Some AI companies are building software. Others are building services businesses disguised as software. As the AI market matures, revenue quality and operating leverage may matter far more than headline ARR growth.

The Great AI Services Bubble

For the better part of two decades, venture capitalists have been trained to recognize a particular type of business. The ideal company builds a product once, sells it repeatedly, and scales revenue far faster than costs. Gross margins expand, customer acquisition becomes more efficient, and each new customer increases the economic value of the business. The software industry produced some of the most valuable companies in history because these characteristics created extraordinary operating leverage.

The current AI cycle is producing companies that appear to fit that model. Revenue is growing at a pace rarely seen in prior technology waves. Startups are reaching revenue milestones in months that once required years, capital continues to pour into the sector, and investors are assigning premium valuations to businesses that seem positioned to reshape entire industries.

Yet beneath many of these growth stories lies a question that deserves far more scrutiny than it is receiving today: how much of this revenue is actually software revenue, and how much is services revenue being valued as software?

The distinction may prove to be one of the defining fault lines of the AI era. In many cases, investors are assigning software multiples to businesses whose growth remains heavily dependent on human implementation, customization, and support. Those economics can look remarkably similar in the early years. But, over time, they often become very different businesses.

The Hidden Labor Layer Behind AI Revenue

The bullish case for AI adoption is well established. According to Menlo Ventures’ 2025 State of Generative AI report, enterprise spending on generative AI reached approximately $37 billion in 2025, more than tripling from the prior year. More than half of that spending flowed into AI applications, reflecting the growing willingness of enterprises to move beyond experimentation and deploy AI into production environments.

The growth is real, the demand is substantial, and the value being created is increasingly measurable. What receives far less attention is how much human effort is often required to deliver that value.

Many enterprise buyers are not simply purchasing software. They are purchasing workflow transformation. Successful deployments frequently require integration with fragmented data systems, redesign of business processes, governance frameworks, security reviews, employee training, and extensive customization to fit the realities of large organizations. The model itself may be capable of performing remarkable tasks, but generating measurable business outcomes often requires significant work beyond the model.

Researchers at MIT have highlighted this challenge in their analysis of enterprise AI adoption. While AI capabilities continue to improve rapidly, many organizations struggle to translate those capabilities into meaningful business results because implementation, integration, and organizational readiness frequently prove more difficult than the technology itself. As a result, many AI companies find themselves deploying armies of solution architects, implementation specialists, forward-deployed engineers, customer success teams, and workflow consultants to bridge the gap between technical capability and business value.

None of this is unique to AI. Enterprise technology has always required varying degrees of implementation support. Companies like Palantir built enormously valuable businesses by embedding deeply within customer organizations and helping them solve complex operational problems. Investors generally understood the tradeoffs involved because the implementation intensity was visible.

The challenge in today’s AI market is that rapid revenue growth can obscure those same dynamics. Investors often focus on ARR, customer growth, and valuation expansion while paying far less attention to the amount of human labor required to acquire, deploy, and support each customer. That distinction matters because businesses that scale primarily through software tend to produce very different economics than businesses that scale primarily through people.

Why Revenue Quality Matters

During the SaaS era, investors learned that identical revenue figures could represent very different businesses. A recurring subscription revenue stream generated by a highly standardized product deserved a fundamentally different valuation than revenue generated through custom development, professional services, or implementation-heavy engagements. The distinction was not accounting. It was economics.

That distinction is becoming increasingly blurred in today’s AI market. Many AI companies are reporting extraordinary growth rates, but growth alone reveals surprisingly little about the underlying quality of a business. Two companies may each generate $50 million in annual recurring revenue, yet the path by which they arrived there can lead to dramatically different outcomes.

One company may deploy customers in days, require minimal support, and see margins improve as scale increases. Another may require months of implementation work, extensive customization, and ongoing human involvement long after the contract is signed. Both companies may report identical ARR figures today, but only one is systematically building operating leverage into the business.

That difference matters because investors are not purchasing current revenue streams. They are purchasing future economics. The central question is not how much revenue a company generates today, but whether each incremental dollar of revenue makes the business more scalable tomorrow.

The Old SaaS Playbook Does Not Automatically Apply

Part of the confusion stems from the tendency to apply traditional SaaS valuation frameworks to a new generation of businesses. Historically, software companies earned premium valuations because they exhibited powerful operating leverage. Once a product was built, each additional customer could be served at relatively low incremental cost, allowing margins to expand and revenue to grow faster than the underlying cost structure.

However, many AI companies are operating under a different set of economic realities than the SaaS businesses investors have traditionally rewarded. Infrastructure and inference costs remain meaningful, but the more important distinction is that customer deployments often require substantial human involvement. The result is that many high-growth AI companies exhibit margin profiles that look materially different from traditional software businesses, particularly during their early stages.

