CoreWeave vs Nebius: prices and Q2 results, Sep 2026
Nebius lists lower H100, H200 and B200 on-demand GPU-hour prices in the 28 September check, including its separately listed October rates. CoreWeave reports greater Q2 revenue; revenue backlog and customer commitments are not equivalent measures. Matched quotes are needed for committed usage.
Nebius lists cheaper H100, H200 and B200 GPU-hours than CoreWeave: H100 is $3.85 versus a normalized $6.16, checked 28 September 2026.1,2 CoreWeave reports greater Q2 2026 revenue, at $2,575 million versus Nebius Group’s $582.3 million.3,4 Their contract-demand disclosures need separate labels: CoreWeave reports revenue backlog; Nebius reports customer commitments.3,4
TL;DR
- Nebius also lists higher rates effective 1 October 2026, though each remains below CoreWeave’s checked on-demand equivalent.1,2
- CoreWeave’s GPU-hour equivalents come from eight-GPU node prices.1 A normalized rate doesn’t establish a general single-GPU rental option.1
- CoreWeave’s approximately $104 billion revenue backlog includes estimated future revenue under committed contracts, subject to delivery and service availability.3
- Both companies advertise commitment discounts.1,2 Compare matched quotes before treating the on-demand price order as a contract-price verdict.
How were the prices and results compared?
CoreWeave and Nebius are compared using their official price pages and Q2 2026 disclosures, checked 28 September 2026.1,2,3,4 Financial results cover the quarter ended 30 June 2026.3,5 GPU prices retain the listed dollar amounts and normalize CoreWeave’s node-hour unit to GPU-hours; this does not establish matched capacity or workload cost.
Nebius explicitly accepts USD and excludes applicable taxes.2 For CoreWeave’s quoted dollar prices, the comparison’s USD convention is an assumption where the captured page does not state the currency; tax treatment is not established. No currency conversion or customer tax calculation is applied.
The subject sits within compute and AI infrastructure research. The billing comparison divides CoreWeave’s listed node-hour price by its GPU count and retains Nebius’s stated GPU-hour unit.1,2 Our research methodology provides the broader evidence standard.
Which company lists cheaper H100, H200 and B200 GPUs?
Nebius lists lower on-demand prices than CoreWeave for H100, H200 and B200, both at the checked September rates and in its separately listed rates effective 1 October 2026.1,2
| GPU | CoreWeave normalized dollar figure/GPU-hour; tax treatment unconfirmed | Nebius USD/GPU-hour, before tax, checked 28 September 2026 | Nebius listed USD/GPU-hour, before tax, effective 1 October 2026 |
|---|---|---|---|
| HGX H100 | $6.161 | $3.852 | $4.502 |
| HGX H200 | $6.311 | $4.502 | $5.402 |
| HGX B200 | $8.601 | $7.152 | $8.502 |
CoreWeave lists eight-GPU nodes at $49.24 an hour for H100, $50.44 for H200 and $68.80 for B200.1 The GPU-hour figures are node-price equivalents rounded to cents, with half-cent ties rounded up: H200’s exact $50.44 ÷ 8 = $6.305 displays as $6.31.1 The node-hour amounts remain the listed billing basis. CoreWeave’s separate GPU-based inference pricing applies exclusively to its inference customers.1
Keep the two Nebius date columns separate. Its captured price page labels the second set “Effective October 1, 2026”; the September check doesn’t establish a later transaction price or a customer’s negotiated rate.2 The lower listed rate is a procurement starting point, with neither page establishing which company completes your workload at lower total cost.1,2
What did CoreWeave and Nebius report in Q2 2026?
CoreWeave reported Q2 2026 revenue of $2,575 million, compared with Nebius Group’s $582.3 million for the same quarter.3,4 CoreWeave therefore has the larger reported quarterly revenue base in this comparison.3,4
| Q2 2026 revenue, USD millions | CoreWeave | Nebius |
|---|---|---|
| Company/group revenue | $2,5753 | $582.34 |
| Nebius AI cloud segment revenue | Not a corresponding CoreWeave segment comparison | $574.94 |
Nebius’s group and cloud-segment figures answer different questions.4 The group includes other businesses; Nebius reports that its AI cloud accounted for approximately 98% of group revenue in the quarter.4,5 Use group revenue for the company-level comparison and the separately reported cloud figure when discussing Nebius’s GPU business.
