The Old Playbook: The Beauty of SaaS
For years, valuing a software-as-a-service (SaaS) company was a refined science. Investors prized the business model for its beautiful predictability. A SaaS company builds a product once and sells it many times over through subscriptions. The key was 'recurring
revenue.' Metrics like Annual Recurring Revenue (ARR) and low customer churn created a stable, forecastable financial future. Best of all, SaaS companies enjoyed incredible gross margins, often in the 75-85% range. Once the software was built, the cost to serve an additional user was almost zero. This created immense operating leverage, where revenue could scale exponentially while costs remained relatively flat, justifying high valuations.
Problem 1: The Crushing Weight of Upfront Costs
AI infrastructure flips the script. Before an AI company can even think about revenue, it faces staggering upfront capital expenditures (CapEx). The race to build powerful foundation models requires massive investment in R&D and, most notably, thousands of specialized GPU servers. This is not the capital-light world of SaaS. We're talking about industrial-scale spending on hardware and data centers before a single customer signs up. J.P. Morgan projected global AI CapEx to hit $900 billion in 2026 alone. This hardware-like spending profile breaks the first rule of the SaaS playbook, making AI infra look more like a manufacturing operation than a software business.
Problem 2: The End of Predictable Revenue
The second pillar of SaaS valuation—predictable, recurring revenue—also crumbles. While some AI companies use subscriptions, many operate on usage-based or consumption models where customers pay per API call or per 'token' processed. A token is the basic unit of data an AI model processes, like a piece of a word or an image. This revenue model is inherently volatile and difficult to forecast. Unlike a fixed monthly seat license, a customer's spending can swing wildly from one month to the next based on their activity. This unpredictability makes it challenging to apply traditional models that rely on smooth, linear growth assumptions.
Problem 3: The Mystery of Gross Margins
Perhaps the most significant departure from SaaS economics is the gross margin. A SaaS company's cost of goods sold (COGS) is minimal. For an AI company, it's a huge and variable expense. Every time a user query hits a model—a process called 'inference'—it consumes real, expensive compute power. This inference cost is a direct COGS. As a result, AI company gross margins are structurally lower than SaaS, often landing in the 45-60% range, compared to the 80%+ investors are used to. More usage means more revenue, but it also means proportionally more cost, destroying the near-zero marginal cost structure that made SaaS so attractive.
The New Calculus: Betting on a Revolution
Given these challenges, why are investors pouring billions into the sector? The answer is that they are not using the old models. Valuing AI infrastructure is less about analyzing current cash flow and more about pricing a potential technological revolution. Investors are making a bet on market dominance. The belief is that the AI infrastructure layer is a winner-take-most market where a few foundational platforms will capture immense value. The valuation isn't based on a multiple of this year's revenue, but on the potential to own a piece of the next fundamental computing paradigm. It's a high-risk, high-reward game of assessing technological moats, proprietary data, and the strength of research teams, rather than stable financial metrics.













