Lilly Chen groups generative AI businesses into B2B SaaS, B2C, open source, and foundation model companies, with each model facing different funding conditions.
2
B2B startups can beat established products through fast iteration and ground-up design, while incumbents have more resources and partnerships with major foundation model providers.
3
AI makes it easy to launch a consumer product, but venture funding still depends on finding something people genuinely want.
Summary
Lilly Chen uses a video-game theme to explain how venture investors evaluate generative AI companies. She divides the market into B2B SaaS, B2C products, open source projects, and foundation models. B2B startups can challenge incumbents when they redesign a workflow from the ground up and move faster, as she argues with Notable Health and Epic EHR. Consumer AI products are now easy to build, but investors remain wary because customer acquisition and churn make growth difficult. Open source can provide a low-risk way to test whether developers will adopt a technology, with LangChain as her example. New foundation model companies face a difficult funding environment because training costs are so high. Chen ends with a point from Kyle Harrison: founders need to understand what people actually want. AI lowers the cost of building products, but it does not create demand.
Investors first ask what kind of business is being built
Lilly Chen says the first question investors usually ask is the business model. She names four common categories: B2B SaaS, B2C, open source, and foundation models. B2B SaaS companies sell to other businesses, usually face longer sales cycles, and can pursue larger enterprise contracts. B2C products sell a single experience directly to individuals, such as turning an ordinary photograph into a professional headshot. Open source companies begin with a public project, while foundation model companies build the underlying models used by other products.
B2B startups can win when they redesign a workflow from the ground up
Chen says B2B SaaS startups compete with incumbents that are adding AI to existing products. She compares Notable Health with Epic EHR. Epic has a large installed base and established hospital relationships, so it cannot quickly replace its interface. Notable Health can design a new way for people to interact with healthcare and iterate faster. Chen says incumbents still have more resources and may receive discounts through partnerships with large foundation model providers, which makes the competition difficult.
Consumer AI products are easy to launch but hard to fund
Chen says most venture capitalists are unwilling to fund a true B2C play because it is difficult to gain meaningful traction without high churn. At the same time, the tools now available make these products much easier to build. A developer who knows React, an image-generation API, and prompting could theoretically create a headshot product over a weekend. Chen has seen claims of strong monthly recurring revenue for narrow consumer products, but she says she does not know how much to believe those claims.
Chen recommends GitHub trending repositories as a way to see what is gaining attention in open source. She uses LangChain as an example of a project that began as a GitHub repository and became a company. Open source can show whether developers are willing to build on an idea. In her view, developer adoption suggests that the technology is useful and difficult to replace. It also gives investors visible evidence of interest before they commit capital.
New foundation model companies face a steep funding problem
Chen is doubtful that a new foundation model company can easily raise venture funding. She points to the amount of money already spent on GPU infrastructure and training by companies such as OpenAI, Anthropic, and Midjourney. Open source models create some overlap with the open source category, but Chen still sees it as difficult to pitch a new foundation model business against companies with far greater resources.
Product demand matters more than the presence of AI
Chen quotes Kyle Harrison of Contrary Capital, who argues that there are not enough startups to absorb all available capital because too few people build things that customers truly want. Harrison describes a finite amount of attention, time, disposable income, and corporate spending. Chen applies that idea to AI products: AI may make a previously difficult product possible, but founders still need to learn what people want. The technology helps build the product, but it does not define the demand.
