Business Spending Data Contradicts SaaS Apocalypse Narrative

The "SaaS Apocalypse" is a Myth, According to Real Business Spending Data

The dominant narrative in the tech world is that AI is poised to decimate Software-as-a-Service (SaaS) companies, fundamentally altering how businesses purchase software and consolidating power around a few AI model providers. However, this conversation with Ara Karazian, Lead Economist at Ramp, reveals a stark contrast between this pronouncement and actual business behavior. The data, drawn from $100 billion in annual spend across 50,000 businesses, suggests that the "SaaS apocalypse" is not only premature but largely unsupported by current market trends. While AI is undoubtedly influencing software adoption, it's doing so in nuanced ways that empower new entrants and drive innovation rather than outright destruction. This analysis is crucial for founders, investors, and business leaders who need to understand the real dynamics shaping the software market to make informed strategic decisions and avoid being misled by speculative pronouncements.

The Unseen Resilience: Why Traditional SaaS Isn't Collapsing

The prevailing doomsday narrative for SaaS companies often hinges on two major predictions: that AI will cause a mass exodus from traditional software providers to model-centric solutions, and that the fundamental way businesses buy software will shift dramatically. Yet, the data from Ramp paints a different picture. Seat-based contracts, the bedrock of traditional SaaS, still command a substantial majority of business spend, accounting for 65-75%. Flat platform fees remain a significant portion, and even the much-hyped token-based pricing models, offered by giants like Adobe and HubSpot, represent a mere half-percent of overall spend. This suggests that despite the rapid evolution of AI capabilities, established purchasing behaviors and pricing models are proving remarkably sticky.

This resilience isn't merely inertia; it's often a reflection of value and integration. Consider Figma, a design tool that faced predictions of disruption from AI-native competitors. Ramp's data, however, showed Figma as one of its fastest-growing vendors. This indicates that businesses aren't abandoning established tools wholesale. Instead, they are often integrating AI capabilities into their existing workflows or adopting new AI-powered tools that address specific, unmet needs, rather than replacing entire software stacks.

"My main take is that the "SaaS apocalypse" as a pronouncement has come way too soon and is typically not informed by actual business behavior."

-- Ara Karazian

The implication here is that the competitive landscape is more dynamic and layered than the "winner-take-all" AI narrative suggests. New entrants like Anthropic are indeed making waves, even surpassing OpenAI in business adoption according to Ramp's data. Similarly, Cursor has emerged as a strong contender against GitHub Copilot. These shifts, however, are not necessarily indicative of a market collapse for incumbents. Instead, they highlight a vibrant ecosystem where innovation is driving rapid, albeit often marginal, changes. The challenge for legacy players, and the opportunity for newcomers, lies in understanding where these shifts are truly occurring and capitalizing on them before they become mainstream. This requires looking beyond the hype and focusing on the tangible data of business spend.

The Tokenization Mirage: A Slow Burn, Not a Wildfire

The transition to token-based pricing for AI services is often cited as a primary driver of the SaaS apocalypse. The theory is that as AI becomes more capable, businesses will shift from paying per seat to paying per usage, with model providers capturing the lion's share of this spend. However, Ara Karazian points out that this shift is happening at a glacial pace. While more software companies are experimenting with token-based options, the actual adoption rate remains minuscule.

This slow adoption suggests several underlying factors. Firstly, many AI tasks integrated into existing software are additive, enhancing current functionalities rather than replacing them entirely. This means that while a token cost might be associated with AI usage, the traditional seat-based model often remains in place. Secondly, the "AI apocalypse" proponents may be overestimating the immediate need for frontier models for every task. As Karazian notes, "You don't need the frontier model all the time." Many tasks can be handled by less expensive, more efficient models, or even by optimizing existing processes.

The true disruption, if it comes, will likely be driven by competitive responses. A company might adopt a token-based model to gain market share or increase usage. However, this would require a significant strategic gamble, and the current data shows few are willing to make that leap, at least not yet. The market is still largely operating on familiar terms, with seat-based contracts dominating. This creates a fascinating tension: the theoretical potential of tokenization versus the practical reality of business purchasing habits. For companies betting on a rapid AI-driven pricing revolution, the data suggests patience--or a significant shift in market dynamics--is required.

