Nebius’s price page, read on 8 October 2026, lists higher on-demand GPU-hour prices in a column headed “GPU-hour (Effective October 1, 2026)” than in its “On-demand, GPU-hour” column for its NVIDIA HGX H100, H200, B200 and B300 instances. NVIDIA HGX H100 goes from $3.85 to $4.50 per GPU-hour and HGX B300 from $7.85 to $9.50.
On our arithmetic, the increases are about 17% for H100, 20% for H200, 19% for B200 and 21% for B300.
Business Pill · RENTING THE CHIPS BY THE HOUR
A one-minute explainer of neoclouds: companies that rent AI chips by the hour to anyone who needs compute. It teaches the general idea only and says nothing about any company in this story.
The key insight: As we read it, the increases fall on the GPUs Nebius lists by the hour for training and serving models, and include the oldest generation on its page. H100 rises by about 17% and B300 by about 21% on our arithmetic, while the RTX PRO 6000 and L40S lines do not change.
The page lists NVIDIA HGX H200 at $4.50 in the on-demand column and $5.40 in the column effective 1 October, and NVIDIA HGX B200 at $7.15 and $8.50.
Not everything moves. NVIDIA RTX PRO 6000 is $1.80 in both columns, and the L40S instances are listed from $1.55 with Intel CPUs and from $1.82 with AMD CPUs in both. GB200 NVL72 and GB300 NVL72 are listed as “Contact us”.
Preemptible, or spot, prices on the same page start from $0.99 per GPU-hour for B300 and B200 and from $0.79 for H200, H100 and RTX PRO 6000; that table has no separate column for 1 October.
Nebius on-demand prices per GPU-hour, as listed on its price page read on 8 October 2026: the ‘On-demand, GPU-hour’ column versus the ‘GPU-hour (Effective October 1, 2026)’ column. H100 $3.85 vs $4.50; H200 $4.50 vs $5.40; B200 $7.15 vs $8.50; B300 $7.85 vs $9.50.Change from Nebius’s on-demand column to its column effective 1 October 2026, our arithmetic: H100 +16.9%, H200 +20%, B200 +18.9%, B300 +21.0%.
What Was Posted on X
$NBIS spot pricing for $NVDA B200 GPUs has more than tripled in the past 7 days.
Thatโs huge heading into year-end because Nebius annualizes December revenue to calculate ARR meaning higher GPU pricing could help beat its $9B target.
The account @StockSavvyShay posted this at 00:39 UTC on 20 October, with a screenshot of what appears to be Nebiusโs B200 spot-price card for us-central1 showing $3.95, up 203.8% on seven days earlier. That is the postโs figure; when we read Nebiusโs live page later that morning it listed $3.35. The argument that pricier spot capacity could lift Nebius past a $9B ARR target is the postโs view. Nebiusโs own guidance is year-end 2026 ARR of $7 billion to $9 billion, and Nebius computes ARR from the last month of the quarter times 12.
CPU Instances Too
The page shows the same two-column layout for CPU-only instances: AMD EPYC Genoa from $0.10 per hour in the current column and from $0.13 in the column effective 1 October, and Intel Ice Lake from $0.05 and from $0.06.
On our arithmetic, that is about 30% for the Genoa starting price and 20% for Ice Lake.
What It Means for a Full Day
On our arithmetic from the listed per-GPU-hour prices, eight H100s for 24 hours cost $864 at the 1 October rate, against $739.20 at the rate in the on-demand column. Eight B300s for 24 hours come to $1,824, against $1,507.20.
The page lists each GPU-hour with its vCPU and memory allocation: 16 vCPUs and 200 GB of RAM for H100 and H200, 20 vCPUs and 224 GB for B200, and 24 vCPUs and 346 GB for B300.
On-Demand Against Spot
The gap to the preemptible floor widens with the new rates. On our arithmetic, the 1 October on-demand price is about 5.7 times the spot starting price for H100, 6.8 times for H200, 8.6 times for B200 and 9.6 times for B300.
