Tech investors have been panicking about software stocks lately, and it's not hard to understand why. But one of the people with the most skin in the game isn't worried at all.
Ray Shu leads the technology, media, and telecom lending team at Capital One Commercial. There, he oversees loans to the exact types of businesses investors are running away from right now, such as software-as-a-service and subscription platforms. His team has financed these companies through multiple market ups and downs, claims not to have lost money on a loan in nearly a decade, and notes that the industry's default rate has hovered at a mere 1% for over twenty years.
Shu believes that Wall Street is conflating short-term anxiety over AI with actual business weakness. In his view, because software is deeply embedded in how modern companies operate, it will survive the downturn just fine.
Prior to joining Capital One in 2016, Shu had helped establish financing for established tech and media companies at companies such as GE Capital and Bank of America.
In this interview, he shares the real-time unit economics he believes keep enterprise software resilient, how the collapse of the taxi medallion completely changed the way he evaluates a company's defensible moat, and how to spot software platforms built to survive an AI transition.
The interview has been edited for clarity and brevity.
This Week in Fintech: Equity markets are pricing in long-term structural fragility in SaaS due to AI disruption. But your credit data shows near-zero losses. What metrics are you seeing in enterprise software health, such as net revenue retention or customer churn, that prove the underlying business models are more resilient than public stock charts suggest?
Shu: As a lender, we're looking at this differently than the equity markets. Public markets tend to price in forward-looking AI disruption, but we focus on top-line resiliency, meaning sectors that are cyclical or resilient against downturns. Software is one of the clearest examples. Throughout our nearly 25 years of banking in the TMT sector, the default rate in software has been around 1%, holding across multiple cycles, including the dot-com crash, the telecom correction, and the volatility of the last few years. That track record is what the equity market's forward-looking fear doesn't fully account for.
The core reason is the recurring revenue model. These are largely B2B companies selling annual subscriptions to Fortune 500 and Fortune 1000 enterprises. Customers continue to pay regardless of the macro environment because these tools are core to operations and growth. We are watching how AI might reshape software development and pricing, but our approach is to focus on companies where recurring revenue, customer retention, and deep enterprise integration provide a buffer against that disruption.
From an operating standpoint, we’re consistently seeing enterprise-focused platforms maintain mid-80s% in gross revenue retention, high 90s%+ in net revenue retention, and annual gross churn in the low single digits, particularly in mission-critical systems.
Are you seeing a divergence in health between legacy SaaS providers that are trying to bolt on AI and newer, AI-native platforms? How do their credit risk profiles look different to you right now?
Shu: From a credit perspective, it's less about AI-native versus legacy and more about the durability of the underlying business model. Frankly, there are very few AI-native software companies. Legacy SaaS companies that perform well sell annual subscriptions to large enterprises, with high renewal rates and predictable cash flows. Where platform integration runs deep, AI is largely being adopted as an extension of an existing system rather than a wholesale replacement.
From what we observe, legacy SaaS platforms with strong enterprise exposure continue to show stable renewal trends and expansion within existing accounts, whereas AI-native companies are still in the phase of proving repeatability of revenue and long-term retention metrics.
AI-native companies may have strong growth prospects, but many are early stage, not consistently profitable, and still proving revenue durability. We think about AI the way we've thought about every major technology shift. There are real benefits, but also real costs and unknowns that the market tends to underestimate early on. What we're seeing more of is businesses applying AI to established models, enabling AI solutions across verticals like advertising, marketing, and enterprise software. SaaS companies with strong recurring revenue and deep customer integration continue to present lower credit risk, while AI-native platforms are still proving they can generate consistent, repeatable cash flows.
When underwriting a software loan five years ago, predictable recurring revenue from multi-year contracts was the gold standard. In 2026, with code generation becoming cheaper and software switching costs potentially declining, how has Capital One evolved its definition of a 'defensible moat' when assessing a company's ability to repay debt?
Shu: Five years ago, recurring revenue from long-term contracts was a great option, but today we’re looking beyond just whether revenue recurs to how durable it is. With AI lowering the cost of building software, we’re focused on whether a platform is embedded in customer workflows, how difficult it would be to replace, and whether it benefits from proprietary data and ongoing usage. Recurring revenue is the starting point, but the real question now is whether that revenue is supported by structural advantages, like integration, switching costs, and data, that make it resilient over the loan term.
In terms of underwriting, that translates into lower leverage thresholds, more conservative cash flow assumptions, and tighter alignment between debt service and recurring revenue visibility.
