Nvidia (NVDA +2.59%) decoded the secret of thriving in the artificial intelligence (AI) era: build a full-stack platform comprising AI-optimized hardware, software, networking, and other components; release new and improved GPU architectures every year; and focus on capturing share in the multitrillion-dollar AI factory build-out now underway.
Image source: Getty Images.
While not yet widely known, a smaller AI player is showing strong momentum and is rapidly scaling its AI cloud infrastructure. That company is Nebius Group (NBIS +6.32%).
Cloud provider to an AI infrastructure platform
Nebius' performance in the second quarter of fiscal 2025 (ending June 30) highlights its growing strength. Revenues were up 625% year over year and 106% sequentially to $105.1 million, driven by strong demand in its core AI cloud infrastructure business. The company has also become adjusted earnings before interest, taxes, depreciation, and amortization (EBITDA) positive ahead of previous projections.
NASDAQ: NBIS
Key Data Points
Management has also raised Nebius' annualized run-rate revenue (ARR) guidance from a range of $750 million to $1 billion to a range of $900 million to $1.1 billion.
Much of this momentum is coming from its core GPU compute clusters business. The company is witnessing near-peak utilization rates and stable pricing for its AI clusters based on Nvidia's Hopper architecture GPUs, despite the next-generation Blackwell architecture GPUs expected to become available by the end of 2025.
Management said Nebius is scaling from a niche AI cloud provider into a full AI infrastructure platform. The company has expanded its cloud infrastructure capacity with large GPU clusters, doubled its networking speed, and made improvements in its software cloud platform to support large-scale AI training and inference workloads.
Nebius is seeing rapid expansion in its customer base, which now includes technology giants such as Cloudflare, Shopify, and Prosus, and leading AI-native start-ups, including HeyGen, Lightning AI, and Photoroom. The addition of such large enterprise customers has improved Nebius' long-term revenue visibility.
Capturing share in global AI infrastructure build-out
Nebius continues to witness demand far outpace its available data center capacity. To resolve this challenge, the company aims to reach 220 megawatts of connected power by the end of 2025, including roughly 100 megawatts of active power across its New Jersey and Finland data centers. The company also began expanding into the U.K. and Israel.
Nebius is aiming for 1 gigawatt of secured power by the end of 2026, supported by expansion in existing data centers as well as new data center sites. New data center sites offer full control over design and deployment and can reduce the total cost of ownership by 20% compared to current data center sites.
NASDAQ: NVDA
Key Data Points
Nvidia also took a similar path during its early AI build-out phase. It moved beyond selling GPUs and became a full-stack platform provider with integrated networking, software, and support. It has been rapidly scaling to capture a bigger share of the $3 trillion to $4 trillion AI infrastructure market.
The company's focus on introducing new architectures annually, which are also compatible with previous ones, has given enterprises a predictable upgrade path. Subsequently, Nvidia estimates that in a typical 1-gigawatt AI data center, about $35 billion of the $50 billion build cost comes from its platform.
Nebius is now following the same strategy from an earlier starting point: rapidly adding capacity with larger GPU clusters and building a software layer on top. This may help the company create a platform that can scale as seamlessly as Nvidia's. The company is also expanding its client reach by partnering with Nvidia DGX Cloud Lepton and Nvidia AI Enterprise, which can further boost enterprise adoption.
Software ecosystem
Nebius is also improving its software stack to support its expanding capacity and improve the performance of larger-scale GPU clusters. Besides accelerating network speed, the company has also improved reliability with automated health check software.
Nebius is also developing an inference-as-a-service platform to work with AI models like Llama, Qwen, and Flux on its infrastructure. This platform has been designed to ensure high performance and reliability in inference workloads, while also tackling challenges such as unpredictable latency, GPU scarcity, and scaling rigidity. The inference-as-a-service platform can result in a shift in revenue mix toward higher-margin software and services.
Just as Nvidia has created a solid competitive moat with its software ecosystem, Nebius may also see impressive demand for its software offerings in the coming years.
Valuation
Despite its rapid growth, Nebius is still in the early phases of its AI journey. Shares of the company gained 222% so far in 2025. The company carries a market cap of about $22.3 billion and trades 88 times sales. The premium pricing highlights Wall Street's high expectations about the company's ability to scale rapidly.
Nebius, however, faces significant execution risks. The company is building massive power and data center capacity, which should not miss timing and budget deadlines. Delays in securing power, GPU supply, or large enterprise deals could pressure margins and slow growth. The company also faces significant competition from other hyperscalers and AI players.
To reach Nvidia-like scale in the 2030s, Nebius will need to lock in multimegawatt contracts with enterprises and AI enterprises, keep releasing larger GPU clusters at a regular pace, and deepen its software moat to boost margins and create stickiness.
This definitely is not an easy task. However, if Nebius manages to implement this strategy efficiently and effectively, it can transition from a fast-growing cloud player into a crucial infrastructure player in the AI economy -- just as Nvidia has transitioned from a gaming-focused GPU player to an AI giant.
