#pymntsAI

The shift reflects a broader realization within the industry that the next phase of the AI race will be determined not only by breakthroughs in research but also by access to massive user bases that generate data and adopt new digital services at scale.

In February, the chief executives of Google, Anthropic and OpenAI appeared together in New Delhi alongside Prime Minister Narendra Modi at India AI Impact Summit, underscoring the country’s growing role in the global AI ecosystem.

The symbolism was difficult to miss. As Bloomberg reported, major technology companies are pouring capital into India as they compete for influence in a market increasingly seen as central to the next stage of AI development.

The financial commitments are enormous. Amazon, Google and Microsoft have collectively committed more than $67.5 billion to deepen their AI footprint in India. The Indian government expects the broader AI sector to attract more than $200 billion in investment over the next two years as global companies build infrastructure, partnerships and local ecosystems.

Individual companies are already making massive bets. Google announced a $15 billion data center investment in southeastern India last year, calling it its largest AI hub outside the United States, according to The Wall Street Journal. Microsoft followed with a $17.5 billion investment commitment, its largest ever in Asia, while Amazon has pledged $35 billion across its India operations through 2030.

Scale, Data Are New AI Currency

The surge of investment reflects something AI companies struggle to find elsewhere: large volumes of human-generated data.

Frontier AI model developers rely heavily on human text, images and interactions to train systems. But many existing datasets in Western markets have already been heavily mined. According to Bloomberg, the next wave of training data is likely to come from countries with massive populations adopting digital services for the first time.

India stands out because of its digital scale. The country has more than 700 million smartphone users and among the highest mobile data consumption rates in the world, according to Ericsson data cited by The Wall Street Journal.

That scale generates enormous volumes of digital activity. India already produces roughly 20% of the world’s data while storing only about 3% of it, according to CareEdge Ratings data cited in the Journal.

For Western technology companies, India has an additional strategic advantage. China, the only other market with comparable scale, has largely been closed to Western AI companies for years, making India one of the few remaining sources of massive new training data.

India’s data advantage also goes beyond sheer volume. The country has at least 121 major languages and extraordinary cultural and economic diversity, Bloomberg reported. Training AI systems on that variety of data can make models more robust when deployed across emerging markets.

Adoption rates are already rising quickly. Roughly 62% of Indians now use generative AI tools from at least one provider, as cited by the Journal.

Talent and Developers

India’s technology workforce is another major factor attracting global AI companies.

The country produces roughly 1.5 million engineering graduates each year, creating one of the world’s largest technical talent pipelines. In Stanford University’s Global AI Vibrancy Index, which evaluates talent, infrastructure and governance, India ranks third globally behind only the United States and China.

That workforce has long powered the global technology services industry. Now it is shifting toward AI development, machine learning engineering and data science roles.

Companies are already expanding their local presence to tap into that talent pool. Anthropic, for example, recently opened its first office in Bengaluru and said India has become the second-largest market for its Claude AI assistant after the United States.

Industry experts say the country’s talent base is particularly suited to working with AI systems that operate at large scale.“Strategically, not only is India a huge market, it also has the largest pool of skilled engineers and mathematicians to work on AI,” Saikat Datta, chief executive of DeepStrat, told Livemint.

For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.

In recent weeks, FinTech lenders reported that automated underwriting systems, with AI in the mix, are enabling them to process large numbers of applications and expand loan originations across consumer and small-business credit markets.

Loan Originations Expand Across Digital Platforms

Upstart said its platform processed approximately 456,000 loan transactions during the fourth quarter, an 86% increase year over year, bringing more than 300,000 new borrowers onto the platform. Personal loan originations rose 41% from the prior year, while the company’s newer secured lending products expanded rapidly. Auto and home loan originations each increased roughly fivefold compared with the previous year.

SoFi said it originated $10.5 billion in loans in the fourth quarter alone. Personal loans accounted for $7.5 billion of that total, while student loans reached $1.9 billion and home loans totaled roughly $1.1 billion.

