Quality Thesis

At first glance, the AI boom ought to be a windfall for America’s software champions. More computing, more automation, more digital work should mean more demand for the tools that run modern business. Yet in boardrooms and on earnings calls, a less comfortable story is emerging: AI is not merely a feature upgrade. It is a new competitor—sometimes a substitute—for the very idea of packaged software.

Investors have begun to treat the sector as if it’s in a holding pattern, waiting for proof that incumbents can defend their economics while AI-native rivals sprint ahead. In one widely circulated set of investor takeaways, the mood is blunt: application software faces fresh competition, and infrastructure players must justify a surge in AI-related capital spending with a commensurate revenue payoff. Confidence, the note argues, may require several quarters of stable fundamentals—and even then, the bear case could simply be deferred rather than defeated.

When “software” becomes a layer, not a product

The core threat isn’t that enterprises will stop buying software. It’s that they may buy less of it, pay less for it, and demand that it behaves more like a flexible capability than a fixed suite of modules. AI changes what users value. If a task can be completed through a conversational interface—draft the proposal, reconcile the invoice, build the dashboard—the premium once attached to mastering a complex workflow starts to erode. That dynamic has already haunted creative tools, where AI-assisted design narrows the gap between amateur output and professional-grade work, and the pricing leverage of incumbents can weaken even if the most demanding power users remain loyal.

In enterprise software, the shift is even more structural. Large SaaS vendors have long benefited from being “systems of record”—central repositories of customer, employee, finance, or operational data. AI doesn’t negate that value, but it can siphon off the interface layer where differentiation and pricing power often lived. If the user’s primary interaction becomes an “agent,” then the winner is whoever owns the orchestration: the layer that routes requests across apps, enforces permissions, maintains governance, and delivers reliable outputs in the context of a company’s data and workflows.

That framing helps explain why the most unsettling competitors for U.S. software firms are not always the obvious ones. It isn’t just the next SaaS startup with a narrow feature set. It’s also frontier model providers and AI-native apps that can underprice incumbents because their cost structures are more dynamic and because their profit pools are still largely incremental. Meanwhile, incumbents still have advantages—installed bases, distribution, enterprise relationships—but those moats look different when innovation cycles compress from years to months.

The build-versus-buy debate returns—with a twist

For decades, software vendors have relied on a familiar argument: building enterprise-grade systems in-house is expensive, risky, and distracting. AI revives that debate in a more tempting form. With coding copilots, agent frameworks, and cloud primitives, companies can assemble bespoke solutions faster than in prior eras. The allure is strongest where the “workflow” feels unique—legal review, specialized customer service, industry-specific compliance—and where leaders believe AI can turn proprietary processes into an advantage.

But the twist is that AI also raises the penalty for getting it wrong. Governance, security, auditability, and data controls become the bottleneck. The investor note highlights a key potential inflection: as the packaged ecosystem matures, customers may shift away from bespoke builds toward standardized offerings precisely because enterprise-grade controls matter more in an agentic world—and because DIY projects can create new security and compliance headaches.

That creates an opening for established software firms—if they can productize AI in a way that feels safer, simpler, and measurably valuable. The problem is that many vendors are still in the messy middle of transformation: migrating architectures, adding semantic layers, integrating vector search, and retraining salesforces to sell outcomes rather than seats. In other words, “SaaS plus AI” is not a patch; it’s a rebuild. Some firms are demonstrably further ahead than others, and the market is beginning to price that gap.

The near-term casualty of this transition may be the industry’s favorite metric: recurring revenue per user. If AI reduces headcount needs in certain functions—support agents, basic analysts, junior creators—seat growth can slow or even reverse. Vendors will try to offset this by charging for “agents” alongside humans, or by shifting toward consumption pricing. That move, however, forces a harder conversation about unit economics. If an AI agent triggers costly inference on every interaction, vendors must either absorb compute costs or pass them on. The report flags “evidence of pricing power” as a decisive signal: can vendors monetize agents without simply reclassifying existing revenue, and can the whole company stabilize as AI monetization ramps?

The new scoreboard: proofpoints, not promises

In public, nearly every software chief now speaks fluently about copilots, agents, and AI road maps. The market is no longer rewarding vocabulary. It is demanding proofpoints.

One proofpoint is stabilization in fundamentals even when traditional software budgets are flat to down. If revenue accelerates despite that backdrop, the inference is that enterprise AI spending is filtering into packaged platforms rather than being confined to experiments. Another is customer stories that AI projects are moving from custom builds to standardized software—an early sign that governance and reliability are pushing buyers back toward vendors.

A third is clarity on where domain knowledge matters. Not all assistants are created equal: a general model that drafts emails may be fine for many workers, but a CRM agent that proposes next-best actions needs business context, permissions, and structured data. Investors are increasingly asking not “Does it work?” but “Why is your agent better than the frontier model in a clean room?” The answer—if incumbents can deliver it—often rests on the unglamorous assets they already have: workflow history, data models, identity systems, and enterprise trust.

The report even floats a scenario that would have sounded far-fetched a few years ago: an AI-native company or model platform acquiring a SaaS vendor to accelerate domain experience and distribution, and to reassure customers about being “enterprise grade.” If that happens, it would validate the industry’s new reality: in AI, the interface and the distribution channel are strategic assets, and the quickest path to both may be buying rather than building.

Conclusion

For U.S. software companies, the AI threat is therefore not a single cliff but a re-rating of what matters. Durable moats will be less about feature breadth and more about orchestration, trust, data context, and the ability to monetize intelligence without letting costs run wild. The sector can adapt—and some leaders likely will. But the easy era, when software companies could raise prices on complexity and call it innovation, is ending. AI is making software simpler to use, faster to copy, and harder to defend. The next winners will be the firms that turn those forces into leverage rather than erosion.