The AI sector is in a massive "invest mode," spending over $600 billion on CapEx annually while generating only $110 billion in revenue. This $500 billion gap, fueled by the belief in scaling laws, makes the industry vulnerable to market hiccups and sudden investor sentiment shifts, even if the long-term potential is real.
The software layer that manages an LLM's context, actions, and outputs—termed the 'harness'—is becoming the standard architecture for AI apps, akin to the LAMP stack for web development. While the framework is consistent, its specific implementation will vary significantly across different business domains, creating opportunities for specialized value.
The launch of ChatGPT was a mass extinction event for a subset of SaaS. Roughly 10% of companies that solved problems now easily handled by large models became obsolete overnight. The survivors are either insulated, able to add AI as a feature, or are now threatened and must pivot to avoid the same fate.
To survive against foundation models, startups need moats that are structurally different from what large AI labs will build. This includes integrating with physical sensors, creating marketplaces with network effects, or building full-stack businesses that become the service provider (e.g., an AI-powered wealth management firm), not just a software vendor.
The recent crash in public SaaS valuations isn't just investor pessimism; it's a rational reaction to a fundamental decline in business performance. The average public SaaS company's growth rate has plummeted from a healthy 30% to under 10%, breaking the compounding model that previously justified high multiples.
An AI company can sustain high compute costs (e.g., 60% of revenue) only if the product is so effective that it serves as its own sales and marketing engine. If the AI drives viral adoption and the product 'leaps into users' hands,' you can spend on tokens. You cannot afford both high compute and a large GTM team.
For an existing SaaS company to become a credible AI player, it must do more than add features. It requires a fundamental reinvention, akin to 'open heart surgery.' This means re-architecting to be agent-first, shifting to outcomes-based pricing, and even rebranding—as seen with Intercom becoming Finn—to attract new funding.