Nebius just raised GPU prices AGAIN and it is the clearest sign yet that AI compute demand is still outpacing supply. The company is increasing pay as you go prices for Token Factory’s dedicated inference endpoints by 16% to 20% starting October 1. H100 pricing will rise from $4.05 to $4.70 per GPU-hour, H200 from $4.70 to $5.60, B200 from $7.40 to $8.70 and B300 from $8.10 to $9.70. This is separate from the cloud GPU price increase Nebius announced earlier. That earlier change raised standard on demand cloud prices by roughly 17% to 21%, taking the H100 to $4.50 per hour, H200 to $5.40, B200 to $8.50 and B300 to $9.50. Nebius is now charging an additional $0.20 per GPU hour for every comparable GPU inside a Token Factory dedicated endpoint. The difference exists because customers are not merely renting a raw GPU. Token Factory’s dedicated endpoints provide isolated model deployments, reserved capacity, predictable latency, a 99.9% uptime commitment and customer controlled autoscaling. Nebius handles much of the infrastructure required to deploy and operate the model, while customers access it through an OpenAI compatible API. That makes Token Factory a higher value managed service layered on top of Nebius’ underlying GPU cloud. Customers pay more, but they avoid managing clusters, orchestration, scaling, model serving and availability themselves. The new rates therefore show Nebius trying to monetize both layers of its platform. It is raising the price of the underlying GPU infrastructure while also raising the price of the managed inference product built on top of it. That matters because Token Factory can potentially generate more revenue per GPU than simply renting the same hardware as raw compute. A B300 running continuously at the new dedicated endpoint rate would generate approximately $6,984 per month, compared with $5,832 at the old rate, before discounts and assuming full utilization. That is roughly $1,152 of additional monthly revenue per fully utilized B300, or nearly $13.8 million annually across 1,000 continuously running GPUs. The economics are similar across the lineup. At full utilization, the new pricing adds approximately $468 per month for each H100, $648 for each H200 and $936 for each B200. Not all GPUs will run constantly at list price, and large customers can negotiate discounts but the increases demonstrate the potential operating leverage. Most of Nebius’ data center, server and GPU costs are fixed or committed in advance. If Nebius can charge 16% to 20% more for the same installed hardware without suffering a comparable increase in operating costs, much of the additional revenue can flow into gross profit. The fact that Nebius is protecting contracted prices also matters .Existing reservations and previously negotiated contracts will not be affected, meaning the immediate revenue benefit will primarily come from new customers, renewed contracts and flexible pay as you go usage. That limits the near term impact but reduces the risk of angering major customers that already committed to the platform. The company also introduced dynamic spot pricing for preemptible virtual machines earlier today and those prices will automatically move according to available capacity and real time demand for each GPU type and region. Customers can set a maximum price and allow their workload to stop when the spot rate exceeds it, or follow the market price to improve their chances of maintaining access. This gives Nebius three distinct ways to monetize the same GPU fleet. Long-term customers receive reserved capacity and discounted pricing, on demand customers pay more for flexibility, and interruption-tolerant customers bid for spare capacity through the spot market. Token Factory then adds another premium layer for customers who want Nebius to manage model deployment and inference. To me, this is one of the clearest signs yet that demand is still running ahead of supply. If Nebius can continue raising pricing, improve utilization and move more customers up the stack into managed services, the economics of every new megawatt and every new GPU they bring online become significantly more attractive. Bullish on Nebius!