agents run lending: where does U.S. AI in fintech land in 2035?

The U.S. AI in fintech market ends 2035 between USD 19.7 billion and USD 39.0 billion across our three cases, around a base of USD 28.5 billion. Credit cycles and state AI rules set the spread.

In the slower case a recession cuts loan volumes, more states pass Colorado-style rules and bank mergers accelerate. In the faster case federal preemption settles the legal question, AI agents run most of the loan file from application to closing and credit unions adopt AI networks at scale. One extra point of yearly spend-per-buyer growth adds USD 2.55 billion to 2035 U.S. AI in fintech value; one point less removes USD 2.36 billion.

Our 15.30% sits just below the 15.76% North America rate and the 17.24% worldwide rate in our global model, because U.S. buyer numbers shrink while Asia Pacific and Latin America still add new digital lenders.

Douglas Exclusive: the U.S. AI Lending Yield Ledger

The U.S. AI Lending Yield Ledger is a Douglas Insights model, not an official register, built from 16 figures in 7 company and regulator documents dated 2026, tracking what U.S. lenders pay AI networks per loan.

The finding, a Douglas Insights estimate: the two disclosed U.S. AI lending networks routed about USD 7.74 billion of loans in the June 2026 quarter, an annual pace near USD 31 billion, at a combined revenue yield of about 9.5%. If that yield compresses to about 7% by 2035 as more banks compete for the same borrowers, credit scoring and underwriting still reaches USD 9.37 billion because AI-routed volume grows faster than yield falls. The money a bank once spent on loan officers moves into per-decision fees, and the buyer that wins is the one with the cheapest funding, not the best model.

How does the methodology turn 15,400 U.S. institutions into USD 6.85 billion of AI spend?

Our methodology multiplies about 15,400 U.S. financial institutions buying AI tools in 2025 by an average USD 444,800 of annual AI spend to reach USD 6.85 billion. Spend comes from buyer-type averages checked against vendor revenue.

The model covers 9 states or regions, 6 applications, 5 buyer types and 2 deployment modes, and uses 96 data points, 21 sourced and 75 modelled. Arithmetic: 15,400 x USD 444,800 = USD 6.85 billion; 6.85 x (0.9940 x 1.1600)^10 = USD 28.5 billion. Buyers reach about 15,300 in 2026 and 14,500 in 2035; spend per buyer reaches USD 516,000 in 2026. Bank and credit union counts come from the FDIC Quarterly Banking Profile and the NCUA quarterly data summary.

Two cross-checks agree. First, USD 6.85 billion is 90.6% of the USD 7.56 billion North America row of our worldwide AI in fintech model. Second, the six named vendors’ estimated USD 2.23 billion equals 32.5% of the total, consistent with a market where card networks and AI lenders are large but most spend goes to many smaller vendors and in-house builds. AI tools help draft the text and automated checks test the arithmetic and links; the Douglas Insights research desk chooses the sources, builds the model and approves publication, as our editorial standards explain.

Sources

1. Upstart Holdings Upstart announces second quarter 2026 results (2026)
2. Pagaya Technologies Pagaya reports second quarter 2026 results (2026)
3. FICO, filed with the SEC FICO third quarter fiscal 2026 results (Exhibit 99.1) (2026)
4. nCino nCino reports second quarter fiscal year 2027 financial results (2026)
5. Bank of America Bank of America fast facts (2026)
6. NCUA Quarterly credit union data summary, 2026 Q2 (2026)
7. ABA Banking Journal, reporting FDIC data Quarterly Banking Profile: banking net income USD 90.1 billion in Q2 2026 (2026)
8. Ropes & Gray Colorado scales back AI law (S.B. 26-189) (2026)
9. Paul Hastings President Trump signs executive order challenging state AI laws (2025)
10. Cozen O'Connor Section 1033 compliance date: open banking rule enjoined and under reconsideration (2026)
11. Visa Visa completes acquisition of Featurespace (2024)

What counts in the U.S. AI in fintech market, and what is excluded?

The U.S. AI in fintech market counts AI software, platform subscriptions, decision-network fees and AI services that U.S. financial institutions buy, valued at vendor revenue. Bank staff costs, general cloud computing, credit bureau data and loans themselves are excluded.

The by-application split covers the six applications above; the by-buyer split covers banks, credit unions, insurers, broker-dealers and wealth firms, and payment firms and fintech lenders; the by-deployment split covers cloud and on-premises; the by-state split covers New York, California, Texas, North Carolina, Florida, the rest of the Northeast, the Midwest, the rest of the South and the rest of the West. Value is counted where the buying institution is headquartered.