Author: Mario Stefanidis, Research Director at Artemis Analytics; Source: Artemis; Compiled by: Shaw Golden Finance

Introduction

According to data from the International Financial Association (IIF), global debt outstanding reached an all-time high of $348 trillion by the end of 2025. Of this, government debt is about $107 trillion, corporate debt $101 trillion, household debt $65 trillion, and financial-sector debt $76 trillion. The share of digital and financial-technology lending platforms in total debt ranges between $590 billion and $680 billion—equivalent to less than 0.2%.

This largest credit market in human history, still runs today on infrastructure designed decades ago (FICO launched in 1989, MERS went live in 1995). Based on data from the Mortgage Bankers Association in the US, the average origination cost for a single U.S. mortgage is about $11,000. Despite huge technological progress and the widespread adoption of artificial intelligence, this cost is still double what it was in the early 2010s.

Source: Freddie Mac

Clearing and settlement for standard wire transfers still take about 28 hours, while most banks’ credit-approval decisions still rely on a committee process, depending on a black-box scoring model built from 20 to 30 variables. All of this is already public fact, but what’s less obvious is the exact way solutions are getting implemented.

The credit industry isn’t being reshaped by a Silicon Valley-style romantic disruption narrative—no startup can completely replace globally systemically important banks like JPMorgan. Real change is more subtle and more structural: the end-to-end credit workflow that was previously vertically integrated within banks—loan origination, distribution, risk review, capital provisioning, and the underlying infrastructure—has been decomposed into a horizontal, modular architecture, with each component controlled by specialized players.

This architectural shift is exactly like the change in cloud computing from monolithic systems to microservices, and like the media industry shifting from the studio model to streaming and a creator ecosystem. Now, this transformation finally arrives in the credit space.

In this wave of re-integration, the winners are not the biggest institutions by balance sheet size, but core-layer businesses sitting at critical bottlenecks—players that everyone else can’t bypass. There are two positions whose importance far exceeds all others: first, the smart decision layer, where AI-driven risk review and risk scoring determine where money flows and the terms of lending; second, the clearing-and-settlement channel layer, where blockchain infrastructure is massively compressing loan origination costs and settlement time by orders of magnitude.

As long as you occupy these two types of “water-seller” core positions, other lenders will pay you a usage fee. If neither is true, you can only compete on price in a commoditized market—where there is already $3.5 trillion of private credit capital chasing yield.

Here, Artemis maps 40 companies across 15 sub-sectors, classifying them into five major layers to analyze where structural value is concentrating.

Five major layers of the new credit architecture

First layer: Loan origination

The loan-origination layer is the source of credit business, covering categories such as consumer loans, mortgages, small-business loans, and crypto-collateralized loans. This area is also becoming increasingly commoditized. Today, having the ability to originate loans is no longer a competitive moat—it’s merely a basic entry requirement. The key that separates winners from other participants lies in loan origination cost and approval pass rates.

SoFi, with an estimated valuation of about $24 billion, and Rocket Companies (Rocket Mortgage), with a market cap of $48 billion, both have massive loan origination volumes, but the core profit logic is how to fund lending at lower cost. Figure, with a market cap of $6 billion, relies on its Provenance blockchain-native issuance of home equity lines of credit (HELOCs) and first-lien mortgages, removing the multi-layer intermediary steps that make traditional mortgage origination slow and expensive.

In the crypto space, Aave (market cap of $2.7 billion) and MakerDAO/Sky (market cap of $1.6 billion) have completely blurred the boundary between financial technology and decentralized finance (DeFi) in the loan-origination layer.

Second layer: Channel distribution

The distribution layer is where demand is aggregated, and embedded finance and “buy now, pay later” (BNPL) models are reshaping this area. The embedded finance market is expected to grow from $156 billion in 2026 to $454 billion in 2031, a CAGR of 24%. The BNPL model is expected to cover 13% of digital transactions, up sharply from 6% in 2021.

Affirm (market cap $15 billion) and Klarna (market cap $5 billion) are well-known players in the industry, but the real structural trend is that credit services have become deeply embedded into checkout flows, software platforms, and merchants’ consumer experiences. Even though both companies’ stock prices have fallen significantly from historical highs, they are not “water-seller” type businesses that can win major market share. Lenders that borrowers can’t perceive are often the ultimate winners.

Today, major software companies are adding financial products: Shopify, Amazon, Square, and Stripe all need API infrastructure-layer services, and the organizations providing these services will extract fees from each incremental transaction volume.

