Investing in Software in the Age of AI: A New Equation for Value Creation

The Market Reckoning: What the Numbers Reveal

In early 2026, the global software market is undergoing a transformation unlike anything seen before. One thesis has emerged with striking clarity: what determines a software company's value in the AI era is no longer whether it sells subscriptions — it is whether the company can harness AI to deliver measurable ROI, embed itself into customer workflows, and build durable data moats.

The scale of the correction is hard to overstate. According to the Sapphire Ventures Enterprise Software Index, which tracks 252 enterprise software companies, aggregate market capitalization has shed over $2 trillion from its late-October 2025 peak. Valuation multiples have collapsed nearly 80% from the highs of 2020–2021.

But the most telling signal is not the decline itself — it is the convergence. Historically, “Pure SaaS”¹ companies (those deriving more than half their revenue from subscriptions) have always traded at a meaningful premium to “Broad Software” companies with mixed license, on-premise, and hardware revenue. At their respective peaks, Pure SaaS traded at 15.2x NTM EV/Revenue (December 2020) versus 11.3x for Broad Software (October 2021). As of late February 2026, both groups trade at an identical 3.1x. The premium that SaaS commanded for roughly five years has vanished.

This raises a fundamental question: has the market simply stopped rewarding the subscription model as a structural advantage? Or are we witnessing a deeper realignment — one in which the relevant divide is no longer SaaS versus non-SaaS, but AI winners versus AI losers? This report tackles both questions. We begin by diagnosing the sources of the market’s anxiety, then outline the structural conditions that separate software companies positioned to thrive in the AI era from those at risk of being left behind.

Understanding the Fear — and Where It’s Warranted

What makes this selloff unusual is that fundamentals have held up surprisingly well. FCF margins² across the SaaS universe have actually improved, and Rule of 40³ scores are trending higher year-over-year. The sell pressure is not being driven by deteriorating earnings — it is being driven by existential anxiety about what AI means for the industry’s future. That anxiety breaks down into four distinct fears.

Fear #1: “AI Agents will make existing software obsolete.”

There is some truth to this. As AI Agents⁴ grow more capable, single-function and workflow-automation tools — especially generic, horizontal SaaS products — face real displacement risk. But the picture is more nuanced in practice. Enterprise buyers still require compliance guardrails, security controls, and human-in-the-loop oversight. Industries like healthcare, financial services, and manufacturing — where domain-specific heuristics and conservative procurement cycles dominate — are unlikely to see wholesale software replacement any time soon.

Fear #2: “Falling development costs will lead companies to build instead of buy.”

AI-powered development tools are undeniably making it cheaper to build lightweight internal applications. At the margins, this will eat into adoption of generic SaaS. But software value is not reducible to code. Years of accumulated business logic, regulatory compliance frameworks, and deep integration with surrounding systems constitute a moat that cannot be replicated with a coding copilot. Deeply embedded systems of record — especially those approaching ERP-level criticality — enjoy structural defensibility through switching friction and compliance fit. The more realistic threat is not build-versus-buy, but vendor-to-vendor switching: customers are increasingly likely to migrate to competitors that have integrated AI capabilities more effectively.

Fear #3: “Software revenue growth will structurally decelerate.”

Of the four fears, this may be the most credible. Most software companies still price on a per-seat basis. If AI proves capable of replacing white-collar headcount at scale, enterprises will inevitably cut seats — and software categories where ROI is hard to quantify will take a direct hit. Compounding this, CIOs and CFOs are re-prioritizing budgets toward AI initiatives, potentially crowding out spend on incremental software adoption.

That said, the threat is not universal. Where AI-generated productivity is measurable in real time, usage-based pricing and new AI-native features offer a genuine path to revenue expansion. The transition to new pricing models may itself become a growth catalyst.

Fear #4: “The software sector faces a permanent multiple reset.”

This fear is already materializing. With seat-based growth under question and AI adoption separating leaders from laggards, the market is applying a broad, conservative re-rating across the sector. Funds that entered software at peak multiples in 2020–2021 now face a painful reality: even if portfolio companies execute flawlessly on their growth plans, exit multiples may be structurally lower than entry multiples.

However, this reset is unlikely to be uniform. Companies that can demonstrate tangible productivity improvements and healthier economics through AI will see their multiples re-differentiate. Paradoxically, the current compression may represent an attractive entry window for PE investors willing to underwrite quality software assets at conservative valuations.

AI-Native Companies: Fast Growth, Fragile Moats

Over the past few years, a new generation of companies has emerged that are AI-native from the ground up. Whether they build applications on top of commercial foundation models⁵ or develop proprietary model layers, these companies share a defining trait: the AI model is the product.

The growth trajectories have been remarkable. Charts now familiar across the investment community show AI-native companies like Cursor and ElevenLabs reaching $100M ARR with a fraction of the headcount and a fraction of the time that legacy SaaS companies required. Part of this is sheer development velocity, but it also reflects a structural business model shift: AI-native products tend to adopt Usage-Based Pricing, which creates fundamentally different unit economics compared to the traditional seat-based model. As AI-driven productivity becomes measurable in real time, customers are naturally gravitating toward consumption-based billing — and hybrid models blending usage and seat pricing are proliferating.

