The Monetization Question That Has Not Gone Away
OpenAI and Anthropic are building extraordinarily capable large language models. ChatGPT is widely recognized. But the question asked persistently in startup and investor circles — "what exactly are they selling?" — has not been as cleanly answered as the technology's capability would suggest.
This article examines the current state of AI monetization through the lens of discussions from "This Week in Startups," and explores what the latest developments suggest about where durable value is being created.
Where Value Accumulates in an AI Ecosystem
The fundamental question the AI industry is working through is which layer of the stack captures the most value.
The model layer — companies like OpenAI and Anthropic developing foundation models — has demonstrated enormous capability but is discovering that capability alone does not translate automatically into a defensible business. Models are becoming commodities faster than expected. The gap between the best models and good-enough models is narrowing.
The application layer — companies building specific products on top of foundation models — is where significant commercial traction is emerging. The pattern being established: model developers do not necessarily need to build the final product themselves. They can provide the platform and acquire the most promising application builders as they emerge. OpenAI's interest in acquiring Cursor and Windsurf (formerly Codium) illustrates this dynamic.
The device layer — where operating systems like Windows, iOS, macOS, and Android integrate AI deeply — represents a different kind of threat and opportunity. If device makers build AI seamlessly into the user experience at the OS level, they can absorb value from the application layer. Microsoft's AI agent strategy and Apple's AI integration efforts are both attempts to win at this layer.
Interested in leveraging AI?
Download our service materials. Feel free to reach out for a consultation.
LM Arena: A Case Study in Emerging Business Models
One interesting example of a new type of value creation is LM Arena — a platform originally from academic research that became a startup based on a simple idea: let users compare AI models anonymously and vote on which performed better.
The "arena" mechanism — showing users two model outputs without identifying which model generated each and asking them to pick the better one — generates model evaluation data that is distinct from existing benchmarks. LM Arena has attracted over a million monthly visitors and significant credibility as an arbiter of model quality.
The business model challenge is typical of research-turned-startup projects: the core product is valuable and well-used, but the revenue model is not yet established. The most credible paths being discussed:
Premium analytics. Trend data about which models perform best for specific query types has commercial value for both model developers (who want to understand their competitive position) and enterprises (who want to make informed model selection decisions). Selling detailed analytics at enterprise prices — potentially tens of thousands of dollars annually per subscriber — could cover costs and fund growth.
Community investment. A more novel idea: equity crowdfunding from the user community that has made LM Arena valuable. Users who care about the platform's survival and independence might provide capital that is less demanding than traditional VC, in exchange for a stake in the outcome.
M&A Activity: The Data Acquisition Logic
The AI M&A landscape in early 2026 is being driven less by technology and more by data.
OpenAI's interest in acquiring Cursor (or Windsurf as an alternative) is illustrative. Cursor has built a user base of developers who use AI coding assistance actively, generating real-time data about how developers interact with AI-generated code. That data, flowing continuously from millions of development sessions, is valuable for training and improving coding-specific models.
The acquisition logic:
- Data flow — continuous real-world usage data that no static training dataset can replicate
- Team and customer base — experienced builders with an established user base in the target segment
- Competitive disruption — removing a customer from a competitor's platform (Windsurf had published case studies featuring Anthropic's Claude)
Google faces constraints from antitrust scrutiny that limit its acquisition options. XAI, backed by the Tesla ecosystem and potentially gaining access to public equity as a currency if integration proceeds, may have a structural advantage in acquisition financing over private companies like OpenAI.
The Autonomous Drone Disruption
A different kind of startup is emerging from the hardware-AI intersection. Thesius, a defense-technology company, has developed a 3D-printed drone priced around $500 that navigates without GPS by comparing camera imagery to pre-loaded map data.
The capability — autonomous navigation in GPS-denied environments — has obvious military applications in conflicts like Ukraine, where GPS jamming is standard practice. The same capability raises serious questions about autonomous lethal weapons systems (LAWS) and the adequacy of current international frameworks to address them.
Y Combinator's recent batch included 43 drone-related companies, suggesting venture capital's assessment that the market opportunity is substantial regardless of the regulatory uncertainty.
What This Means for Startups and Investors
A few patterns worth noting for anyone making startup or investment decisions in the current environment:
Value is accumulating in specific applications, not general capability. The startups that are generating real revenue have specific user bases with specific use cases — not general AI assistants.
The model commoditization trend is real and accelerating. Building a startup whose primary differentiation is using a particular model is increasingly risky. The differentiation needs to come from data, workflow integration, or domain expertise.
M&A exits remain plausible at early stages. The major model companies and platform companies are actively seeking acquisitions for data and talent reasons, which creates exit opportunities for well-positioned startups even before they reach the scale that public markets require.
Branding in an AI-Saturated Market
A recurring theme in conversations about AI startups is the increased importance of branding — not as aesthetics, but as the mechanism by which a company creates differentiation in a market where the underlying technology is increasingly similar.
The key principles that separate durable brands from undifferentiated products:
Differentiation — a clear, specific claim about what makes this offering different. Not "we use AI" but "we are the AI that does X specifically for Y."
Consistency — repeating the same message across all touchpoints until it becomes automatic recognition. The test: when someone hears the company name, do they immediately know what it does?
These are not new marketing principles. But they become more important, not less, in a market where dozens of companies are using similar technology to build similar-sounding products.