Third layer: Risk review and risk pricing

This is the first core layer in the entire credit architecture. The institution that controls borrower credit scoring controls how returns are distributed across the whole credit industry.

Currently, the credit bureau and scoring field in the credit world is an oligopoly dominated by three giants: Experian, TransUnion, and Equifax. Combined, the three generate about $18 billion in annual revenue by scoring borrowers using 20–30 variables.

AI risk models can assess more than 1,600 variables (data from Upstart). Upstart’s published data also shows that, while maintaining the same bad-loan rate as traditional models, its approval volume increases by 44%, default rate falls by 53%, and the annualized interest rate (APR) decreases by 36%. With mortgage rates surging to close to 7% today, every basis point matters a great deal for first-time homebuyers.

Upstart currently achieves fully automated loan decisioning for 92% of loans, completing approvals within minutes, whereas traditional risk review takes 3 to 5 days. The U.S. Consumer Financial Protection Bureau (CFPB) is pushing for alternative scoring approaches to FICO that are less discriminatory. The EU’s AI Act also classifies credit scoring as a high-risk scenario, requiring explainability. These regulatory moves are all favorable to explainable machine-learning models, giving them an edge over traditional credit bureaus that rely on black-box models.

The value of this layer is extremely high because whoever owns the scoring engine owns the entire return curve across the chain above it. But at the same time, this domain’s moat still needs to be continuously validated—rapid AI progress means that as long as sufficient resources and time are available, “any institution” can build scoring models.

Fourth layer: Capital and funding supply

In the post-pandemic era, overall capital has been abundant. Although the current environment is challenging, private credit management has expanded to $3.5 trillion, and Morgan Stanley expects it to reach $5 trillion by 2029. The total value locked (TVL) of decentralized finance (DeFi) lending protocols ranges from $5 billion to $78 billion, about half of all DeFi activity. The size of non-transactional perpetual assets (NPE) has grown from zero growth in 2021 to more than $200 billion.

In an environment with abundant capital, the most core capability is smart allocation—directing capital flows efficiently. Therefore, even though the funding layer is huge in sheer size, its structural position still depends on the smart decision layer above and the infrastructure layer below.

Private credit institutions such as Ares, Blue Owl, and Golub are important capital allocators, but they rely heavily on upstream scoring systems and downstream clearing channels to execute lending efficiently. In DeFi, Ape holds an absolute dominant position in liquidity, accounting for more than half of lending volume; meanwhile, protocols such as Maker, Morpho, Maple, and Kamino compete for the remaining market share.

Fifth layer: Infrastructure

Infrastructure is the second core layer in the whole architecture. Whoever controls the financial licenses or the clearing-and-settlement channels everyone has to pay a “toll” to. According to disclosures from management, the bank license held by SoFi reduces its cost of funds by 170 basis points, cutting annualized interest expense by more than $500 million. Figure, leveraging its Provenance blockchain, has processed more than $50 billion in total transaction value; its cost to originate a single loan is under $1,000, while the average cost of traditional channels is about $11,000. Final confirmation on-chain settlement requires only seconds, whereas traditional wire transfers take about 28 hours.

SoFi’s Galileo and Technisys technology stack, along with platforms such as Blend Labs, provide the underlying technology support for remaining lending-as-a-service (LaaS) operations. Cross River Bank, the “invisible” partner bank behind dozens of fintech companies, has already issued more than 96 million loans totaling over $140 billion through partnerships.

Companies that can win long term either occupy a critical bottleneck—becoming indispensable to all participants—or vertically integrate multiple layers to form compound competitive advantages. Companies that lose will get stuck in commoditized business layers, lacking structural bargaining power, and can only compete on price until profits approach zero.

Winners: Core-layer controllers and compound advantage businesses across multiple layers

SoFi: An end-to-end compound toolkit

SoFi is the only company that covers four out of the five layers:

• Directly originates consumer loans and mortgages.

• Through the Galileo platform, exports lending infrastructure to third parties, supporting about 160 million activated accounts.

• Conducts loan underwriting using its own risk-control models, with core evaluation dimensions being repayment willingness, repayment ability, and stability.

• Holds a banking license and, within the infrastructure layer, has the core banking technology systems of Galileo and Technisys.

SoFi set a record for 2025 revenue of $3.6 billion, up 38% year over year. The platform has 13.7 million members and $20.2 billion in financial product volume. Management guidance indicates that 2026 revenue will reach $4.7 billion, with EBITDA of $1.6 billion. This business is not only growing strongly on revenue but also has excellent profitability, with a profit margin of 34%. Just the bank license alone enables SoFi to fund loans through deposits rather than the wholesale market, directly lowering funding costs by 170 basis points.

