Verda invests $117 million in AI cloud infrastructure as AI workloads fragment across multiple providers.

Finland-based AI infrastructure provider Verda has raised $117 million in a combined equity and debt round to expand into the US and other markets as AI demand fragments across distinct workload types. The round includes equity led by Lifeline Ventures, with participation from byFounders, Tesi, and Varma, alongside debt financing from a group of Nordic financial institutions. Verda said it will use the funding to speed platform development and expand across Europe, the US, and Asia. Rather than competing directly with hyperscalers such as AWS, Microsoft Azure, and Google Cloud, Verda is targeting a narrower opportunity: AI workloads that benefit from tightly optimized infrastructure and more direct access to GPU-backed compute.

Industry analysts attribute this opportunity to fundamental shifts in how AI workloads are being placed. "The conversation is shifting from cloud versus cloud to how AI workloads are starting to fragment," commented Matt Kimball, vice president and principal analyst at Moor Insights & Strategy. That fragmentation is opening space for smaller providers, particularly in training and emerging inference workloads. "The high-intensity, performance-sensitive work – large training runs and at-scale inference – favors vertically integrated systems," Kimball said. "Smaller providers can optimize those stacks more easily because they don't carry the same multi-tenant complexity as hyperscalers." He noted that agentic workloads pose particular challenges for hyperscalers: "Agentic workloads break the hyperscaler model. They're long-running and non-linear, which means GPUs aren't fully utilized every minute. Providers that separate prefill from decode or optimize for sustained usage can create real advantages."

Workload placement is already shifting, particularly in training where performance and compute access remain constrained. "These providers are taking real share, particularly for training workloads," said Holger Mueller, vice president and principal analyst at Constellation Research. "Risk awareness is driving that shift, as companies look to spread AI workloads across multiple clouds rather than rely on a single provider." However, hyperscalers retain structural advantages when AI workloads closely align with data and existing systems. "Anything tied to enterprise data gravity – RAG pipelines, governed data access, applications embedded in workflows – is still likely to stay with hyperscalers," Kimball said. "They already own the control plane – identity, security, and data services."

Verda said it is cash flow positive, with revenue run rate doubling to more than $60 million in the first quarter of 2026. The company participates in the Nvidia Preferred Partner Program, reflecting its reliance on Nvidia GPUs, and operates vertically integrated infrastructure across data centers and software in Finland, leveraging renewable energy and cooling advantages. The company counts customers including Nokia and ExpressVPN. CEO Ruben Bryon said the company is focused on building infrastructure tailored to AI workloads as it expands globally. "We're building the next generation of AI cloud infrastructure for pioneering teams across the globe," Bryon said in a statement.

Despite these openings for smaller providers, significant uncertainties remain about the durability of their competitive position. It remains unclear how much of Verda's demand reflects sustained workload placement versus short-term GPU constraints, or how its workload mix breaks down across training, inference, and emerging agentic use cases. "Right now, a lot of decision-making is still constrained by GPU availability," Kimball said. "If capacity opens up, workloads follow. That's not strategy – it's supply chain behavior." Ultimately, as Mueller observed, "Training is the initial foothold, but the real competition expands as inference grows. Providers are racing to build out full stacks, which puts them on a broader collision course with traditional cloud platforms." Kimball cautioned that success will depend on real differentiation: "The ones that last will be those with real architectural differentiation – not just access to GPUs."