Nebius raises GPU cloud rates 17-21% amid AI demand surge

Nebius has announced a second round of price increases for its on-demand GPU cloud services in three months, citing sustained demand for AI workload capacity. The adjustments, effective October 1, will raise rates for select NVIDIA GPU instances by 17% to 21%, alongside smaller increases for CPU and memory resources. The move reflects broader industry pressures as cloud providers navigate accelerator scarcity and rising infrastructure costs for high-performance computing customers.

What changed

The pricing adjustments target Nebius’s most in-demand GPU configurations. NVIDIA H100 instances will see hourly rates climb from $3.85 to $4.50, a 17% increase. H200 pricing rises 20% from $4.50 to $5.40 per hour. The newer B-series accelerators face steeper adjustments: B200 instances move from $7.15 to $8.50 (19%), while B300 pricing jumps 21% from $7.85 to $9.50. CPU and memory charges also increase, though the company did not disclose specific figures for those components.

Nebius attributes the increases to persistent supply constraints for high-end GPUs, which have given specialist cloud providers greater pricing leverage even as enterprise customers grapple with escalating infrastructure budgets. The company’s first price adjustment occurred in June 2026, suggesting a rapid acceleration in cost pressures for AI-focused cloud services.

Market context

The pricing move arrives amid a broader shift in the AI cloud landscape. Competitors like Crusoe have recently secured substantial funding—$3.9 billion in a Series F round—to expand data center capacity and modular infrastructure, reflecting aggressive investment in GPU-dense environments. Crusoe’s valuation now stands at $30.9 billion, with over 6 gigawatts of contracted capacity globally, underscoring the scale of capital required to meet AI workload demand.

Huawei’s recent projections further illustrate the strain on infrastructure. The company forecasts a 100,000-fold increase in global token consumption by 2035, driven by AI agents, which could dominate machine-to-machine traffic within a decade. Such growth would necessitate dramatic expansion of compute clusters, exacerbating existing challenges around power, cooling, and network capacity.

For enterprise buyers, the pricing adjustments complicate budgeting for AI initiatives, particularly for workloads already struggling with cost efficiency. The increases may also accelerate migration to alternative architectures or providers, though capacity constraints remain a bottleneck across the industry.

What to watch

The pricing trend at Nebius and similar providers signals broader industry dynamics. As AI workloads consume an increasing share of cloud resources, providers are likely to continue adjusting rates to reflect supply-demand imbalances. Buyers should monitor:

• Capacity expansion timelines: New data center builds and GPU supply agreements could ease pressure, but lead times remain long.
• Alternative architectures: Some customers may shift workloads to more cost-effective configurations or providers, though options remain limited for cutting-edge accelerators.
• Contract structures: Enterprise agreements with committed spend or reserved instances may offer partial insulation from on-demand price volatility.

The adjustments also highlight the growing divergence between traditional cloud services and AI-focused infrastructure, where pricing power increasingly favors providers with scarce accelerator capacity.