SoFi is building the “Amazon Web Services (AWS)” of lending—an enabling platform that both competes with other lenders and empowers them. Galileo itself has already been built into a billion-dollar revenue engine. Technisys, acquired in 2022 for $1.1 billion, provides core banking system-layer infrastructure to third parties. A banking license forms a structural moat that most fintech lenders cannot replicate—even as the industry tries to copy: in 2025 alone, the U.S. Office of the Comptroller of the Currency (OCC) received 14 applications to establish new bank licenses, signaling that the contest over the infrastructure layer is accelerating.

Upstart and Pagaya: Smart decision layer

Ironically, winning in the lending industry doesn’t necessarily require doing the lending business yourself. Upstart and Pagaya both center around a risk-review underwriting engine; their risk-control performance is better than the lenders’ own in-house models, and they don’t need to rely on their own balance sheet to run the business. This is exactly where the “water-seller” logic is implemented in the credit decision layer.

Compared with traditional FICO-based risk-control models, Upstart’s model can approve 44% more borrowers at the same bad-loan rate, reduce default rates by 53%, and simultaneously offer borrowers a materially lower annualized interest rate. Currently, almost all newly originated loans on the platform are fully automated, significantly reducing manual intervention. This is fundamentally different from traditional consumer-credit risk-control workflows.

Pagaya is in the same track, but faces tougher market reality. The company does not originate loans directly; instead, it authorizes banks to use its AI risk engine. Since its founding in 2016, Pagaya has evaluated approximately $2.6 trillion in loan applications across 31 partner banks. Its structural positioning is very clear: it doesn’t need borrowers to know its brand—only needs banks to rely on its scoring system. But today, the market has not validated this logic. In Q4 2025, network transaction volume rose only 3% year over year; revenue missed market consensus expectations, forward guidance also came in below expectations, and the stock price plunged by nearly a quarter in a single day. The value of the smart decision layer is entirely constrained by the credit cycle—when bad-loan rates rise across partner networks, even excellent AI can’t withstand pressures from deteriorating asset quality.

But the core logic still holds: FICO builds a single cross-sectional score from a small number of historical variables, whereas as consumers’ financial conditions become increasingly complex and diverse, AI underwriting systems become ever more critical. Unlike FICO, these systems keep learning and improving after each scoring.

Figure: Next-generation clearing-and-settlement channel

With traditional channels and the Mortgage Electronic Registration System (MERS), the cost to originate a single loan is $11,000. But using Figure’s technology stack, which includes the Provenance blockchain and the DART system, that cost can be reduced to $717. These new-channel infrastructure setups enable an order-of-magnitude reduction in lending costs.

Figure has originated more than $21 billion in home equity products (primarily HELOCs) through the Provenance blockchain; cumulatively, it has processed more than $50 billion in transaction volume on-chain. Loan origination volume reached $2.7 billion in Q4 2025, up 131% year over year. The company holds more than 180 lending licenses and has registrations as a U.S. SEC-registered broker-dealer; it has a compliant foundation for scalable operations. It also has more than 300 white-label lending partners. Since submitting its S-1 listing filing last September, it has added partners at a pace of about 1 per day. Its revenue grew from a quarterly annualized $28.5 million in Q1 2023 to today’s $146.8 million.

Figure’s core business isn’t closely tied to crypto assets, yet its stock-price movement is highly similar to Bitcoin. The company’s settlement system reflects the logic of rebuilding the cost structure: final settlement confirmation takes only seconds, while traditional methods take more than a day; loan origination costs are only a fraction of the traditional model. Across the entire lending lifecycle, costs related to asset securitization are saved by more than 100 basis points—within the $3 trillion annual asset securitization market, this implies potential cost reductions of more than $30 billion.

Aave: The core controller in the DeFi space

Aave captures more than half of the DeFi lending market share. Liquidity breeds more liquidity, and borrowers keep clustering toward the platform with the deepest pool (network effects). Its cumulative loan origination volume has already surpassed $1 trillion; last month, the protocol officially crossed the cumulative $1 trillion mark in total loans.

Beyond its dominant position in DeFi, the structurally most interesting part of Aave is its institutional lending business line, Horizon. Horizon has attracted $580 million in deposits, targeting breaking $1 billion by 2026. It serves as a bridge connecting DeFi liquidity with traditional credit demand. If Aave can bring on-chain capital into institutional-grade lending products, it will become the capital-supply layer for traditional lenders—unlocking a potential total addressable market far larger than the retail-focused DeFi market.

