Scaleway H100 Pricing 2026: Sovereign AI GPU Cost vs Spheron

Scaleway is a real, current H100 provider, not a footnote in a European regions table. Its on-demand H100 SXM rate runs $3.17-3.31/hr per GPU out of Paris and Warsaw, and it backs that price with a sovereign-AI pitch few other providers can match: French incorporation outside CLOUD Act jurisdiction, an in-progress SecNumCloud qualification, and production workloads from Mistral AI already running on its infrastructure. Our own European GPU cloud roundup covers the wider regulatory landscape but never priced Scaleway out on its own. This post does, and it does not dodge the parts where Scaleway beats us.

We pulled Scaleway's numbers from independent pricing trackers and Scaleway's own product pages, and pulled Spheron's numbers live from our current GPU pricing API on 20 July 2026. Every dollar figure below is dated and sourced so you can verify it yourself before budgeting.

TL;DR: Scaleway H100 Pricing vs Spheron (20 Jul 2026)

Pricing fluctuates based on GPU availability. The prices above are based on 20 Jul 2026 and may have changed. Check current GPU pricing → for live rates. Scaleway figures are cross-referenced from independent pricing trackers and Scaleway's own product pages; verify directly with Scaleway before budgeting.

The short version: if your workload has to stay on non-US-incorporated infrastructure inside the EU, Scaleway's SXM rate is genuinely competitive, and right now it beats Spheron's EU on-demand SXM floor. If you need PCIe-class H100 access, a single GPU, spot pricing, or capacity outside Paris and Warsaw, Spheron comes out ahead. For the wider Blackwell pricing picture across providers, see our B200 cloud pricing comparison.

Scaleway H100 and B300 Pricing in 2026

Scaleway sells H100 capacity through two connection types out of its Paris data center, with a second region for the smaller configs. It is a narrower catalog than a hyperscaler's, but every SKU is priced and billed by the minute with no reservation contract required.

H100 PCIe and SXM Instance Configs (1x/2x/4x/8x GPU)

Scaleway's H100 lineup splits into PCIe and SXM instance families, per Scaleway's H100 SXM product page:

Every SXM config carries 80 GB of HBM3 per GPU at 3.35 TB/s of memory bandwidth, and CPU allocation scales with the node: 32 to 128 vCPUs on Xeon Platinum 8452Y chips and 240 to 960 GB of DDR5 RAM depending on GPU count, per the same product page. Warsaw (WAZ2) currently only carries the smaller PCIe configs; if you need the 4x or 8x SXM node, that capacity is Paris-only.

Per-GPU Cost Breakdown and Bulk-Discount Curve

Here is where the SXM lineup gets interesting. Per-GPU price actually drops as the node gets bigger, the opposite of what you'd expect from a scarce-GPU market:

Figures per independent tracker gpucompare.com. The pattern holds across the SXM tier: buying the full 8-GPU node is about 4% cheaper per GPU than buying a 2-GPU SXM slice. That's a real, if modest, bulk-discount curve, not a rounding artifact. The PCIe 2x config actually undercuts every SXM config on a per-GPU basis, so if your workload doesn't need NVLink-class GPU-to-GPU bandwidth, PCIe is the cheaper door in.

HGX B300 and L40S/L4 Pricing for Context

Scaleway also lists Blackwell Ultra B300 capacity and older-generation inference GPUs. HGX B300 (288 GB) runs around $10.84/hr, per independent tracker computeprices.com, last updated 19 July 2026. L40S runs about $1.47-1.68/hr and L4 about $0.79-0.90/hr depending on config, per the low end from gpucompare.com and the high end from computeprices.com. For a full breakdown of Blackwell-class pricing across the market, including where B200 and B300 land on Spheron, see our NVIDIA B200 cloud pricing comparison.