The page lists the spot prices under a separate heading, “Preemptible NVIDIA GPU Instances”, from the on-demand table, so the two are different classes of instance rather than two prices for the same one.
Update, 20 October: Spot Prices Now Move, and the B200 Jumped
The spot starting prices quoted above are no longer fixed. Starting 8 October 2026, Nebius switched preemptible pricing for its B300, B200, H200, H100 and RTX PRO 6000 platforms to dynamic spot pricing, according to its pricing documentation. Nebius says the spot price “may change as often as every 15 minutes based on factors such as capacity availability, cluster utilization and demand”, and that the top of each platform’s price range is lower than the regular pay-as-you-go price, per its pricing policy.
Nebius’s live spot-price page listed the B200 in its us-central1 region at $3.35 per GPU-hour, effective since 06:21 UTC on 20 October, when we read it. A screenshot of what appears to be the same page’s B200 card for us-central1, posted on X at 00:39 UTC by the account @StockSavvyShay (Shay Boloor), showed $3.95 and “+$2.65 (+203.8%)” against seven days earlier. That implies a starting level of about $1.30 and roughly a tripling over the week, on our arithmetic. We did not see the seven-day history ourselves, and spot prices can move again within minutes.
Why this could matter for the company’s targets: Nebius calculates ARR by taking AI cloudrevenue from the last month of the quarter and multiplying by 12, according to its Q2 2026 shareholder letter, which also said: “We are reaffirming our full-year 2026 guidance across all metrics.” In its 12 February letter, Nebius said it was on track to end 2026 with ARR of $7 billion to $9 billion. The post argues that higher GPU pricing could help Nebius beat a $9B target. We have not seen Nebius disclose how much of its revenue comes from preemptible capacity, so the size of any effect is not established.
The Structural Read
The top of the range is not on the price list. GB200 NVL72 and GB300 NVL72 are listed as “Contact us”, so the largest systems are priced privately.
The cheap option is still there, but further away. Spot prices start from $0.79 for H100 and $0.99 for B300, which puts the new on-demand rates at about 5.7 and 9.6 times those floors on our arithmetic.
CPU instances move as well. The page lists AMD EPYC Genoa from $0.13 per hour in the column effective 1 October, against from $0.10.
The framework’s starting point: “It is simultaneously the unit of cognition the model produces, the unit of compute the data center serves, the unit of price the lab charges, and the unit of value the enterprise extracts.”
As we read it, a GPU-hour sits underneath the second of those roles: it is what a cloud charges for the compute that produces tokens. When the list price of that hour rises, the cost under every token served on rented on-demand capacity rises with it, before any change in what the lab or the enterprise sees.
What Is Not Established
We read Nebius’s public price page on 8 October 2026. The page does not give a reason for the new prices or say how many customers pay list prices rather than contracted rates, and these are one provider’s list prices, not a market index. We did not contact Nebius.
Business Engineer Framework
Tokenomics: The Economics of AI
A Business Engineer framework on the token as the unit of compute, price and value across the AI stack.
Nebius’s price page lists on-demand prices effective 1 October 2026 that are about 17% to 21% higher across H100, H200, B200 and B300 on our arithmetic, with H100 at $4.50 and B300 at $9.50 per GPU-hour, while RTX PRO 6000 and L40S prices are unchanged.
A note on sourcing. We read Nebius’s public price page on 8 October 2026. The 20 October update rests on Nebius’s pricing documentation and pricing policy, its live spot-price page (read 20 October 2026, about 07:10 UTC), its Q4 2025 and Q2 2026 shareholder letters, and a screenshot posted on X; the seven-day change is the screenshot’s figure, which we could not re-check. Prices are Nebius’s list prices; percentage changes, day costs and ratios are our arithmetic. We did not contact Nebius. Nothing here is a forecast, and nothing here is financial or investment advice.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