DeFi lending also has a structural risk advantage that is often underestimated. Overcollateralization ratios in DeFi are typically in the 150%–180% range, while traditional peer-to-peer lending is only 50%–70%. Bad debts in DeFi mostly come from oracle failures or technical malfunctions, rather than creditworthiness defaults.

Affirm: Channel lock-in

Affirm is a leader in the buy now, pay later (BNPL) space by deeply embedding merchant payment settlement infrastructure. Critics focus on its consumer-credit risk, but ignore the core structural logic: Affirm is not a consumer-lending company in the traditional sense—it’s a credit distribution channel for the sales terminal. Its moat is the system integration with merchants. Given that BNPL is expected to cover 13% of all digital transactions, large-scale embedded checkout platforms will charge a structural “channel fee” from the commercial transactions themselves.

Losing scenarios: Four structural failure modes

We intentionally do not name the companies that fit these patterns. If you’re an investor or operator in credit, you naturally know who they are. More important than specific names is understanding why these structural positions are destined to fail—because in the next cycle, the same patterns will create new victims again.

Loan institutions focused only on the balance sheet

The only competitive advantage of these firms is their ability to obtain funding. They originate loans using traditional risk controls, provide funding from their own balance sheets, and do not have dedicated technology layers. They are merely “dumb pipelines” for capital.

In a world where private credit management has reached $3.5 trillion and is moving toward $5 trillion, capital is not scarce—what’s scarce is smart decisioning and infrastructure. These businesses can only compete on price, compressing profits to zero in each interest-rate cycle and forcing them to take on excessively high risk. In the end, these lenders extend credit to high-risk borrowers and suffer losses when the cycle turns.

These participants are often traditional consumer-lending companies, small-scale banks, and fintech lenders that have never built a technology moat beyond their initial loan products. When capital becomes commoditized, and they lend only with their own balance sheets without technology advantages, it’s essentially a slow transfer of shareholder equity to borrowers.

CeFi lending casualties

The centralized crypto lending (CeFi) platforms that collapsed explosively in 2022 were not victims of the bear market. They fell due to the oldest failure modes in the lending industry: maturity mismatch, misuse of customer funds, lending against illiquid assets without transparent risk management.

Decentralized finance (DeFi) protocols that automatically enforce collateral discipline via smart contracts and make on-chain collateralization ratios publicly visible did not blow up. What really went wrong were those CeFi platforms that relied on human judgment and had non-transparent balance sheets. Any lending platform—whether in crypto or traditional finance—if it only makes you trust its balance sheet but does not show you the collateral, is essentially repeating a structural old path that has already failed before.

Ghost protocols

There’s a type of DeFi lending protocol that is still alive technically, but dead structurally. After going live, they attract initial locked capital through token incentives, but once incentives fade, they stall. The code can still run, and total value locked (TVL) is not zero—but utilization curves level off or continue to decline, with no clear path for organic demand growth.

The reason is that DeFi lending exhibits extreme power-law distribution: liquidity concentrates toward platforms with network effects—Aave dominating the market share is clear proof. Protocols that can’t break through critical scale fall into a structural “no-man’s-land”: too small to attract organic liquidity and supporting integrations, yet not small enough to wind down in a respectable way. As profit-seeking capital flows to the top platforms, their locked value slowly leaks away—an irreversible process. These are zombie protocols that can only barely survive through the sunk cost of governance tokens.

Lenders that missed the platformization transition

Some companies built strong loan origination businesses in the last cycle, but never developed platform capabilities. They have no API distribution channels, no embedded finance partnerships, and no technology-licensing model. Their loan-origination capability is strong, but they can’t output that capability externally.

As the credit industry moves toward modularity, whether you can become a component within someone else’s system is just as important as directly originating loans yourself. Firms that can only lend directly to end borrowers face growth limits tied to the coverage of their own channels; while firms that can provide lending capability support to other institutions have no upper bound to potential market space (TAM). Pure loan originators may have decent unit economics on single customers, but their growth curve flattens because the addressable market is limited to their own brand and channels. In modular architectures, being an excellent lender is a necessary condition, but becoming the kind of lender that other lenders can plug into is the true winning position.

Watch-list assets

The winning companies above have become market consensus or close to consensus, while the following companies are not. They have structural traits that could make them core-layer controllers, but have not yet been validated at scale. These are assets worth continuous tracking.

