Flow Capital’s Managing Director, Josh Axler, joined Horizon Advisors’ podcast to discuss what AI is changing about SaaS funding in 2026. This post pulls together the key points from that conversation.
For a decade, lenders used three numbers to decide if a SaaS company was worth backing: 90% gross retention, 110% net revenue retention, and 40% year-over-year growth. Hit all three and term sheets appeared. Those numbers still matter, but they don't carry the same weight on their own. AI is changing how software gets built and how defensible it is, and lenders are asking founders a harder second question: not just whether the metrics are strong, but how durable they are.
For a long time, software companies defended themselves in a familiar way: the product hooked into a customer's stack, and moving off it cost more than staying. AI is pulling that logic apart.
Andreessen Horowitz published a piece arguing that AI agents could eventually handle migrations themselves, which would weaken that lock-in. In their view, brand may become more important as a moat because agents still report to humans, and humans still need someone to trust. Josh’s view on that was directionally aligned, but more measured. He said we’re still very early when it comes to the human-in-the-loop side of working with agents, getting them set up, and trusting them with company data. The capability is coming, but it’s not here yet.
The deeper shift, in his view, is not just about switching costs. It is about what made software defensible in the first place. Software has been secured by the fact that writing code was hard. You needed top talent, and small companies struggled to attract that talent away from the largest players. AI is pulling that defence apart. People across an organization can now do more of the early building work themselves. CEOs can communicate more clearly with engineering teams, mock up ideas, and compress parts of the product workflow that used to take much longer. Things move faster, and companies don't need to budget for as much engineering talent. The flip side: you still need top talent and great people when it comes to turning ideas into execution.
For lenders looking at SaaS companies in 2026, strong retention numbers now invite a follow-up question: how exposed is that retention to a new AI-native competitor, or to customers who start building what you offer into their own stack?
The sharpest question being asked of SaaS founders today is whether they are one GPT release away from being irrelevant. It is no longer enough to point to growth and retention alone. Founders also have to explain whether what they have built is a real workflow customers depend on, or something that could be absorbed into the model layer quickly.
For companies built on top of foundation models, the cost question runs alongside the product question. Wrapper businesses built on top of the major LLMs are being subsidised by those LLMs today, because the model providers themselves still seem to have unlimited access to capital. Nobody knows how durable that cost structure is. What looks viable today could look very different in 6, 12, or 24 months.
That uncertainty shows up directly in the P&L, where many management teams still do not have a clear answer for where AI costs belong. Does it sit in COGS? As a fixed cost? As a variable line tied to token usage? Founders often can’t answer the question because the model providers themselves haven’t priced for the long term yet. Companies built around today’s cost structure are exposed to a shift nobody can time.
So, the question keeps coming back: is the product a feature, or a real business? A real workflow that customers rely on and can't easily replicate has staying power. A thin layer on top of a model is one release away from being absorbed.
The framework hasn't changed as much as the headlines suggest. The same criteria is still relevant. What changed is how hard lenders stress-test each one.
Three things matter most:
Hit all three and venture debt is worth a conversation. Miss on unit economics and top-line growth won't fix the answer. The bar sits this high because leverage cuts both ways. Debt accelerates a strong company, but it also puts pressure on a business whose systems are not ready for it. Covenants track performance after the fact, but the pressure of debt starts much earlier than that.
The AI shift doesn't only change how lenders underwrite, it changes what founders should fund and how, especially in the current market. AI is reshaping what equity investors want to fund, and plenty of healthy SaaS businesses are getting caught in the gap: strong enough to keep growing, not the zero-to-$100M-ARR story that commands today's premium valuations. Raising equity in that environment means accepting worse terms than the same company could have got a few years ago. Venture debt fills the gap.
Venture debt sits between a bank and equity, and it fits a specific kind of founder: one who'd rather keep their company than dilute it. Many of these businesses are growing 40-100% year-over-year, a strong metric by any reasonable standard, but it doesn't excite venture investors chasing the next zero-to-$100M-ARR AI-native story.
If these founders go out to raise equity, they do it at a lower valuation and give up more of the company to get the round done. Banks don't fill the gap either, because banks don't lend to a company still investing ahead of profitability. Venture debt funds the same growth and leaves the cap table alone, provided the business has the retention, go-to-market maturity, and financial discipline to carry leverage.
A common mistake is to treat venture debt like another round of venture equity. It isn't.
Equity investors underwrite the upside story. A debt provider underwrites the downside: can this company keep generating the revenue needed to service the loan in a market where AI is reshaping competitive dynamics? That's why the bar on systems, forecasting, and unit economics is higher than many founders expect when they first reach out.
What the debt structure buys the founder is optionality, which runs in both directions. During the loan, the founder keeps ownership and gets to invest in growth, whether that means building AI into the product, defending retention against new competitors, or extending runway until the market rerates. At repayment, the company usually has real choices: a bank refinancing at a lower cost of capital, the next equity round on stronger terms, growth capital from private equity or a strategic, or a full M&A event.
That optionality is also where the "$20M exit beats $200M" framing from the episode title comes from. Smaller exits aren't inherently better and exit value alone doesn't tell the founder's story. A founder who gives up too much ownership along the way can walk away with less from a $200M exit than one who built with less dilution and kept more of a smaller outcome. The AI layer is changing what's fundable at the top of the market. For everyone else, how the company gets funded along the way shapes what's left at the end.
The rules of SaaS funding haven't broken. Companies that qualify for venture debt in 2026 look a lot like the ones that qualified in 2020: strong retention, a clear go-to-market, and financial management that shows the founder understands their own business. The extra work happens before the term sheet. Lenders now spend more time asking how durable that retention is against the AI layer in the company's category, and whether the cost structure in the forecast reflects where the industry is heading.
For founders, the takeaway is narrower. The AI shift has made equity more expensive to raise for most SaaS businesses, and more expensive to give up once raised. Capital structure is now part of the strategy, not a paperwork exercise at the end. The founders who come out ahead aren't the ones who raised the most. They're the ones who raised what they needed, on terms that let them keep the outcome.
Q: What metrics does a company need to qualify for venture debt? Most venture debt lenders want to see multiple millions in ARR, gross retention around 90%, net revenue retention above 110%, and growth of at least 40% year-over-year, with a path to cash-flow positive within the loan term. Underneath the headline metrics, lenders dig into capital efficiency (sales and marketing spend per new customer), unit economics (CAC, payback period, contribution margin), and forecast quality. Growth assumptions have to match sales capacity and a hiring plan that can be executed.
Q: How is venture debt different from a bank loan or equity? Venture debt sits between the two. Banks usually won't lend to a company that's still cash-flow negative. Equity dilutes the founder. Venture debt lenders look at revenue quality, retention, and growth trajectory, and structure repayment over 24-36 months against the company's growth plan rather than current cash generation. The lender is underwriting the trajectory.
Q: Is a $20M exit really comparable to a $200M exit for founders? It depends on cumulative dilution and deal terms, mostly liquidation preferences. Equity is the most expensive form of capital because the cost shows up at exit. Give up 20% at one round, then dilute again across subsequent rounds, and a much bigger headline exit can produce less for the founder than a smaller one with low dilution. The point is to run the scenarios before the round, not after.
Q: How do AI costs affect venture debt eligibility? Where AI inference and model costs land in the P&L is an open question for most companies built on foundation models. Lenders want to see that founders have thought about it, can explain where the costs sit today, and have a view on what happens if model providers change their pricing. The model providers are subsidising wrappers right now. That subsidy isn't permanent.
