This article is exclusively available to Business Insider subscribers. If you're not yet a member, consider joining today to gain access to valuable insights.

A significant transformation is currently underway in the software-as-a-service (SaaS) industry, as companies move away from traditional monthly "per seat" licensing models to adopt a more flexible usage-based, pay-as-you-go pricing approach. This shift is largely driven by the increasing implementation of artificial intelligence (AI), particularly the introduction of advanced reasoning models that require substantial computational resources and can be expensive to operate.

As businesses integrate AI into their core operations, the cost of running AI-powered software services is becoming a pressing concern, pushing companies to reconsider their pricing strategies. The emergence of costly inference processeswherein AI models evaluate data, cross-check their outputs, and refine their responseshas led to mounting operational expenses that traditional pricing structures may no longer accommodate.

In previous articles, I've highlighted the potential for the generative AI revolution to spark significant pricing changes across various internet businesses. In January 2024, I noted that the expenses involved in developing AI models were rising sharply, prompting major tech companies to seek new revenue streams, such as subscription services. Today, we face a new reality where sophisticated "reasoning" AI models demand even greater computational power. These models dont simply generate straightforward responses; they engage in complex inference processes that can create numerous "tokens"a fundamental element in the generative AI landscapewhich must be processed accordingly.

For example, OpenAI's state-of-the-art o3-high model was found to utilize approximately 1,000 times more tokens to respond to a single benchmark question compared to its predecessor, the o1 model. The estimated cost associated with generating just one answer from this advanced model is around $3,500, as determined by analysts at Barclays.

These figures illustrate a harsh reality. As enterprises increasingly weave AI capabilities into their workflowsdeveloping tools such as intelligent agents and decision-making aidsthe computational demands associated with each query are growing. This trend is particularly concerning when considering the vast scale of users; expenses can escalate rapidly when millions are engaged simultaneously.

This situation raises critical questions for software companies, many of which may struggle to sustain their traditional flat monthly fees in the face of rising and inconsistent AI usage and computational costs.

Why Traditional Pricing Models are Becoming Obsolete

For decades, SaaS giants like Microsoft and Salesforce relied on a straightforward pricing strategy that charged customers based on the number of users, per month. This model has been effective largely due to the minimal marginal costs associated with usage. However, the advent of generative AI is disrupting this paradigm. With the escalating costs of inference computing, sticking to flat pricing structures could become a financial burden.

A recent study conducted by consulting firm AlixPartners emphasizes that elevated computational costs for AI agents may lead to a higher cost of revenue relative to conventional SaaS offerings, compelling companies to reassess their financial strategies.

The Emergence of Pay-As-You-Go Pricing

In response to these challenges, companies are beginning to pivot towards a pay-as-you-go model, whereby charges are based on activity. This includes pricing mechanisms tied to the number of tokens consumed, queries executed, automations performed, or models accessed. By aligning revenue with actual usage, companies can more effectively cover their variable and rising infrastructure expenses.

Sam Altman, the CEO of OpenAI, recently suggested a pricing model that reflects this shift. He proposed that subscribers to their services could convert their existing $20 monthly plans into credits usable across various features, such as deep research, GPT-4.5, and Sora, among others. Users would have the flexibility to access features without fixed limits and could purchase additional credits as needed.

Vercel, a developer platform, exemplifies this new pricing strategy. Their model charges customers based on site traffic, meaning that as a customer's site grows, so does their payment. Marten Abrahamsen, CFO of Vercel, expressed that this approach aligns company success with customer success. "If our customer does well, we do well," he stated in a recent interview.

Pioneers of the Usage-Based Model

Several innovative companies are leading the charge towards this new pricing paradigm. For instance, Bolt.new, which specializes in low-code platforms powered by AI agents, experienced significant revenue growth after transitioning from a per-seat pricing model to a usage-based tier system. Their current pricing structure scales according to the number of tokens consumed, catering to both casual users and hardcore power users.

Other companies, such as Braze and Monday.com, are experimenting with hybrid pricing models that blend traditional seat licensing with pay-per-use AI credits. Monday.com offers its users a monthly allocation of 500 AI credits, which they can utilize until they run out, at which point they must purchase additional credits.

ServiceNow's Approach to Pricing

ServiceNow, a prominent player in the SaaS arena, has also incorporated usage-based pricing, albeit as a minor add-on to its established seat-based model. CEO Bill McDermott noted that the company invested considerable effort into developing a fast, cost-effective, and secure AI platform, collaborating closely with Nvidia to achieve this. He also remarked that many leading AI models, such as Meta's Llama and Google's Gemini, have recently become more affordable to access.

Despite this, ServiceNow has implemented a usage-based pricing component to safeguard itself against potential spikes in customer activity that could lead to excessive token usage. McDermott stated, "When it goes beyond what we can credibly afford, we have to have some kind of meter." He reassured customers that they can still engage in thousands of business processes before reaching this usage-based pricing threshold, adding, "Our customers still want seat-based predictability. We think it's the perfect goldilocks model, offering predictability, innovation, and thousands of free use cases."

Investor Reactions to the Shift

The financial community is paying close attention to these developments. Analysts at Barclays have recently noted that companies operating on a usage-based model, such as JFrog and Braze, could command premium valuations, especially as traditional seat-based providers may face slower revenue growth from AI features that do not grow in tandem with user counts.

Concerns are emerging among investors regarding the potential for AI agents to disrupt the revenue contributions traditionally associated with seat growth for SaaS vendors. These changes could lead to increased volatility in quarterly revenue reports but may also result in a stronger long-term alignment between pricing and the value of the products delivered.

The Challenges Ahead

However, this transition to usage-based pricing brings its own set of challenges. Customers will now face variable costs, making it difficult to predict monthly expenses. This unpredictability could lead to unexpectedly high charges if user engagement surges or if employees become particularly enthusiastic about utilizing new AI tools. Similarly, for the companies offering these new AI-driven software services, revenue may become more closely tied to customer success and overall activity, leading to fluctuating sales that are less appealing to investors compared to the consistent monthly revenues typically seen with traditional SaaS models.

David Slater, a chief marketing officer with experience at tech companies such as Salesforce and Mozilla, expressed concerns about cost overruns while using platforms like Bolt.new. He mentioned that heavy usage or excessive tweaking could lead to inflated costs. The key appeal of SaaS services has always been their predictability for both customers and providers. Any disruption to this model could raise significant concerns, particularly among end users. "A pricing model that's not predictive for the company and the consumer cannot stand," Slater emphasized during our conversation.

What Lies Ahead for SaaS Pricing

The shift from seat-based pricing to usage-based models is not solely an AI-driven phenomenon, but AI advancements are certainly fueling this transformation. As software evolves to become smarter, more adaptable, and increasingly reliant on computational resources, tying costs to actual usage could emerge as a more sustainable approach moving forward.

We can anticipate that more companies will roll out token credit systems, pay-per-query pricing, or hybrid models throughout 2025. This shift is not only about efficiency; it may also become essential for survival as AI adoption continues to accelerate.

Of course, the landscape could change again if the costs associated with generative AI computation decline over time. Historical trends in the computing sector indicate that such a scenario is possible, and some experts are optimistic about this potential outcome. "Sooner or later, AI costs are going to plummet, and then this usage-based model dies, replaced with an anchor like seats, or time, or a monthly subscription that's understandable," Slater remarked, hinting at the cyclical nature of technological advancements.