Scaleway GPU Pricing 2026: H100 Cost vs Spheron

Scaleway GPU pricing starts with an H100 PCIe rate of roughly $3.19-3.22/hr per GPU, which puts it among the cheapest openly published H100 rates anywhere in Europe. That number is real and it's live on Scaleway's own pricing page today. What it doesn't tell you is that the capacity backing it lives in exactly two regions, that the SXM tier most training workloads actually want is confined to a single availability zone, and that the SecNumCloud qualification Scaleway's marketing leans on is still in progress, not granted. This post prices the hardware, scopes the availability and compliance claims, and compares the total cost against Spheron's live rates.

TL;DR: Scaleway GPU Pricing vs Spheron (Nov 2026)

• Price: Scaleway H100 PCIe (H100-1-80G) runs about $3.19-3.22/hr per GPU as of 3 Nov 2026, per independent trackers GPU Finder and GPU Tracker.
• SXM tier: Scaleway's H100 SXM node pricing actually drops with scale, from about $3.55/GPU-hr on a 2-GPU node to about $3.40/GPU-hr on an 8-GPU node, roughly a 4% discount PCIe configs don't get.
• Regions: H100 PCIe, L40S, and L4 run in Paris and Warsaw only, per Scaleway's own product availability documentation. H100 SXM and B300 SXM run in a single Paris availability zone and nowhere else.
• Compliance: Scaleway is still "undergoing SecNumCloud (ANSSI) qualification... not yet granted," per its own security and compliance page, not a completed certification.
• Spheron: H100 SXM5 runs $2.98/hr on-demand and $2.20/hr spot as of 06 Nov 2026, aggregated across 5+ providers with no region lock-in. See the GPU cloud pricing comparison for rates across more providers.

Scaleway GPU Pricing: H100 PCIe and SXM Rates, Region by Region

Scaleway sells H100 capacity in two distinct shapes: a PCIe tier at one or two GPUs per instance, and an NVLink-connected SXM tier that scales from two GPUs up to a full 8-GPU node. They're priced differently, and only one of them gets cheaper as you add GPUs. Scaleway isn't the only openly-priced sovereign-cloud option in Europe either; our OVHcloud GPU pricing breakdown prices the other one.

H100 PCIe. The single-GPU H100-1-80G lists at $3.19/GPU-hr per GPU Finder's Scaleway pricing page, with GPU Tracker pricing the same instance at $3.2206/hr. The gap is capture-date noise between trackers, not a real pricing discrepancy, so $3.19-3.22/hr is the working range. The two-GPU PCIe config holds flat at the same per-GPU rate, roughly $6.38-6.44/hr total for both GPUs, per the same sources. There's no bulk discount for buying the bigger PCIe box.

H100 SXM. This is where Scaleway's pricing actually moves with scale. Per GPU Tracker's listing, the 2-GPU SXM node runs $7.0994/hr total ($3.5497/GPU), the 4-GPU node runs $13.6963/hr ($3.4241/GPU), and the full 8-GPU node runs $27.1661/hr ($3.3958/GPU). That's about a 4.3% per-GPU discount from the 2-GPU config to the 8-GPU config, a real if modest volume incentive that the PCIe tier doesn't offer.

Everything else in the catalog. L4 runs about $0.88/GPU-hr (GPU Finder lists $0.8848, GPU Tracker $0.8848, matching), L40S runs about $1.64/GPU-hr per GPU Finder, and the newer B300 SXM 8-GPU node lists at $70.782/hr total, or $8.8478/GPU, per GPU Tracker. Scaleway also charges $0.011/GB for network egress on its GPU instances, per GPU Finder, a real line item most published per-hour rates leave out of the headline number.

None of these configs run on a single flat region, which is the part a rate card alone won't show you.

The Catch: Availability and Stockouts in Paris and Warsaw

Here's the part the price card doesn't say. Per Scaleway's own product availability documentation, H100 PCIe, L40S, and L4 GPU Instances are available in exactly two regions: Paris (availability zone PAR-2, plus PAR-1 for L4) and Warsaw (availability zone WAW-2). Amsterdam and Milan, two of Scaleway's other compute regions, carry none of these GPU products at all.