But speed alone is not a moat. The “AI-native” label does not automatically confer competitive advantage over traditional SaaS incumbents, which are themselves layering AI capabilities at pace. And in enterprise sales environments, even AI-native vendors are often compelled to offer seat-based pricing alongside usage models.

More fundamentally, many AI-native companies may lack structural defensibility altogether. Thin wrapper⁶ copilot products — those that layer minimal UI over a foundation model without proprietary data, logic, or workflow integration — are acutely vulnerable to commoditization as OpenAI, Anthropic, Google, and other model providers expand their native feature sets. And the risks extend well beyond wrapper displacement:

•       Inference costs that must be absorbed within gross margins

•       Dependency on a small number of foundation model providers

•       Revenue volatility inherent to usage-based pricing

•       Enterprise adoption friction driven by security, audit, and data governance requirements

These are fundamental vulnerabilities that top-line ARR growth can obscure but not eliminate.

So what does a defensible AI company look like? The most durable players will likely combine three attributes: domain depth built on proprietary industry data and vertical-specific logic; trustworthiness rooted in security and regulatory compliance; and product architecture that is genuinely embedded in real business processes — not merely a tool bolted onto a workflow. An alternative model with strong lock-in potential is the AI implementation partner: akin to a cloud MSP, these firms customize and operate AI systems within client environments, creating deep integration ties that are difficult to unwind.

The bottom line: competitive advantage in the AI era is not about using AI. It is about whether AI enables you to build a moat that others cannot replicate.

Implications for PE: Investment Selection and Portfolio Value-Up

The criteria that determine software company competitiveness — and, by extension, valuation — are shifting at the foundation. Functional completeness and recurring subscription revenue are no longer sufficient. What matters now is the degree to which a software platform can deliver meaningful decision-making automation and productivity gains through AI.

Companies with deep, vertically specific data — proprietary assets accumulated through years of domain expertise and inaccessible to outside competitors — are uniquely positioned to deliver AI-driven performance that generic, horizontal foundation models cannot match. Beyond the product layer, whether a company has successfully re-architected its pricing, go-to-market, and operational KPIs will be a defining driver of enterprise value.

For PE investors evaluating buyout targets, the following questions should guide due diligence:

• Does the company’s data have enough domain specificity to create meaningful AI performance differentiation?

• Can AI be embedded deeply into actual customer workflows, rather than sitting as a surface-level feature?

• Is the regulatory and compliance landscape stable and navigable?

• Does the product deliver clear, measurable ROI from the end user’s perspective?

• Do these factors collectively produce high switching costs?

These criteria map to four evaluation dimensions:

• Product Moat: proprietary data depth, workflow embeddedness, regulatory compliance

• Unit Economics: pricing power, gross margin profile, capacity to absorb inference costs

• Commercial Viability: customer adoption speed, ROI visibility, switching costs

• Strategic Risk: foundation model dependency, pace of competitor AI integration

The key is to evaluate simultaneous strength across all four — not to treat them as an independent checklist.

On the value-up side, the imperative is to drive AI-led operational transformation and business model evolution across the existing portfolio. This means structuring domain data assets and embedding AI implementation capabilities directly into portfolio companies. In practice, five execution priorities should anchor the roadmap:

• Structure and label domain data assets systematically

• Prioritize AI feature integration around high-impact core workflows

• Migrate pricing from seat-based to outcome- or usage-based models

• Redesign sales, CS operations, and organizational structure around AI-enabled delivery

• Shift performance measurement from ARR as a single KPI to a multi-layer framework incorporating AI ROI, gross retention, and gross margin

The market is in a transitional phase — technology is advancing on a weekly, sometimes daily basis, and volatility remains elevated. But history has consistently demonstrated that periods of disruption and creative destruction are also periods of outsized opportunity. Centroid intends to capitalize on this moment by identifying undervalued, structurally advantaged software assets and by actively deploying an AI transformation framework across its portfolio to drive sustained value creation.

Source:

Sapphire Ventures, 2026 Software x AI Report: Software’s AI Inflection Point (March 2026). All figures and charts cited in this report are drawn from the above source. Analysis and implications reflect the views of Centroid Investment Partners.

Footnotes

[1] Software as a Service: a cloud-based delivery model in which software is accessed via the internet on a subscription basis, typically priced per user seat or feature tier.

[2] Free Cash Flow Margin: free cash flow (operating cash flow less capital expenditure) as a percentage of revenue; a measure of actual cash generation capacity.

[3] Rule of 40: a benchmark for software company health combining revenue growth rate (%) and EBITDA margin (%). A combined score of 40 or above is generally considered indicative of a financially sound business.

[4] AI Agent: an AI system capable of autonomously planning and executing multi-step tasks toward a defined objective — encompassing activities such as web search, code execution, and external system integration beyond simple Q&A.

[5] Foundation Model: a large-scale, pre-trained AI model with broad applicability across domains (e.g., OpenAI GPT-4, Anthropic Claude, Google Gemini), commonly used as the base layer for AI product development.

[6] Thin Wrapper: a product that adds minimal UI and functionality on top of a foundation model without proprietary data or technical assets — effectively repackaging the underlying model’s capabilities and highly vulnerable to displacement as model performance improves.