Nebius also reported annualized run-rate revenue of $3.0 billion at the end of June.4 That measure has a different period basis from quarterly revenue; substituting it for the $582.3 million quarter would distort the comparison.4
Can their reported backlog figures be compared directly?
CoreWeave’s revenue backlog and Nebius’s customer commitments are different disclosed measures, so they don’t support a like-for-like backlog ratio.3,4 CoreWeave reported approximately $104 billion at 30 June 2026; Nebius’s Q2 shareholder letter reported more than $40 billion of customer commitments without presenting that figure as an equivalent quarter-end revenue-backlog measure.3,4
CoreWeave defines revenue backlog as remaining performance obligations plus other amounts it estimates will become revenue under committed customer contracts.3 Recognition depends on satisfying delivery and service-availability requirements.3 Its quarter-end figure excludes more than $25 billion of net new customer commitments added in early Q3.3 Keep those later commitments outside the Q2 figure.
Nebius’s commitment statement appears in its financing discussion alongside a secured financing transaction completed in July.4 The letter also says the majority of its Q2 deals were signed for delivery arriving in late 2026 and would contribute primarily to 2027 revenue.4 Contract value and earned quarterly revenue therefore describe different stages of the business.3,4
Retain each company’s exact measure and date. Calling both figures “backlog” would erase the definition and timing differences the disclosures leave unresolved.3,4
How do the contract models change the price comparison?
CoreWeave and Nebius both publish on-demand prices and offer discounts for committed usage, so their list-price comparison cannot determine the cheaper negotiated contract.1,2 CoreWeave advertises discounts of up to 60% over on-demand prices; Nebius advertises up to 35% for reserving large-scale clusters for multiple months.1,2 Those are advertised ceilings, not matched offers for the same order.1,2
Nebius’s Q2 letter describes three contract types:4
- Shorter-duration agreements, typically three to six months, for time-bound needs and priced at a premium.4 Its first such deal was signed in Q3.4
- Mid-term agreements, described as its core business, with average durations of one to three years.4
- Long-term agreements with investment-grade customers, which management says support financing.4
Nebius also reported that roughly 70% of deals closed in Q2 included customer prepayments.4 A buyer evaluating a commitment should request the payment schedule alongside the hourly quote. CoreWeave’s captured pricing page advertises reserved compute but supplies no equivalent duration breakdown.1
Request quotes for the same GPU, region, GPU count and term. Specify the required start date, payment schedule and storage charges. Treat maximum advertised discounts as questions to resolve in those quotes, not inputs for an assumed saving.
What should a GPU buyer decide from this comparison?
Glamdring Research’s verdict is to start a list-price-led shortlist with Nebius: it lists lower H100, H200 and B200 GPU-hour rates than CoreWeave in the checked comparison.1,2 For committed usage, request matched quotes from both companies because each advertises discounts whose applicability depends on the order.1,2
Keep company size separate from the rental decision. CoreWeave’s greater reported revenue establishes a larger revenue base; the financial disclosures don’t test your workload’s speed, availability or total bill.3,4 Compare the same workload under the proposed terms before choosing on cost.
For a wider shortlist, use the H100 price comparison by company and the B200 price comparison by company. The decision changes when a matched quote or workload test overturns the listed-price advantage.
Frequently asked questions
Can I rent one CoreWeave GPU at the normalized price?
The normalized node price alone doesn’t establish that option.1 CoreWeave lists eight-GPU H100, H200 and B200 nodes, while its separate GPU-based pricing is restricted to inference customers.1 Confirm the purchasable configuration and applicable rate before budgeting for a single GPU.
Are spot prices fixed alternatives to on-demand rates?
Nebius’s spot prices are dynamic: its price page says they can change with available supply, utilization and demand during active use.2 CoreWeave also lists spot prices in a separate column from on-demand prices.1 The on-demand comparison therefore doesn’t establish a fixed spot-price winner.1,2 Ask about the relevant interruption terms before using spot for a workload that needs uninterrupted service.
Can companies pay Nebius by bank transfer?
Companies can pay Nebius by bank transfer.2 Both individuals and companies can pay for resources by bank card.2 Companies can also change their payment method by contacting support.2
Sources checked
Company-owned pages establish what a company says. They do not prove a market conclusion. Each source is dated so readers can judge each claim.
