The Token Tax: A Research-Driven View on AI and Software Across Equity and Credit

The Token Tax: A Cross-Asset Conversation (Equity and Credit) on AI, Software, and Credit

Jeff Helsing: Tony, let’s start with the basic idea. We keep hearing that AI is a product cycle for software companies. But you have framed it as something bigger—a margin structure change. What do you mean by the “token tax”?

Tony Wang: The token tax is the cost of embedding intelligence into software. Historically, software companies had a very attractive model: build the product once, sell it many times, and enjoy very high incremental margins. AI changes that. Every time a user asks for intelligence, someone has to pay for compute, tokens, memory, inference, storage, and networking. The product may become more valuable, but the cost to serve the customer can also rise.

That is why the market is starting to look beyond AI as a feature race. Investors are asking whether companies can monetize AI faster than AI raises their cost base.

Jeff: So the issue is not that AI is bad for all software necessarily. It is that AI may make the economics more complicated?

Tony:  Exactly. AI may be very good for many SaaS companies over time. While it may displace some, it can improve products, deepen workflows, and increase automation. But the market is asking a harder question now: who captures the economics first?

If a software company has to spend more on R&D, more on infrastructure, and more on AI capabilities just to remain competitive, investors need to see the return. If pricing power is not immediate, then AI spending can show up as lower incremental margins or higher operating expenses. That can pressure valuation multiples even when the underlying business is still growing.

Jeff: Mike, from the credit side, does this same issue show up differently?

Michael Trivino: It does. Equity investors have already seen enterprise valuation multiples compress amid disruption risk and uncertainty. The token tax is another potential headwind for credit investors to incorporate into earnings and cash flow models. Over time, these pressures could contribute to rising restructuring activity in both the public and private lending markets, where software exposure is meaningful.Many of these companies were financed under the assumption that software offered durable growth, strong margins, and predictable cash flow. Those assumptions supported premium enterprise-value-to-EBITDA multiples and, in turn, relatively low loan-to-value ratios at origination. If the token tax raises costs, pressures earnings, or creates uncertainty around pricing power, lenders may demand a higher return. That raises the cost of capital.

This becomes especially important as companies approach upcoming maturities. Investors become more cautious when they are being asked to extend capital into a business model whose earnings profile may be changing.

Jeff: That sounds like the token tax could hit both the equity multiple and the credit spread.

Mike: That is the concern. In credit, the problem is not always immediate default risk. Many software companies will survive. But the equity cushion can thin out. Loan-to-value ratios can rise. Ratings pressure can increase. Refinancing can become more difficult.

Some companies may end up being treated less like high-growth software platforms and more like mature infrastructure businesses—or what we sometimes call “dumb pipes.” Dumb pipes can still be refinanceable, but usually not at the same valuation or cost of capital.

Jeff: Tony, where does that leave the equity opportunity? Is the answer simply to avoid software and own semiconductors?

Tony: I would not say it that broadly. The application layer can still create enormous value. But active investors have to separate companies that can pass through AI costs from companies that absorb them.

The more straightforward opportunity today may be closer to the bottleneck. If every company wants AI, then every AI workflow needs compute. As we move from chatbots to agents, the infrastructure burden grows. Agents do not just answer questions. They reason through steps, call tools, retrieve data, use memory, and keep working through a task. That means more tokens, more memory, more inference, more storage, and more networking.

That is why semiconductors, memory, and AI infrastructure remain structurally advantaged. They sit closer to the constraint point of the system.

Jeff: So for investors, the question is not “Is AI good or bad?” It is “Who pays, who passes it on, and who collects?”

Tony: That is the right framing. Some companies will pay the token tax. Some will pass it through to customers. Some will collect it by selling the infrastructure that everyone else needs.

Mike: And from the credit side, some companies will have the balance sheet to manage the transition, while others may see their cost of capital rise right when they need flexibility.  This will likely create opportunities in public and private markets for new lenders at elevated yields and better underwriting than say 5-6 years ago too.

Jeff: That seems like a strong argument for active management.

Tony: Very much so. Passive exposure may own the AI theme, but it will not necessarily distinguish between companies that benefit from AI and companies whose economics are being challenged by it. Active management can focus on pricing power, margin durability, reinvestment returns, and competitive positioning.

Mike: And credit adds another layer. Two companies can have similar AI narratives but very different levels of cash flow volatility and capital structures. One may have time, liquidity, and flexibility. Another may have maturities, leverage, and less room for error. Those differences matter.

Jeff: Final thought: what should investors watch from here?

Tony: Watch incremental margins, AI pricing models, infrastructure intensity, and whether customers are willing to pay for AI-enabled products.

Mike: Watch refinancing windows, downgrade trends, free cash flow, and whether lenders are getting paid enough for the uncertainty.

Jeff: So the token tax is not just a technology story. It is an equity story, a credit story, and a capital allocation story. AI may create tremendous value, but the market will care a lot about where that value lands. In this environment, owning the theme may not be enough. Investors need to understand the economics underneath it.