For years, investors loved Software-as-a-Service (SaaS) for its beautiful, high-margin business model. But the new darling, AI SaaS, often comes with lower margins that would have been a red flag. So why are investors still excited?
The Gold Standard: Traditional SaaS Economics
To understand the difference,
you have to appreciate what made traditional SaaS so attractive. The model was simple and powerful: build a software product once, then sell subscriptions to thousands of customers. The cost of serving one more user was nearly zero. This led to famously high gross margins, often in the 70-80% range. Once a company covered its initial development and marketing costs, each new customer was almost pure profit. Revenue was predictable, scalable, and incredibly profitable, making it a favorite for venture capitalists and public markets alike. This 'build once, sell infinitely' logic became the bedrock of software valuation for over a decade.
Enter AI: A Fundamentally Different Cost Structure
AI SaaS breaks the 'near-zero marginal cost' rule. Unlike traditional software, every time a user asks an AI-powered product to do something—summarize a document, generate code, or create an image—it incurs a real, variable cost. This is known as 'inference cost', and it’s the computational power required to run the AI model for that specific query. Suddenly, serving a customer is no longer practically free. This single change ripples through the entire financial profile of the company. Instead of the lofty 70-80% gross margins of their traditional peers, AI-native companies often report margins in the 50-60% range, with some projections showing improvement over time.
The Compute Conundrum and the Talent Tax
The primary driver of these lower margins is the sheer expense of running AI models. These models are hungry for specialized processing power, primarily from GPUs (Graphics Processing Units), which are costly to buy and operate. This compute spending is a direct Cost of Goods Sold (COGS) that grows with user activity. A power user who runs thousands of queries can cost a company significantly more than a casual one, even if they pay the same subscription fee—a problem that has led companies like GitHub to shift toward usage-based billing. On top of compute costs, there's the 'talent tax'. The race for elite AI engineers and researchers is intense, driving salaries sky-high and demanding continuous, heavy investment in research and development to maintain a competitive edge.
The Investor's Bet: Why Lower Margins Get a Pass
If AI SaaS has lower margins why are investors pouring money into it often at higher valuation multiples than traditional SaaS? The answer is that they are underwriting a different kind of long term value. Instead of focusing purely on today's gross margin they are betting on a few key factors. First is the potential for a deeper competitive moat; a company with a proprietary model trained on unique data is harder to copy than a standard software application. Second is the expectation of massive scalability and the ability to automate entire workflows not just assist with tasks. Investors are also looking at different metrics. They might prioritize Net Revenue Retention (NRR)—the ability to grow revenue from existing customers—and the long term potential for margins to expand as AI models become more efficient and cheaper to run. Some are even focusing on new metrics like Annual Recurring Margin (ARM) to better capture the profitability of this new model. The bet is that while the initial costs are high the long term value and defensibility will ultimately create a larger more durable business.
