Nebius pricing

About

Nebius — operating as Nebius AI Cloud under parent Nebius Group N.V. of Amsterdam — is an AI-native GPU cloud, one of the new wave of “neoclouds” competing with hyperscalers and rivals like CoreWeave and Lambda. Its lineage is unusual: Nebius is what remained of the former Yandex N.V. after that company divested its Russian operations in 2024. The international assets — cloud, data centers, and several AI ventures — were renamed Nebius Group and relisted on Nasdaq as NBIS, making Nebius one of the few publicly traded pure-play GPU clouds.

The company rents NVIDIA GPUs by the GPU-hour to AI developers and teams, with a full-stack platform spanning large-scale GPU clusters, managed Kubernetes, storage, and developer tooling, plus a managed inference offering. Nebius has scaled aggressively on the back of the AI-compute shortage: by late May 2026 its market capitalization was roughly $58 billion, and it guided to $3.0-3.4 billion in 2026 revenue, riding the same GPU-scarcity wave that has lifted its neocloud peers.

For current pricing, see Nebius’s pricing page. Note the canonical pricing URL is nebius.com/prices — nebius.com/pricing 404s.

Pricing summary : How Nebius’s pricing model works

Nebius is pure usage-based: you pay per GPU-hour for the instances you launch, with no monthly subscription. The meter runs whenever a GPU is allocated to you. Pricing splits across a few surfaces:

1. On-demand instances — self-serve configs of B300, B200, H200, H100, RTX PRO 6000, and L40S, with list rates published openly per GPU-hour.
2. Preemptible instances — the same GPUs at roughly 45% lower rates (H100 at $2.15 vs $3.85 on-demand), in exchange for the instance being reclaimable when capacity is needed.
3. Explorer Tier — a self-serve entry point: up to 8 GPUs on demand from $1.50/GPU/hr for up to 1,000 GPU-hours/month, with no contract or sales call.
4. Committed clusters / reserved capacity — multi-month and large-scale commitments cut up to 35% off on-demand rates; flagship GB300/GB200 NVL72 racks are sales-quoted.

CPU-only instances start as low as $0.05/hr (Intel Ice Lake) and $0.10/hr (AMD EPYC Genoa), storage is billed separately per GiB/month, and Kubernetes, networking, and public IPs are free.

What makes this different: Nebius pairs a published, hyperscaler-style rate card with two unusual levers — a preemptible tier that undercuts its own on-demand rate by ~45%, and a self-serve Explorer Tier from $1.50/GPU/hr that democratizes access most neoclouds reserve for sales-qualified buyers.

Pricing by product

On-demand and preemptible list prices, per GPU-hour, as of June 2026 (excludes tax):

CPU-only and storage (separate meters):

Sales motions across products: on-demand, preemptible, and the Explorer Tier are fully self-serve (PLG); multi-month and large-cluster commitments and the GB300/GB200 NVL72 racks steer to “contact sales” (sales-led reserved). Kubernetes, networking egress/ingress, and public IPs are free.

Hidden costs : What Nebius users actually pay

Nebius’s headline rates are clean — free networking and Kubernetes remove two of the most unpredictable hyperscaler line items — but real bills include items beyond the GPU-hour:

The biggest real-world cost decisions are the preemptible-vs-on-demand trade-off (a ~45% discount that you pay for in reliability) and storage tier selection — block volumes with 3x mirroring cost roughly 8x more per GiB than object storage. As with peers, capacity for the newest Blackwell GPUs is the practical constraint, with GB300/GB200 NVL72 racks gated behind sales.

Want to estimate your own Nebius bill? Use the Nebius pricing calculator to model your costs based on GPU type and hours.

Pricing evolution : Nebius pricing history and changes

Cadence

Tracked range: 2024 H2–present. Nebius’s primary pricing surface moved to nebius.com/prices; the legacy nebius.com/pricing path now 404s.

Notable changes

• Oct 2024 — Explorer Tier launched at $1.50/GPU/hr self-serve, up to 8 GPUs and 1,000 GPU-hours/month, no contract or sales call — a deliberate “democratize AI compute” move for new customers.
• 2025 — Full per-GPU-hour rate card published across the NVIDIA fleet, including a preemptible tier ~45% below on-demand and multi-month commitment discounts of up to 35% on large clusters.
• June 2026 — Blackwell on the rate card: on-demand B300 $7.85, B200 $7.15, H200 $4.50, H100 $3.85, RTX PRO 6000 $1.80, L40S from $1.55. Preemptible H100 $2.15, B200 $3.95. GB300/GB200 NVL72 racks remain sales-quoted.

The direction of travel mirrors the wider neocloud market: scarce frontier GPUs (Blackwell) carry premium, sales-gated pricing, while the older fleet and the Explorer Tier keep self-serve access cheap to win developer mindshare.

What’s unique : Nebius’s distinctive pricing mechanics

1. A self-serve Explorer Tier from $1.50/GPU/hr. Nebius lets anyone launch up to 8 high-performance GPUs on demand without a contract or sales call — explicitly framed as democratizing access, where most neoclouds gate meaningful GPU counts behind sales.

