Investor Sarah Guo: Inference is Already a Scale Game
Investor Sarah Guo believes that previous analyses regarding the low defensibility of the inference layer do not align with reality, pointing out that inference has now become a scale game.
She stated that as long as one has actual workloads, financing capabilities, and operational and technical teams, they can compete; the earlier judgment about customer churn being extremely high is at least half completely wrong.
Leaders rarely lose customers, and even when they do, they usually switch to other leaders rather than any new entrants.
She added that if the next company attempts to sell $20 million worth of inference services at a negative gross margin without a path to scale supply, they can try freely, but it will be difficult to form a sustainable business.
This response directly addresses the viewpoint that the inference layer lacks barriers due to easy access to capital and low switching costs, emphasizing the decisive role of scale, customer retention, and supply capability.
Market Mechanism: Leaders with scaled workloads and operational capabilities attract and retain customers, while capital and computing power prioritize companies that already have paths for supply expansion; competitors with negative gross margins lacking scale paths face dual pressures of funding and supply, while the application layer can still choose among leaders, but migration mainly occurs among the top players.
Source: Public Information
ABAB AI Insight
After leaving Greylock in 2022, Sarah Guo founded Conviction, investing early in Baseten and continuously supporting its transition from data science tools to a production-grade inference platform; Conviction's portfolio also includes Harvey, Cognition, Mistral, covering both application and infrastructure layers. Previously at Greylock, she participated in transactions like Figma Series A, focusing on software infrastructure and enterprise software pathways.
The capital path concentrates on directing funds to teams with existing workloads and operational capabilities rather than purely conceptual inference companies; Conviction emphasizes early deep involvement, helping portfolio companies acquire customers, computing power, and talent to establish scale. The motivation is to capture the scale effect as inference demand spreads from the lab to the application layer, with resources leaning towards companies that can convert financing into actual supply expansion.
Similar cases include AWS in early cloud infrastructure forming customer stickiness through scale and operational capability, making it difficult for later entrants to significantly poach core customers even with price reductions; the current inference layer is transitioning from expansion to a focus on scale, with leading companies consolidating their positions through workload accumulation and supply capability.
Essentially, this is about capital concentration. Under the scale game mechanism, financing capability, workloads, and operational teams jointly determine who can continuously expand supply and retain customers; negative gross margin competitors lacking scale paths struggle to secure follow-up capital, leading resources to concentrate on validated scale leaders, resulting in limited competition among the top rather than a full-blown price war.
ABAB News · Law of Cognition
- In a scale game, workloads are harder to replicate than prices.
- The real customer flow occurs among leaders.
- Negative gross margins without supply paths are just temporary noise.