The clearest illustration of what we want to avoid is what happened with taxi medallions. Before ridesharing, a taxi medallion was a valuable, seemingly durable asset. Then a single app completely disintermediated that value almost overnight. We think about that lesson constantly when we're evaluating software businesses. The question isn't just whether revenue recurs today, but whether anything is coming that could disintermediate the value of what that company provides. A company that delivers a single feature or capability is far more exposed to that risk than one whose capabilities are woven into its customers' daily operations.
We've seen some large private credit funds pull back on software exposure, and there's talk of rising default rates across the broader private credit ecosystem. Since your team hasn't taken a credit loss in nearly a decade, what are you doing differently to avoid the landmines that other tech lenders are starting to hit?
Shu: We focus on deep sector knowledge rather than broadly increasing volume. We have always underwritten defensible moats with longer-term barriers to competition, which is why we skew toward vertical players that bring domain expertise, regulatory/compliance needs, or payment integration, versus the horizontal, one-size-fits-all companies that many lenders have focused on thus far. We spend significant time understanding software, infrastructure, and digital media businesses and aligning financing structures to those specific models. We also tend to work with companies that have already scaled and are operationally mature.
Part of what separates us from private credit funds now pulling back is that we never competed with them on their terms. Private credit operates outside the regulatory framework that governs us, allowing them to offer more aggressive structures and flexible terms than we're comfortable with. That said, we don't view them as competitors; we actually find private credit firms to be great partners. Typically, we provide senior secured debt at the top of the capital structure, while private credit assumes mezzanine and subordinated debt. While many in private credit are currently managing software exposure down, similar to real estate office exposure several years ago, we've consistently walked away from deals where the structure didn't meet our standards. The zero-loss track record is partly a function of the deals we didn't do.
Ray Shu, Senior Managing Director & Head of Originations - Technology, Media & Telecom Banking at Capital One
Historically, private equity buyout shops have been massive consumers of software debt to fund acquisitions. If public software valuations remain depressed, does that create a gold rush for PE buyers utilizing your capital, or are lenders being more cautious about financing highly leveraged SaaS buyouts right now?
Shu: Sponsor finance remains at our core, with about two-thirds to 70% of our books being PE-backed. While PE buyers still aim to maximize leverage, lenders are now significantly more cautious about financing highly leveraged SaaS buyouts. This shift is driven by a reset in long-term sustainable growth rates and by the impact of rising interest rates, which have necessitated lower leverage and higher debt pricing than 6–12 months ago.
Furthermore, the competitive landscape has evolved. While the biggest competition for these deals often comes from private credit, those firms are currently trying to reduce their software exposure, similar to the real estate office exposure several years ago. Consequently, we are seeing increasing redemption rates in the private credit sector. This has made the credit environment more selective, with a heightened focus on structure and cash flow alignment.
While we continue to see active demand from sponsors, capital structures are increasingly being built around durable cash flow profiles and realistic growth assumptions rather than aggressive leverage. Our approach remains rooted in deep TMT expertise and tailored financing that aligns with industry dynamics to ensure long-term success.
Because you can see corporate banking data, pipeline health, and drawdowns on credit lines in real time, what is your portfolio telling you about enterprise tech spending for the rest of the year? Are corporations actually cutting software budgets, or are they just shifting dollars around?
Shu: Enterprise IT budgets are growing, particularly in AI, security, and cloud infrastructure, but that growth is more targeted. Within our portfolio, that shows up as continued baseline demand for core software alongside increased scrutiny of usage, vendor rationalization, and tool consolidation. Companies aren't cutting software budgets; they're shifting toward doing more with the same amount, with capital flowing to areas that drive efficiency and automation and away from tools viewed as non-essential.
What we’re seeing across portfolios is continued renewal of core systems alongside increased scrutiny of seat counts, usage levels, and vendor overlap, leading to some contraction in non-essential tools.
In practice, that means sustained demand in software with deep enterprise integration, telecom infrastructure, and data centers, while single-feature tools and point solutions are seeing pressure. The businesses holding up are embedded in core workflows, which is exactly the profile we've been selecting for.
Can you name one specific characteristic of a software company that makes it "fragile" in your eyes today versus one that you consider "protected?"
Shu: The biggest difference is between companies built around a single feature versus those embedded in a customer's workflow. Fragile businesses rely on tools or capabilities that can be quickly copied or replaced, and AI is accelerating that risk by making it cheaper and faster to replicate what used to take years to build.
The protected companies are integrated into customers' day-to-day operations, part of core workflows, and difficult to remove without real disruption. Replacing a fragile tool is inconvenient. Replacing an embedded platform means rebuilding processes, retraining teams, and accepting operational risk. That consequence is what creates durable credit quality.
Another way to think about it is that fragile companies tend to have low switching costs and limited differentiation, while protected businesses benefit from high customer dependency, integrated workflows, and accumulated data that would be difficult to replicate or migrate.