It gets narrower at the top of the catalog. H100 SXM and the newer B300 SXM instances run in a single Paris availability zone, PAR-2, and nowhere else per that same documentation. No Warsaw failover, no Amsterdam overflow. If PAR-2 is out of SXM capacity on the day you need it, there's no second EU region to check on Scaleway's own platform.

That concentration is exactly the kind of setup that produces stockouts rather than a graceful queue. GPU Finder's own stock tracker for Scaleway shows only 9 of the 15 GPU configurations it monitors have fresh availability data; the one SKU with a clean, fully-available track record over the past month is the older P100, not the H100 or SXM tiers most AI teams are actually trying to book. A single-zone SKU with inconsistent stock tracking is a very different buying proposition than "GPU capacity in Europe," even when the headline rate is genuinely competitive.

Scaleway vs Spheron: Cost at Equivalent H100 Configurations

Here's the direct comparison, holding Scaleway's published rates against Spheron's live pricing as of 06 Nov 2026.

Scaleway's PCIe rate and Spheron's on-demand rate are close enough that the deciding factor usually isn't the sticker price, it's which one actually has a GPU free in the region you need on the day you need it. Spheron spreads that risk across 5+ providers and multiple regions instead of two; Scaleway concentrates it into Paris and Warsaw, with the SXM tier narrowed to one zone.

Pricing fluctuates based on GPU availability. Scaleway figures above are from GPU Finder and GPU Tracker as of 3 Nov 2026 and may have changed; Spheron rates are live as of 06 Nov 2026. Check current GPU pricing → for live rates, and verify Scaleway's numbers directly on its own pricing page before budgeting.

When Scaleway's EU-Only Footprint Is Worth the Tradeoff

Scaleway earns its place in a European GPU shortlist. Per its own security and compliance page, it's already HDS-certified for health data, and it's one of the few openly-priced H100 providers running entirely out of French and Polish data centers rather than a US-incorporated hyperscaler region. For a team whose only hard requirement is non-US-incorporated EU infrastructure, that's a real, currently-priced option, not a promise.

Where it stops being the obvious pick is the SecNumCloud claim specifically. Scaleway's own compliance page says the qualification is "not yet granted" and that the company is "engaged in this demanding process," with no committed date and no published scope of which services or datacenters it will eventually cover. If your procurement checklist says "SecNumCloud-qualified GPU workload" rather than "EU-incorporated provider," Scaleway doesn't clear that bar today. OVHcloud, the other sovereign-cloud option mentioned above, runs into the identical gap: a genuine EU sovereignty story that hasn't yet extended SecNumCloud qualification down to the actual GPU rental SKU.

The regional concentration is the other half of the tradeoff. If your job needs a single H100 PCIe GPU and Paris or Warsaw has stock, Scaleway's rate is hard to beat on paper. If it needs an 8-GPU SXM node, that demand funnels into one availability zone with no regional backup, which is a stockout risk a flat rate card doesn't disclose. For workloads where provider and region flexibility matter more than a single EU-sovereign checkbox, Spheron's aggregated marketplace, spot tier, and per-minute billing cover more ground; our best GPU for LLM training guide and GPU cloud benchmarks are good next stops for sizing the actual job before you lock in a region.

Teams coming from a hyperscaler quota queue, where Azure's ND H100 v5 pricing sits well above either Scaleway or Spheron, will find both of these EU options cheaper outright. Teams coming from the marketplace side, comparing Runpod against Vast.ai, should weigh Scaleway's fixed list price and narrow footprint against a marketplace's wider but more variable stock.

If a two-region EU footprint and an in-progress compliance certification don't match what your workload actually needs, Spheron's H100 on-demand and spot rates, aggregated across 5+ providers with no region lock-in, are worth pricing out first.

Check H100 availability on Spheron →