Morpho

Morpho’s total value locked (TVL) is already $6.6 billion, up 164% year over year, and its market cap exceeds $800 million. Its structural logic is entirely different from Aave: Aave is like a commercial bank in decentralized finance (using a unified lending pool model), while Morpho is building a modular lending layer that allows institutional participants to customize their own lending markets based on their risk parameters, collateral types, and interest-rate models. If the lending system truly moves toward modularity, Morpho will become a lending-as-a-service protocol on-chain.

Maple Finance

In 2025, Maple originated $11.3 billion in loans in total, served 65 active borrowers, and grew its assets under management (AUM) from $516 million to $4.6 billion—a 767% increase. The company aims to reach $100 million in annual recurring revenue (ARR) in 2026. Maple is one of the few protocols truly committed to bringing real-world enterprise lending into blockchain infrastructure, connecting institutional credit demand with on-chain capital and settlement systems to execute business. The explosive growth in its managed assets suggests that institutional interest in the on-chain credit market is shifting from theory and concepts to real-world deployment.

Cross River Bank

Since 2008, Cross River has issued more than 96 million loans through partnerships, totaling over $140 billion. It is the partner bank behind Affirm, Upstart, and dozens of other fintech lenders. Reports suggest the bank is preparing for an IPO. Cross River is an “invisible bank” that supports a significant portion of fintech lending activity as an infrastructure-layer backbone. As the partner-bank model matures, the bargaining power it brings is something no single fintech lender can replicate. The bank’s winning key is to make it so fintech companies can’t do lending without its support.

License battle

The U.S. Office of the Comptroller of the Currency (OCC) received 14 applications to establish new bank licenses in 2025 alone—almost equal to the total of the previous four years. The total number of license applications filed by fintech institutions has hit a historical high of 20. Affirm, Stripe, and Nubank are all actively applying for licenses. These companies view licenses as the core competitive advantage that brings the endgame of rebuilding the credit business.

Companies that started as technology service providers are now capturing the economic value across the full industry chain by obtaining regulatory credentials. In lending, the status of a bank license is comparable to regional nodes in cloud computing, because:

• Building costs are extremely high;

• Industry participants can’t bypass it;

• Once obtained, it creates a permanent structural advantage.

The business logic is straightforward: optimizing the cost of funds by 1 basis point can raise pre-tax net return on equity by several percentage points. For scaled enterprises, the advantages brought by licenses are significant. But for small and medium-sized institutions, a license may become a trap: they must bear all compliance costs, regulatory examination pressure, and capital requirements, yet lack sufficient business scale to cover those expenses. Only companies that already have large business volumes can make a license into a growth accelerator.

Credit architecture in 2030

If there is one core analytical framework to remember from this article, it is the following three questions. They apply to all lending enterprises, whether publicly listed, private, or on-chain institutions.

First: Which layer does the company occupy? Loan origination and homogenous capital supply are in red-ocean tracks, where profit margins keep getting compressed with industry cycles. But AI risk control, blockchain settlement, and banking licenses are in the core bottleneck layers, where value compounds steadily over time. If a company gets trapped in a red-ocean track and can’t enter core layers, no matter how strong the team is, its long-term profitability will be continuously eroded.

Second: Is it building a platform or a single product? A single product serving end borrowers scales linearly with the coverage of its own channels; a platform empowers other lenders, and its growth depends on the size of the entire ecosystem rather than only its own business. SoFi has both attributes, while Pagaya is a pure platform-type enterprise. For companies that lend only directly to their own customers, growth has a ceiling; platform-type companies face no such limitation.

Third: Does it have a regulatory moat? This includes bank licenses, lending licenses across 180 jurisdictions, or programmatic compliance achieved through smart contracts. In the lending industry, regulation is not an extra cost—it is core infrastructure. Enterprises that recognize this early will build advantages that competitors would need years and huge capital to catch up with.

By 2030, the credit industry will no longer resemble traditional banking so much as the cloud computing industry. A small number of full-stack platforms will cover multiple layers and form compounding advantages across each stage. The most typical example in traditional finance is SoFi, and in the on-chain space it is Aave. Around these core platforms, many specialized layer service providers will connect through APIs and on-chain channels, each deeply focusing on specific functions and charging service fees.

In the global $348 trillion debt market, fintech penetration is still under 0.2%. This market isn’t meant to be divided up by hundreds or thousands of lenders; it will be led by a dozen-plus platforms, becoming the underlying support for the entire industry.