Small-business lending also expanded. Enova reported $2.3 billion in loan originations during the fourth quarter, representing a 32% increase from the prior year. Of that amount, $1.6 billion came from small-business lending, which grew 48% year over year and now represents the majority of the company’s lending portfolio.

The increases illustrate how digital lending platforms have expanded beyond consumer credit products into broader segments of the lending market.

Automation Is Driving Underwriting Decisions

Executives say those volumes are being enabled by increasingly automated underwriting systems.

At Upstart, machine learning models analyze borrower characteristics using large repayment datasets. The company said its underwriting models now incorporate more than 100 million repayment events, allowing the algorithms to refine approval decisions and better distinguish risk levels across applicants, according to commentary on the conference call.

OppFi described a similar trend toward automation. CEO Todd Schwartz said automated underwriting approved nearly four out of five applications during the quarter.

“The auto-approval rate in the fourth quarter was 79%, which allowed more customers to be approved without human interaction and helped increase originations 48% year over year,” Schwartz told analysts during the company earnings call.

These models are designed to speed application processing and to segment borrower risk, pricing loans accordingly.

Credit Metrics Show Mixed Signals

While origination growth has been strong, credit metrics across the sector show a more complex picture.

At SoFi, personal loan credit performance remains relatively stable. Borrowers in that portfolio have a weighted average income of roughly $158,000 and an average FICO score of 746. Charge-offs in the personal loan segment were 2.8% in the fourth quarter, up modestly from the previous quarter but still more than 50 basis points lower than a year earlier.

Enova reported similar stability in several parts of its portfolio. The company said its consolidated net charge-off ratio was 8.3% during the quarter, down 60 basis points from the prior year. Its small-business portfolio showed particularly steady performance, with net charge-offs at 4.6%, while the consumer segment posted a higher but stable 16% charge-off rate.

OppFi reported stronger profitability but acknowledged rising stress in certain loan vintages. Net charge-offs reached 45% of revenue in the fourth quarter, up from 42% in the prior year period. Executives said some of the increase reflected inflation pressures affecting borrowers’ discretionary income and repayment capacity. During the call with analysts, CFO Pamela Johnson said: “One of the benefits of short-duration loans is that loans work through the system relatively quickly. That means that by first quarter 2026, the majority of the higher default rate loans should be reflected in our earnings.”

Credit Access Remains a Key Driver of Demand

Executives also pointed to structural factors behind the growth in digital lending.

Enova cited survey data indicating that nearly three-quarters of small-business owners bypass traditional banks and instead seek financing from alternative lenders. Among those who initially approached banks, 46% reported being denied credit.

That dynamic has helped sustain demand for lending platforms that can process applications quickly and evaluate borrowers outside conventional credit scoring frameworks.

At the same time, lenders continue to emphasize risk discipline. Upstart executives told investors that recent loan vintages generated returns exceeding Treasury yields by more than 600 basis points on average, which the company attributed to improvements in underwriting models.

The Central Question for AI-Underpinned Lending

The earnings disclosures collectively illustrate how AI-driven underwriting platforms have become a meaningful source of credit in consumer and small-business markets.

Yet the data also makes clear that the central issue for these lenders moves beyond scale. The sector is producing large origination volumes, but the durability of those models will ultimately be determined by credit performance over time.

These features “make selling more efficient so sellers can list more items with less effort, and shoppers can find more of what they’re looking for with every search,” Facebook said in a Thursday (March 12) announcement.

One new feature makes it easier for sellers to list items by having the AI create a draft listing, fill in details and suggest a price — all based item images uploaded by the seller.

Another simplifies shipping by enabling sellers to generate shipping labels in a few clicks and use one dashboard to keep track of all shipped orders.

A third new feature facilitates sellers’ replies to buyers’ inquiries by enabling AI auto replies. Sellers can enable, preview and edit these auto replies while creating their listing and then allow it to send auto replies when buyers ask if an item is still available.

The fourth feature announced Thursday uses Meta AI to generate a summary of the seller’s Facebook profile, including the length of time they’ve been on the platform, the number of friends they have, their listing history and their seller ratings. This summary will be placed at the top of the seller’s profile to build trust and transparency with buyers, the announcement said.