As a result, investors risk evaluating fundamentally different business models through the same lens. Some AI companies are building highly scalable software platforms. Others are building sophisticated technology-enabled services businesses whose growth remains closely tied to implementation, customization, and ongoing customer support. Both models can create significant value, but they produce very different economic outcomes over time.

The Revenue Quality Framework

As AI investing matures, I suspect the conversation will gradually shift away from model performance and toward business-model quality. The central question for investors will become surprisingly simple: is the business becoming more scalable as it grows?

Revenue quality is ultimately a measure of how efficiently a company converts customer demand into durable operating leverage. That requires looking beyond headline ARR figures and examining the mechanics of how revenue is generated. How much of the onboarding process is automated? How long does implementation take? How much customization is required for each deployment? What percentage of revenue depends on services, support, or consulting activities? And perhaps most importantly, does each new customer make the business more efficient or simply require additional human resources to support growth?

The strongest AI companies will gradually convert human effort into product functionality. Lessons learned through customer implementations become software features. Custom workflows become repeatable workflows. Deployment times shrink, revenue per employee expands, and margins improve. Over time, the business becomes increasingly software-like.

That progression is what investors should be looking for. A company that relies heavily on services during its early years may still become an exceptional software business if it systematically transforms implementation work into product capabilities. Conversely, a company that continues to require substantial human intervention with each new customer may discover that revenue growth alone is insufficient to create the economics investors expect.

In many respects, the next generation of AI winners will be determined not by who builds the most impressive models, but by who most effectively converts labor into software.

The Coming Separation

Technology markets have a long history of overlooking business fundamentals during periods of rapid innovation. New technologies generate excitement, capital becomes abundant, and growth often dominates the conversation. Eventually, however, attention shifts from possibility to economics.

The internet era eventually moved from page views to business models. The SaaS era moved from growth at all costs to efficient growth. The AI era is likely to follow a similar path. As the market matures, investors will spend less time debating model benchmarks and more time evaluating the underlying economics of the businesses deploying those models.

When that transition occurs, the distinction between software revenue and services-assisted revenue will become increasingly important. The winners will not necessarily be the companies with the fastest early growth rates, but those that most effectively transform labor into software, customization into standardization, and implementation work into repeatable product experiences. Those are the businesses that create durable operating leverage, expand margins over time, and ultimately justify software valuations.

Final Word

The AI opportunity remains enormous. Enterprises are adopting AI at an accelerating pace, new categories are emerging with remarkable speed, and entirely new forms of value creation are beginning to take shape. Few investors would dispute that we are still in the early innings of a significant technology cycle.

The more interesting question is not whether AI will create value, but where that value will ultimately accrue.

Over the next several years, I suspect one of the most important distinctions in venture investing will be between companies that use services as a temporary bridge to product scalability and companies that rely on services as a permanent component of their growth model. Today, many of those businesses appear remarkably similar. They may report comparable growth rates, raise capital at similar valuations, and compete for the same customers.

Over time, however, their economics will diverge. When that separation occurs, many of today’s AI leaders will prove every bit as valuable as investors currently believe. Others will discover that what appeared to be software economics was, in reality, highly skilled labor hidden behind a software multiple.

The next phase of AI investing will not be defined by who can generate the most revenue. It will be defined by who can generate the most scalable revenue.

***Jonathan Tower has been a global venture investor for over 20 years, having managed more than $5 Billion in AUM, invested in more than 85 companies, and seeded 9 companies that went on to become unicorns across three core investment themes: consumer (marketplaces, ecommerce enablement, digitally native brands), enterprise (software, services, infrastructure, storage, data orchestration) and frontier technologies* (AI/ML, IoT, robotics, Fintech, etc).

Jonathan’s direct investments have resulted in more than $10 Billion in exits, including early bets in Jet.com (acquired by Walmart for $3.5 Billion), Dollar Shave Club (acquired by Unilever for $1 Billion), Freshly (acquired by Nestle for $1.5 Billion), IfOnly (acquired by Mastercard), InsideView (acquired by Demandbase), and MapR Technologies (acquired by HP). Other notable investments, which Jonathan led or helped champion, include Groq (acquired by Nvidia for $20 Billion), Hammerspace, Cohere, TogetherAI, Snorkel AI, Jeeves, SingleStore, Artera, Cart.com, Madison Reed, Qumulo, and many other companies that have gone on to become market leaders.

Jonathan writes frequently on venture capital and technology topics on his blog, Adventure Capitalist, and he’s been a frequent contributor to The New York Times, Fortune, The Wall Street Journal, FastCompany, Forbes, The Washington Post, The LA Times, and other leading publications.

X: @jonathan_tower; instagram: jonathan_tower