The Multi-Model Reality: Diversification as a Competitive Advantage

Contrary to the idea that AI will centralize power around a few dominant model providers, the data suggests businesses are increasingly adopting a multi-model strategy. Early adopters, often the bellwethers of market trends, are more likely to use multiple AI models and integrate a wider array of AI vendors. This diversification is not just about hedging bets; it's becoming a strategic advantage.

As AI spend increases--with token costs for some high-intensity users rising 13x in a year--companies are becoming more cost-conscious. This is leading to a greater reliance on platforms that allow for model selection and optimization, such as OpenRouter. These platforms enable businesses to leverage cheaper, open-source models when appropriate, reducing overall AI expenditure. This trend is particularly pronounced among the heaviest AI spenders, who are actively seeking efficiencies.

"Firms that were early adopters are the most likely to use multiple models and add more software, add more AI vendors over time."

-- Ara Karazian

The implication for model providers like OpenAI and Anthropic is significant. Their revenue is heavily tied to token usage, creating a direct incentive to encourage higher spend. However, as businesses become more sophisticated in their AI adoption, they will naturally seek ways to optimize costs. This creates a potential headwind for providers who cannot offer cost-effective routing or leverage open-source alternatives. The future likely belongs to those who can provide flexible, cost-efficient solutions, rather than those solely focused on maximizing token consumption. This dynamic suggests that a future dominated by a single AI provider is less likely than a more fragmented, optimized ecosystem.

Beyond the Model Labs: Growth in the Application and Infrastructure Layers

While the focus is often on the battle between frontier model providers, much of the real growth and innovation in the AI space is happening at the application, workflow, and infrastructure layers. Companies building tools to optimize AI performance, manage AI workflows, or integrate AI into specific business functions are experiencing significant traction.

Answer Engine Optimization (AEO), for instance, is a burgeoning sector that didn't exist a year ago. These tools help businesses track their performance within AI models and ensure they are being recommended. Companies like Profound are seeing rapid growth in this space, operating entirely independently of the major model labs. This demonstrates that AI is not just about the core models but also about the ecosystem of tools that make those models useful and effective for businesses.

Similarly, AI-native CRM platforms like Adio are gaining market share, offering distinct advantages over legacy players. While these companies leverage AI features, they are not simply extensions of the model providers. They represent unique product offerings that cater to specific market needs. This indicates that the "SaaS apocalypse" narrative, which often assumes a direct takeover by model companies, overlooks the vibrant innovation occurring in adjacent and complementary sectors. The opportunity lies not just in building better models, but in building better applications and infrastructure around them.

• Immediate Action: Re-evaluate current SaaS vendor contracts. Identify any seat-based pricing that could be optimized with token-based or usage-based alternatives as they become more prevalent.
• Immediate Action: Begin experimenting with multi-model strategies for AI tasks. Explore platforms that allow for routing to different models based on cost and performance needs.
• Short-Term Investment (3-6 months): Investigate emerging AEO (Answer Engine Optimization) tools and other AI infrastructure plays. These represent areas of significant growth independent of the major model providers.
• Short-Term Investment (6-12 months): Develop internal expertise in AI cost management. Train teams to identify tasks that do not require frontier models and to leverage more cost-effective alternatives.
• Long-Term Investment (12-18 months): Assess the strategic advantage of adopting AI-native SaaS solutions in key business areas, focusing on those that offer unique workflows and capabilities beyond basic AI integration.
• Discomfort Now for Future Advantage: Actively seek out and pilot less mature, potentially more complex AI solutions that offer significant cost savings or unique capabilities, even if they require more upfront effort to implement. This builds internal capability and foresight that competitors may lack.
• Strategic Review: Regularly review Ramp's AI Index and similar data sources to track shifts in business spending patterns, focusing on adoption rates of specific models and pricing structures.