It was reported in March 2025 that the Facebook Marketplace resale platform remains a draw for younger consumers even as the social media platform loses ground with them.

At that time, the marketplace had 1.1 billion users in 70 countries, competing with the likes of Craigslist and eBay.

Meta, the owner of Facebook, has been deploying AI tools across its platforms.

The company said Wednesday (March 11) that it launched new AI-powered anti-scam tools for its platforms WhatsApp, Facebook and Messenger. These tools are designed to help users of these platforms spot and avoid scammers.

On March 3, it was reported that Meta is testing a shopping research function on its AI chatbot. This function lets shoppers request product suggestions and then responds with a collection of product images with information about the brand, price and website.

The numbers on both sides reflect the amount at play. UnitedHealth Group projects AI could save it nearly $1 billion in 2026, while HCA Healthcare expects roughly $400 million in AI-driven cost savings, partly from automating revenue management, according to Reuters. On the other side of that ledger, Blue Cross Blue Shield has released an analysis suggesting that AI-enabled coding practices may be responsible for more than $2 billion in additional claims spending nationwide.

Hospitals Use AI to Optimize Billing

Hospitals are turning to AI to automate clinical documentation and medical coding, the process of translating care into standardized billing codes submitted to insurers. These tools use ambient listening technology to capture clinical interactions in real time, then analyze physician notes and lab reports to automatically assign billing codes, a workflow that proponents say reduces paperwork and physician burnout.

But the Blue Cross Blue Shield Association analysis of de-identified claims data found patterns that raise questions about accuracy. Researchers tracked a sharp rise in diagnoses of acute posthemorrhagic anemia at hospitals that had publicly disclosed AI adoption. In many of these cases, patients coded with the condition never received blood transfusions, a treatment typically associated with it. That diagnosis spike alone added $22 million to maternity admission costs in one year.

Looking across all facilities, the analysis attributed about $663 million in inpatient spending and at least $1.67 billion in outpatient spending to AI-powered coding practices.

Federal data shows 7 in 10 U.S. hospitals used predictive AI in 2024, with AI use for billing jumping 25% year over year, according to U.S. News and World Report.

Insurers Use Their Own Algorithms

As hospitals automate revenue capture, insurers are deploying AI to audit claims and deny coverage at scale. The share of provider claims denied more than 10% of the time has risen from 30% three years ago to 41% today, according to Experian. Insurers on Affordable Care Act marketplaces denied nearly 1 in 5 in-network claims in 2023, up from 17% in 2021, according to KFF.

UnitedHealth Group has faced scrutiny from federal lawmakers over its use of algorithms to deny care to seniors enrolled in Medicare Advantage, according to industry news site Stat. Humana and other insurers face lawsuits and regulatory investigations over similar practices, said Revenue Cycle Coding Strategies. The industry argues AI improves efficiency and reduces costs by processing high volumes of claims data that would otherwise require extensive manual review.

Patients are beginning to arm themselves with AI tools as well, according to North Carolina Health News. Startups, including Sheer Health and the nonprofit Counterforce Health, have built tools that help patients analyze denial letters, cross-reference their insurance policies and draft appeals. Historically, fewer than 1% of denied claims are appealed, and patients lose more than half of those appeals.

Consumer AI tools are designed to shift that math, though Carmel Shachar, assistant clinical professor of law and the faculty director of the Health Law and Policy Clinic at Harvard Law School, warned that it can be difficult for a layperson to understand when AI is doing good work and when it is hallucinating or giving something that isn’t quite accurate, according to North Carolina Health News.

Regulation Meets Rapidly Scaling Problem

The speed of AI deployment on both sides of the healthcare billing divide is outpacing regulatory frameworks. The site said more than a dozen states passed laws regulating AI in healthcare in 2025, with Arizona, Maryland, Nebraska and Texas among those banning AI as the sole decision-maker in prior authorization denials. Broader federal standards have not kept pace.

The concern for regulators is not simply that AI speeds up billing disputes. It is that automated systems on both sides risk optimizing for financial outcomes rather than clinical accuracy, with patients caught between competing algorithms.