The Sovereign AI Pitch: EU Data Residency and GDPR vs US Hyperscalers

Scaleway markets its H100 capacity explicitly for sovereign AI: keeping proprietary models and training data under EU jurisdiction and immune to extraterritorial regulation, per its own product positioning. That's a claim worth pressure-testing rather than taking at face value, so here's what's actually verifiable.

SecNumCloud Qualification: What It Is and Where Scaleway Stands

SecNumCloud is one of the world's most stringent cloud security and sovereignty qualifications, run by ANSSI, France's national cybersecurity agency. A provider that holds it has demonstrated its offering does not expose customer data to extraterritorial laws, per Scaleway's own announcement.

Scaleway announced in September 2025 that it had begun the qualification process and passed the initial J0 milestone, with a target of completing qualification by the end of 2025. It already holds ISO 27001 and HDS (French healthcare data hosting) certifications while that process runs.

"Security is crucial for a cloud provider. That's why at Scaleway we've made it the absolute priority for our development and operations teams. By starting the SecNumCloud qualification process we're also reaffirming a commitment that goes above and beyond certification," said Damien Lucas, CEO of Scaleway, in the announcement.

Treat SecNumCloud as in-progress rather than granted. If your compliance requirement specifically demands a completed SecNumCloud qualification today, confirm current status directly with Scaleway before committing a workload to it. For the broader compliance framework EU teams navigate alongside data residency, see our EU AI Act compliance guide for GPU cloud.

CLOUD Act Exposure: Why Non-US Incorporation Matters

The US CLOUD Act lets US law enforcement compel US-incorporated cloud providers to produce data stored anywhere in the world, including EU data centers, without notifying the EU government or the data subject. The determining factor is legal incorporation, not physical server location, which is why running workloads in AWS's Frankfurt region or Azure's West Europe region does not remove CLOUD Act exposure, as we detail in our Europe GPU cloud guide.

Scaleway, as a French-incorporated company, falls outside that jurisdiction. That's a structural difference from AWS, Azure, and GCP EU regions, not a policy promise Scaleway is making on top of the same underlying legal exposure. For teams processing GDPR Article 9 special category data (health, biometric, or similarly sensitive records) where CLOUD Act compulsion would itself create a compliance breach, that distinction is the whole ballgame.

Who Actually Runs on Scaleway: Mistral AI and NVIDIA DGX Cloud Lepton

Sovereignty claims are easy to make and hard to prove. Scaleway has real production workloads to point to. Mistral AI has run on Scaleway infrastructure, and Scaleway's own announcement of its NVIDIA DGX Cloud Lepton selection lists Mistral AI's Mixtral, Kyutai's Moshi, and H Company's Holo 1 as models trained or served on its infrastructure.

NVIDIA selected Scaleway to bring European sovereign compute to DGX Cloud Lepton on June 12, 2025, one of eight EU-based providers NVIDIA added at GTC Paris alongside Mistral AI, Nebius, Nscale, Firebird, FluidStack, Hydra Host, and Together AI, per Constellation Research's coverage of the announcement. "Being chosen as part of DGX Cloud Lepton confirms Scaleway's role as a trusted provider of sovereign, high-performance AI infrastructure," Lucas said in Scaleway's own announcement. For the broader picture of what DGX Cloud Lepton is and how its marketplace pricing works, see our DGX Cloud Lepton pricing breakdown.

Scaleway vs Spheron: Cost and Availability Compared

Here's the part we'd rather not sugarcoat: Scaleway wins on price for one specific configuration, and we should say so plainly instead of burying it.

On-Demand H100 Rate Comparison (Scaleway EU vs Spheron Global Marketplace)

We pulled this straight from Spheron's live pricing API on 20 July 2026, splitting every offer by instanceType before computing anything (DEDICATED for on-demand, SPOT for spot) so we don't accidentally mislabel a spot rate as an on-demand one. Spheron's current H100 SXM5 on-demand floor is $3.92/hr per GPU, from an 8-GPU node hosted with Data-Crunch in Finland, one of the 5+ providers Spheron aggregates. That's higher than Scaleway's $3.17-3.31/hr per-GPU SXM rate for equivalent EU-based capacity.

