Disclosed terms
The disclosure and our analysis
What changed
Nebius's pricing page now shows two on demand columns: today's rate and a rate effective October 1, 2026. H100 rises 17%, H200 20%, B200 19% and B300 21%. [1]
Spot rates still start at $0.79 for H100 and H200 and $0.99 for B200 and B300. RTX PRO 6000 stays at $1.80. CPU only instances also rise. Nebius still advertises commitment discounts of up to 35% off on demand for multi month cluster reservations. [1]
Nebius's pricing page does not give a reason for the change.
The second increase this year
This is not the first increase. The Motley Fool reports that in early May, Nebius's pricing page showed H100 at $2.95 per hour and B300 at $6.10. [2] Our own catalog recorded H100 at $3.85 on September 18 and B300 at $7.85 on September 24.
If the May figures are right, the October rates are about 53% higher for H100 and 56% higher for B300 than five months earlier. [2] The trend matters more than any single rate. A budget built on spring pricing may already be out of date.
How the new rates compare
We compared the new rates with published on demand rates in our provider catalog, checked September 18 to 28.
H100: At $4.50, Nebius moves above Runpod ($3.49), Crusoe ($3.90) and Lambda ($4.29). It stays below CoreWeave's 8 GPU instance, about $6.16 per GPU.
H200: At $5.40, Nebius moves above Runpod ($4.59) and Verda ($4.55). It stays below Together AI ($5.99) and CoreWeave (about $6.31 per GPU).
B200: At $8.50, Nebius moves well above Lambda ($6.69) and Runpod ($6.79). CoreWeave's 8 GPU instance works out to about $8.60 per GPU.
B300: At $9.50, Nebius moves above Verda ($8.20) and Scaleway ($7.50). Together AI lists $9.99.
Configurations differ. vCPUs, memory, storage, networking and minimum allocation all vary. A lower hourly rate is not a lower project cost on its own.
What to do before October 1
Ask about locking today's rate. Ask Nebius whether a commitment signed before October 1 is priced at the current rate. Get the answer in writing, with the rate and term.
If you run on Nebius on demand today, your bill goes up October 1 unless your contract says otherwise. Check whether your agreement fixes the rate.
If you have a Nebius quote in hand, ask which rate it uses. A quote issued in September may still show the old price.
If you are planning a project, get a dated quote for your actual start date. Ask what a 3, 6 or 12 month commitment costs against the new on demand rate. Ask whether that commitment reserves GPUs or only lowers the rate.
If you are comparing providers, run the comparison again. On H100 and B200 especially, the gap to lower priced providers is now wide enough to justify a second quote.
Get competing quotes
Tell us your GPU, count, region and start date. We will quickly return competing quotes from other GPU clouds, at no cost to you. Our advisory services are paid by the provider you choose.
What remains unknown
Whether existing reserved or committed customers are affected. Whether committed use discounts change along with on demand rates. Whether other GPU clouds follow.
Companies mentioned: Nebius
Sources
- Nebius AI Cloud pricing page · Published September 28, 2026
- The Motley Fool · Published September 27, 2026
Related pricing guides
WORK THROUGH IT WITH AN ADVISOR
Planning around this news? Talk it through.
Bring your workload, launch date and what could change. We can help you evaluate capacity, compare proposals and identify the deployment support you need.
