Nebius Group is no longer content just renting out GPUs. The Amsterdam-based company announced this month that it will surpass 4 gigawatts of contracted data-center power by the end of 2026, a staggering leap from less than 1 GW a year ago. The expansion is fueled by a newly raised $20–$25 billion capital expenditure budget and a $2 billion strategic investment from Nvidia, positioning Nebius as a serious challenger in the AI cloud market.
That 4 GW target—more than triple the company’s contracted power at the end of 2025—represents one of the most aggressive infrastructure buildouts outside the Big Three cloud providers. According to a Zacks Investment Research report published via TradingView, Nebius hit 3.5 GW in the first quarter of this year and now expects to end 2026 above the 4 GW mark.
The Scale of Ambition: 4 GW and a Pennsylvania Megacampus
The centerpiece of this expansion is a newly disclosed site in Pennsylvania, where Nebius has secured land and power for a facility that could eventually draw 1.2 GW. That would be the company’s second wholly owned gigawatt-scale AI campus in the United States, signaling a deliberate shift from leasing colocation space to building and owning its own infrastructure.
More than 75% of Nebius’s contracted power is now associated with company-owned infrastructure, according to management. That control promises better unit economics and supply-chain predictability, but it also exposes the company to construction delays, equipment shortages, and the notoriously slow process of connecting to the grid.
The Pennsylvania site alone would rival the power consumption of a large metropolitan area. For context, a single gigawatt is enough to power roughly 750,000 homes. Nebius’s ambition, then, is to build a network of AI factories that can train and run models at a scale previously reserved for Microsoft Azure, Amazon Web Services, and Google Cloud.
Beyond Renting GPUs: Building a Full-Stack AI Platform
Nebius is not merely piling up servers. The company is rapidly layering on software to address the entire AI workload lifecycle: bare-metal systems, multi-tenant cloud, managed inference, and agent-oriented services. That is a meaningful departure from a pure GPU leasing business, and it targets enterprise buyers who increasingly demand identity controls, observability, predictable performance, and managed model-serving layers alongside raw compute.
Earlier this year, Nebius released Aether 3.5—not 3.6 as some reports initially stated—which added serverless AI capabilities. The platform lets developers deploy models without provisioning infrastructure, a feature that can dramatically accelerate time-to-value for inference-heavy applications.
Nebius also moved to improve the economics of inference through talent and technology acquisitions. It completed the purchase of Eigen AI in June, and earlier brought in the engineering team behind Tavily, an agentic search startup. In the case of Clarifai, Nebius did not acquire the company outright; instead, it hired Clarifai’s core engineering and research team and licensed its inference and compute-orchestration technology. Those deals aim to increase throughput per GPU and reduce cost per token—critical as AI workloads shift from training massive models once to serving them millions of times in production.
Demand Is Surging, But So Are the Costs
Customer interest appears to be keeping pace with the capacity plans. Nebius reported record pipeline generation in the first quarter, up roughly 3.5 times sequentially. Newly deployed GPU capacity was fully committed, and demand outpaced available supply. To meet that demand—and to honor existing customer commitments—the company boosted its 2026 capital expenditure guidance to the $20–$25 billion range, with most of that investment expected to begin producing revenue in the first half of 2027.
Funding such enormous outlays required an equally enormous fundraising sprint. During the quarter, Nebius raised $6.3 billion, including about $4.3 billion through convertible securities and the $2 billion strategic investment from Nvidia. It ended the period with $9.3 billion in cash, giving it a war chest that few cloud challengers can match.
Still, contracted power is not the same as operational capacity. Much of that power is tied to facilities still being built. Booked demand does not automatically convert into durable revenue. Customer concentration, construction slip-ups, GPU delivery timelines, and volatile financing costs can all derail the most carefully laid plans.
What This Means for IT Teams and AI Developers
For most Windows users, Nebius’s moves won’t change the daily desktop experience. But for IT managers, DevOps teams, and AI developers who rely on cloud-based AI compute, Nebius is suddenly a viable alternative to the Big Three. Its platform now spans the tools enterprises need to train and serve models without stitching together multiple vendors.