On PCIe and NVL, Spheron is cheaper across the board. On guaranteed, on-demand SXM capacity in an EU region specifically, Scaleway currently has the better rate. That's a narrower win than a blanket "Scaleway is cheaper" claim, and it's the accurate one.

Where Scaleway Wins (EU-Only Regulated Workloads, French Sovereignty)

• Regulated data that cannot leave French or EU jurisdiction under any circumstances. Scaleway's non-US incorporation removes CLOUD Act exposure entirely, which no US-incorporated provider, including hyperscalers with EU regions, can offer.
• Guaranteed on-demand H100 SXM capacity in the EU right now. At $3.17-3.31/hr per GPU, Scaleway's SXM rate beats Spheron's current EU on-demand SXM5 floor.
• A completed compliance paper trail once SecNumCloud finishes. ISO 27001 and HDS are already in place; SecNumCloud, once granted, adds a qualification most competitors, including hyperscaler EU regions, don't hold at all.
• Renewable-powered infrastructure. Scaleway's Paris DC5 facility runs on renewable energy with adiabatic cooling and a PUE of 1.16, per Scaleway's own H100 product page, relevant if your procurement process weighs data center efficiency.

Where Spheron Wins (Per-GPU Flexibility, Broader Selection, Spot Pricing)

• PCIe and NVL pricing. Spheron's H100 PCIe floor ($2.01/hr per GPU) and H100 NVL rate ($2.06/hr per GPU) both undercut Scaleway's PCIe rate by a wide margin.
• Spot pricing. Scaleway has no spot tier. Spheron's H100 SXM5 spot rate runs $2.91/hr per GPU, well under even Scaleway's SXM on-demand floor, for workloads that can tolerate reclamation.
• Region and provider breadth. Scaleway is limited to Paris and Warsaw. Spheron aggregates capacity from 5+ providers across a much wider set of regions, which matters if your workload isn't bound to EU-only placement.
• No fixed-node minimums beyond the GPU itself. Both providers let you rent a single GPU, but Spheron's broader marketplace gives you more configurations to choose from at that single-GPU tier.

If your workload doesn't carry a hard EU-sovereignty requirement, Azure's H100 pricing is a useful third data point: both Scaleway and Spheron undercut Azure's on-demand H100 rate by a wide margin, which is the more common comparison most teams outside strict compliance scopes actually need to make. For deployment and API details on Spheron, see the docs.

FAQ

How much does Scaleway charge for H100 GPUs?

H100 PCIe (2x GPU) runs about $2.87/hr per GPU in Paris and Warsaw. H100 SXM runs $3.31/hr per GPU on a 2x node, down to $3.17/hr per GPU on the full 8x node, Paris-only.

Is Scaleway SecNumCloud certified?

Not yet. Scaleway announced in September 2025 that it had started the process with ANSSI, passed the J0 milestone, and was targeting completion by the end of 2025. It already holds ISO 27001 and HDS certifications.

Does Scaleway fall under the US CLOUD Act?

No. Scaleway is French-incorporated, which puts it outside CLOUD Act jurisdiction, unlike AWS, Azure, and GCP EU regions, which remain US-incorporated entities regardless of where the servers physically sit.

Is Scaleway cheaper than Spheron for H100 GPUs?

For SXM configs, currently yes: Scaleway's $3.17-3.31/hr per GPU beats Spheron's $3.92/hr EU on-demand SXM floor. For PCIe and NVL, Spheron is cheaper, and Spheron also offers spot pricing Scaleway doesn't have.

If EU sovereignty isn't a hard requirement for your workload, Spheron's H100 pricing across PCIe, SXM, and NVL configs, plus spot capacity, is worth checking before you commit to a single-region provider.

H100 GPU pricing on Spheron →