DigitalOcean (DOCN) Citi’s 2026 Global TMT Conference summary

Strategic direction and AI platform evolution

• Focus shifted from hardware-intensive AI training to software-driven inference, emphasizing managed inference as a durable, production-grade workload aligned with cloud consumption patterns.

• Platform built from the ground up, integrating five layers for seamless AI-native cloud operations, enabling co-innovation with leading-edge AI-native customers.

• Product-led growth model drives rapid customer adoption, with over 6,000 customers in less than four months, primarily through organic developer engagement.

• Managed inference platform differentiates by abstracting infrastructure complexity, offering API endpoints with customizable SLAs and throughput, supporting advanced reasoning models.

• Industry trend shows a shift from closed-source to open-weight models, with customers increasingly fine-tuning and reinforcing models for domain and customer specificity.

Emerging use cases and customer adoption

• AI-native workloads are expanding beyond coding to generative media, agent-first enterprise applications, gaming, and personal productivity agents.

• Customers leverage features like model routing, model synthesis, and agent orchestration to optimize cost-performance and manage complex, multi-model workflows.

• Adoption is moving up the stack, with AI demand driving attachment of core cloud services at a 70% rate among 100,000+ AI customers.

• The platform's flywheel effect sees customers entering for AI tokens and expanding into broader cloud and agent services, increasing stickiness and revenue per customer.

• Higher-value inference and cloud services yield ARR per megawatt significantly above industry averages, with incremental ARR per megawatt 30–40% higher than neocloud peers.

Financial performance and growth levers

• ARR per megawatt reached $22 million in Q2, with further upside expected from new services, pricing, and hardware advancements.

• Three main growth drivers: newer GPUs generating more tokens per megawatt, increased attachment of higher-layer services, and token optimization to maximize infrastructure utilization.

• Transition from bare metal to token-based consumption enables higher pricing and utilization, with strategies to increase off-peak usage and global demand balancing.

• Capacity expansion is managed through partnerships with tier one data center providers and strong relationships with GPU suppliers, ensuring timely and efficient scaling.

• Gross margin and EBITDA margin face short-term pressure from capacity build-out, but operating leverage remains strong, with continued investment justified by high returns.