Foundation Models Disrupting AI-Native SaaS Companies and Traditional Software Business Models

Foundation model providers—primarily OpenAI and Anthropic—are building vertical products that directly compete with the AI-native B2B SaaS companies that grew by reselling access to their APIs. SemiAnalysis frames this as Anthropic and OpenAI 'killing the AI-native B2B slop-SaaS after charging them $300k for tokens' [1]. Anthropic's reported launch of Claude Science, a purpose-built biopharma research platform, is a specific instance of a foundation model provider entering a domain-specific software market [2]. Established enterprise incumbents, led by Palantir, are simultaneously defending against the same models by arguing that operational depth and cultural trust cannot be reproduced by general-purpose AI [4].

If foundation model providers continue building vertical products, AI-native SaaS companies that built on those APIs may have been involuntary customer-funded R&D for their own future competitors. The investor consensus, articulated by D1 Capital's Dan Sundheim, is that software will become a worse business overall, with survivors being companies that hold genuine distribution and systems-of-record status [7]—a test that most AI-native SaaS startups are unlikely to pass.

Foundation model providers built their businesses partly on API revenue from AI-native SaaS startups—companies that used models like Claude and GPT as the core engine of vertical software products. That arrangement is now collapsing from two directions. SemiAnalysis describes the pattern bluntly: Anthropic and OpenAI are 'killing the AI-native B2B slop-SaaS after charging them $300k for tokens' [1]—meaning API customers helped fund the models that now directly compete with them. Anthropic's reported launch of Claude Science, a purpose-built platform for biopharma research, is a concrete example of this pattern: a foundation model provider entering a domain-specific software market that might otherwise belong to a vertical SaaS company [2]. Brookings has framed the structural problem as what happens when AI companies compete with their own customers [3]—a conflict of interest with no clean resolution as long as foundation model providers need both API revenue and product revenue.

Established enterprise software companies face a different version of the same problem: Wall Street is asking whether general-purpose AI can replicate what they do. Palantir CEO Alex Karp has made the most visible public defense of the incumbent position, arguing in a CNBC appearance that operational complexity in manufacturing, aerospace, and defense creates a barrier that no AI company—however capable its engineers—can bypass quickly [4]. Karp's argument is that the barrier is relational and cultural, not only technical: 'There is not a single high-end enterprise like that that would ever put that in place. That is before you even get to the cultural impasse.' At a July all-hands, Karp pushed 'sovereign AI' as a competitive frame and criticized OpenAI and Anthropic directly [5]. Databricks is taking a different posture, positioning itself as infrastructure that translates fragmented enterprise data into governed AI context using open foundation models as a base—framing those models as a component rather than a threat [6].

The investor community's assessment sits between Karp's confidence and SemiAnalysis's dismissiveness. D1 Capital's Dan Sundheim offered a Walmart analogy: software will 'probably be a worse business model going forward,' requiring enormous investment and margin compression, with new competitors emerging—but incumbents with genuine distribution and systems-of-record status can adapt rather than collapse [7]. The Walmart analogy implies survival is possible, but it required absorbing real pain before finding footing. Meanwhile, hyperscalers are absorbing that pain at scale: Microsoft closed its fiscal year at $331.8 billion in revenue with continued AI infrastructure commitment [8], and AWS showed strong growth in Q2 2026 while AI infrastructure spending consumed cash flow [9].

The competitive map between OpenAI and Anthropic themselves adds a further layer. Multiple analyses characterize them as testing different business models [10] while navigating the same structural conflict—both are moving toward enterprise and vertical products while still serving the API customers those products will eventually displace. The NYT reported that competition between the two companies is deeply personal, extending into Pentagon and enterprise contract battles [11]. Which approach produces a durable business, and at whose expense, remains the central unresolved question for the software industry.