If you’re running GPU-intensive workloads on Azure or AWS, here’s what the Nebius push could mean in practice:
- Pricing pressure: A well-funded third player with Nvidia’s blessing could keep the hyperscalers from raising prices unchecked. Even if you never use Nebius, its existence benefits you.
- Workload portability: Nebius’s platform supports bare-metal instances, Kubernetes, and serverless endpoints. That flexibility may ease migration for teams already using containerized AI pipelines.
- Inference economics: The company’s software acquisitions are squarely aimed at reducing cost per token. If you’re deploying a high-volume inference application, benchmarking Nebius’s managed inference services against your current provider could uncover savings.
- Risk diversification: Relying on a single cloud provider for all AI compute creates concentration risk. A credible second or third option allows you to spread critical workloads across providers and negotiate better terms.
None of this, however, comes with a guarantee. Nebius remains a smaller player that must prove it can deliver on time and operate with the reliability of an AWS or Azure. Early adopters should treat it as an intriguing option for non-mission-critical projects while keeping a close watch on execution.
How Nebius Got Here—and Where It’s Headed
Nebius itself is a relatively new entity, formed from the split of Russian technology company Yandex N.V. in 2024. Starting as a GPU cloud provider, it first offered Nvidia H100 and H200 clusters to AI startups and enterprises. In less than two years, it has evolved into a full-stack AI infrastructure company, mirroring the trajectory of CoreWeave, another rapidly scaling GPU cloud that went public earlier this year.
The Nvidia investment, announced in late 2025, marked a turning point. Nvidia rarely makes direct equity bets on cloud providers, and its $2 billion infusion signaled confidence in Nebius as a strategic outlet for its GPUs. Since then, Nebius has used that endorsement to attract customers and raise additional capital.
2026 has been a year of aggressive scaling. In January, contracted power stood at just over 2 GW. By April, it had reached 3.5 GW. Now, the addition of the Pennsylvania campus pushes the year-end target above 4 GW. The company’s revenue timeline suggests that 2027 will be the real proving ground, when customers should start running production workloads on the new hardware.
What You Should Do Now
If you’re evaluating your organization’s AI infrastructure strategy for the next 12–18 months, consider these steps:
- Audit your AI compute spend. Identify which workloads are most sensitive to GPU pricing and which could be moved without major rearchitecting.
- Add Nebius to your vendor watchlist. Track its progress on the Pennsylvania campus and other builds. Request early access or a trial environment if you have a qualifying project.
- Benchmark performance. Run sample training jobs or inference tasks on Nebius and compare against your current provider. Pay attention to total cost per job, not just per-hour GPU pricing.
- Consider a multi-cloud AI plan. Even if you don’t switch providers today, design your MLOps pipelines to be cloud-agnostic. Containerized training scripts and model-serving endpoints that can run on any Kubernetes cluster will make a future transition easier.
- Watch the 2027 milestone. Nebius expects its $20–$25 billion capex investment to start generating revenue in the first half of that year. If the company hits that target and customers stay, it could become a permanent addition to the AI landscape.
The Road Ahead
Nebius’s blueprint is ambitious, but the AI infrastructure boom is littered with plans that never materialized. The company’s cash balance of $9.3 billion and its close relationship with Nvidia give it a running start, but construction, power-grid connections, and cooling remain physical constraints that money alone can’t solve overnight.
The real test begins when the concrete sets and the GPUs hum. If Nebius can bring its Pennsylvania campus online on schedule, convert its contracted power into operational capacity, and retain its early customers through reliable service and competitive pricing, it could carve out a meaningful slice of the AI cloud market. For cloud buyers, that would be a welcome development—more choice, lower prices, and a hedge against the dominance of a few mega-providers.
For now, Nebius is a story to follow closely, not yet a done deal. But in an era where AI compute is the new oil, having another refinery under construction is news worth watching.