The World Inference Cup | Top Inferences from Inference-as-a-Service Day

Research evidence excerpt

The World Inference Cup | Top Inferences from Inference-as-a-Service Day

isition should help insulate the company across demand scenarios, including

frontier labs, open-source adoption, AI-native customers, and enterprise deployments. INDUSTRY RISKS

2) Right to Win varies by customer segment. For hyperscalers, the differentiator is trust in Macroeconomic, IT Spending, Architectural

execution and delivering contracted capacity on time. For frontier labs, execution remains Shifts, Competition / ASPs, & Valuations

critical, but performance and differentiated compute matter more, helped by CRWV’s strong

Nvidia positioning. For AI-native customers, developer and post-training tools are more

relevant, while enterprises care more about top-of-stack software and ease of deployment.

3) Sounded confident in its ability to reach its 8GW+ active capacity target by 2030,

citing visibility into demand and customers looking to add gigawatts of capacity.

4) Margin: Management also expects margin expansion to accelerate into Q3/Q4 as large

compute deals normalize, with flat margins as the conservative case beyond this year and

potential upside as older chip generations are redeployed.

DigitalOcean (Neutral) - Paddy Srinivasan (CEO) & Radu Patrichi (SVP Corp Dev & IR)

1) Positioning is around a differentiated AI cloud stack for AI-native customers rather

than competing purely on scarce GPU capacity or price. Its platform includes managed

inference, serverless and dedicated inference, managed agents, sandboxed agent

environments, and core cloud pull-through. Management believes this stack should remain

valuable even if raw GPU supply normalizes.

2) Training vs Inference: Coding is the only inference sub-vertical that has clearly taken