Verda raises $189M in Series B to expand full-stack AI cloud capacity

The company develops its own data centers, hardware infrastructure, and a software platform to support various AI workloads, from rapid prototyping to foundation training and scalable inference. Verda reported an annualized revenue run rate of $165 million in July 2026 and currently serves organizations in over 50 countries.

The new capital will be used to expand data center capacity and deepen investment in its platform and its "AI Lab," a dedicated team focused on solving engineering challenges such as improving GPU utilization, inference optimization, and kernel engineering. Verda plans to expand its operations to more than 250 MW by 2027, with upcoming capacity deployments in Europe, the UK, the US, and Asia. The company also expects early deployment of NVIDIA VR200 NVL72 infrastructure in the coming months.

PLUS ULTRAby Amenoyomi

The necessity of a full-stack approach stems from a fundamental shift in how AI compute is consumed. Verda explains that AI capacity was previously managed as a procurement problem, where teams negotiated for fixed environments suited for predictable workloads. However, modern requirements—such as training runs that must start immediately and production inference where latency is non-negotiable—demand infrastructure that can respond dynamically to workload spikes.

Agentic workloads further intensify this need. These workloads often involve context that accumulates over multiple turns and demand that arrives in bursts, peaking specifically when a tool call returns. Managing these fluctuations requires an infrastructure that can scale more fluidly than traditional annual procurement cycles allow.

By vertically integrating the entire stack—from the design of data centers and internal hardware to the overlying platform—Verda aims to optimize the delivery of this capacity. This control extends to the development of proprietary compilers and serving software, as well as performance and reliability engineering at the kernel level.