2. A native preemptible tier ~45% below on-demand. The same H100 is $2.15 preemptible vs $3.85 on-demand, giving batch and fault-tolerant workloads a built-in discount lever rather than a single take-it-or-leave-it rate.

3. Free networking and Kubernetes. By zeroing out egress/ingress, public IPs, and managed Kubernetes, Nebius collapses the bill down to GPU-hours plus storage — removing the line items that make hyperscaler GPU costs hard to forecast.

Strengths & weaknesses

Billing UX : Nebius billing controls and transparency

• Billing controls — Pay-per-GPU-hour on-demand or preemptible; the Explorer Tier caps self-serve usage at 1,000 GPU-hours/month from $1.50/GPU/hr. Multi-month and large-cluster commitments lock up to 35% off.
• Usage visibility — Self-serve console manages instances, clusters, and attached storage; billing starts only after a $25 minimum card deposit enables the account.
• Payment options — Self-serve card checkout for on-demand, preemptible, and Explorer usage; sales-led contracts and invoicing for committed clusters, reserved capacity, and the GB300/GB200 NVL72 racks.

Strategic wins : Why Nebius’s pricing decisions worked

1. Democratizing access with a self-serve Explorer Tier

By letting anyone launch up to 8 GPUs from $1.50/GPU/hr with no sales call, Nebius turned self-serve access into a developer-acquisition wedge — capturing the long tail of builders that sales-gated neoclouds ignore. See how AI companies structure pricing.

2. A preemptible lever instead of a single rate

Offering preemptible GPUs ~45% below on-demand lets price-sensitive batch workloads self-select into a cheaper tier without Nebius discounting its premium on-demand rate. Related: outcome-based pricing trends.

3. Committed clusters layered over self-serve

Up to 35% off for multi-month and large-cluster commitments converts spiky self-serve demand into predictable, forecastable capacity — smoothing planning for both sides. See choosing the right usage metric.

Areas to improve : Gaps in Nebius’s pricing approach

1. No free trial raises the activation bar

Requiring a $25 card deposit before any GPU runs is friction for the very developers the Explorer Tier courts. A modest free credit (as rivals offer) would lower the first-launch barrier. See bill shock and cost unpredictability.

2. Storage tiers are easy to overspend

Block volumes with 3x mirroring cost roughly 8x more per GiB than object storage, and the right choice isn’t obvious at launch. Clearer in-console guidance and defaults would reduce surprise storage bills.

3. Newest GPUs sit behind sales

GB300/GB200 NVL72 racks are sales-quoted, so the self-serve transparency that defines the rest of the rate card stops at the frontier. Publishing even indicative rack pricing would extend Nebius’s transparency edge.

Key takeaways

1. Nebius is pure per-GPU-hour usage pricing — on-demand, preemptible, and a self-serve Explorer Tier — with commitment discounts layered on top. For the underlying model, see the introduction to usage-based pricing.
2. The preemptible tier is a built-in discount lever (H100 $2.15 vs $3.85), letting fault-tolerant workloads trade reliability for ~45% savings.
3. The Explorer Tier democratizes access from $1.50/GPU/hr self-serve — a developer-acquisition wedge most neoclouds skip.
4. Free networking and Kubernetes collapse the bill to GPU-hours plus storage, with the $25 deposit and storage-tier choice as the main hidden costs.
5. As a public company (NBIS), Nebius is unusually transparent — published rates and a $3.0-3.4B 2026 revenue guide — even as frontier Blackwell capacity stays sales-gated.

UBP implications

1. A second “preemptible” price point segments demand cleanly. Offering the same resource at two reliability tiers lets price-sensitive buyers self-select without eroding the premium rate — a reusable pattern for any capacity-constrained usage business.
2. Self-serve access can be a growth lever, not a discount. The Explorer Tier widens the top of the funnel at a published rate rather than via negotiated discounts, trading sales friction for volume.
3. Zeroing out ancillary meters builds trust. Free networking and Kubernetes make the bill predictable; removing unpredictable line items is itself a pricing decision that lowers buyer friction.

Sources

• Nebius AI Cloud pricing (accessed 2026-06-15)
• Nebius Explorer Tier (accessed 2026-06-15)
• Nebius Explorer Tier press release — self-service GPUs from $1.50/GPU-hr (accessed 2026-06-15)
• Nebius Group (NBIS) stock overview (accessed 2026-06-15)
• Captured pricing page text — public/images/blueprint/nebius-ai/2026-06-15-main-validated.txt (nebius.com/pricing 404s; nebius.com/prices used) (accessed 2026-06-15)

Bottom line

Nebius (NBIS) is a publicly traded, pure usage-based GPU cloud born from the international remnant of Yandex: a transparent per-GPU-hour rate card with on-demand and ~45%-cheaper preemptible tiers, a self-serve Explorer Tier from $1.50/GPU/hr, and up to 35% off for committed clusters. Free networking and Kubernetes keep the bill predictable, while the real cost decisions are the preemptible-vs-on-demand trade-off, storage tier choice, and the $25 deposit that replaces a free trial. Like its neocloud peers, the frontier Blackwell racks stay sales-gated. Browse the pricing blueprint for more fully-researched company profiles, or compare Nebius against other Infrastructure, Compute & MLOps companies.