投资人如何区分真正的 AI 产品收入，与伪装成软件收入的服务收入？

“The 20VC panel distinguishes between hollow “AI-influenced” revenue — where existing features are rebranded — and genuine net new bookings generated by AI that customers pay extra for. Bundling AI without separable value creates a risk: upon contract renewal, customers may drop the AI add-on, reducing revenue.”

“A pricing model where users pay a break-even subscription fee and the platform additionally takes a percentage of the revenue generated through the AI agent, plus a portion of the ad spend managed by the agent, thereby aligning incentives on actual business success. This structure ties the platform’s financial success directly to the user’s ability to make money, incentivizing genuine business value creation.”

“Revenue stacking, also called margin stacking, is the phenomenon where a single dollar of end-customer spend is counted as revenue multiple times as it moves through the AI technology stack. The same underlying dollar therefore appears in the top line of at least three different companies, inflating aggregate market-size figures.”

“Much of the revenue reported by AI startups may be little more than a resale of compute, sometimes with negative gross margins. Until a shift to optimization mode occurs, the true quality of AI company revenue remains uncertain.”

“David Friedberg notes that AI is now completing work that would have required a services firm—drug discovery, factory design, engineering projects—enabling a pricing model akin to a services engagement. When software performs tasks humans cannot, charging by outcome rather than per seat becomes possible, collapsing the boundary between software and services and potentially increasing total addressable market dramatically.”

“Immad Akhund argues that companies selling AI as a cost-saving substitute for human labor — typically charging about one-third of the labor cost — will face rapid competitive erosion because all competitors use the same foundation models, creating minimal moats. Akhund contrasts this with more durable AI revenue models, such as Cursor's SaaS-like developer tools, where switching costs are high and the value delivered far exceeds the price charged.”