Framework v1.0
How Compute Compass structures provider shortlists.
The framework below is the structure Compute Compass uses to make EU GPU provider shortlists consistent, transparent, and easier to discuss with technical and procurement teams. It is not a certification system and does not guarantee provider performance or legal compliance.
Tracking 25+ providers across France, Germany, the Netherlands, Italy, the Nordics, Central/Eastern Europe.
Ranking is not for sale. Providers cannot pay to alter shortlist position.
Structured comparison criteria
What the framework considers
| Criterion | What it captures | Example inputs |
|---|---|---|
Workload fit | Whether the provider's infrastructure matches the buyer's technical workload. | GPU model, memory, interconnect, batch size, inference vs training, multi-node needs. |
Total cost | Published on-demand rates used as a directional filter. Reserved pricing negotiated directly with providers. | Published on-demand GPU-hour rate, published storage and egress costs, support tier, stated commitment periods. Reserved pricing and volume discounts are marked "available on request" and not scored. |
Deployment risk | How likely the buyer is to deploy successfully within the required timeline. Includes capacity availability and lead time. | Capacity availability, quoted lead time, onboarding complexity, contract flexibility, region availability, reserved vs on-demand inventory. |
Compliance and data residency | Comparable information relevant to governance, data location, and procurement review. | Hosting jurisdiction, DPA availability, subprocessors, data-handling terms, certifications referenced by the provider, and EU Data Act Chapter VI switching clauses in standard contracts. |
Operational maturity | Day-two reliability and support signals after initial deployment. | Support model, SLA terms, incident transparency, status page, customer references, API, CLI, Terraform, Kubernetes, Slurm. |
What counts as EU for shortlisting purposes: providers with operations and parent ownership in EU member states are first priority. EEA members (Norway, Iceland, Liechtenstein) are treated as equivalent for data-protection purposes since GDPR applies in full. UK and Swiss providers are flagged separately as third-country jurisdictions with adequacy decisions: lawful for most workloads but evaluated on a per-buyer basis. Non-EU/EEA parent ownership is disclosed in every shortlist alongside the relevant provider.
Pricing
How we handle pricing
GPU pricing on Compute Compass reflects published on-demand rates sourced from provider pricing pages and provider-disclosed data. These rates are indicative. Most serious GPU contracts involve reserved capacity, volume commitments, or negotiated discounts that differ materially from published rates. For this reason, Compute Compass uses pricing as a hard-filter signal to identify providers outside your stated budget envelope, not as a ranking criterion. We do not claim to show your actual contract price. Every shortlist marks cost inputs as "published on-demand rate; reserved pricing available on request." To understand your real cost, request a quote directly from the shortlisted providers. We surface comparable cost structure so you know which line items to ask about.
Weighting
How weighting works
Compute Compass does not rank providers using a one-size-fits-all score. AI infrastructure buyers care about different things: some need the lowest viable cost, others need EU compliance, fast availability, specific hardware, or the lowest deployment risk.
That is why each match starts with your priorities. You either select a buyer archetype in the intake form or rank the criteria yourself. Compute Compass then converts that priority order into transparent weights and applies the same scoring method across all providers.
Priority tiers
How priorities translate to weights
Compute Compass uses a four-tier priority allocation. For each criterion the buyer assigns Critical, Important, Nice to have, or Not relevant. Each tier carries a fixed point value:
- Critical = 4 points
- Important = 2 points
- Nice to have = 1 point
- Not relevant = 0 points
The weight for each criterion equals its points divided by the total points assigned across all criteria, normalised to 100%. This produces transparent percentages that appear on every delivered shortlist.
A buyer who marks one criterion Critical, two Important, and two Nice to have receives weights of 40% / 20% / 20% / 10% / 10% in the order they chose. A buyer who marks all five criteria as Important receives 20% on each. Critical is capped at two criteria, so it always functions as a genuine differentiator rather than a label applied to everything, and the weighting always reflects a real trade-off.
Buyers can mark up to two criteria as deal-breakers. A deal-breaker criterion is one where a low provider score signals unacceptable fit, not just a lower ranking. Providers scoring 2 or below on a deal-breaker criterion are eliminated from the shortlist entirely rather than ranked at the bottom. This prevents a structurally weak provider from appearing in a shortlist just because it scored well on other dimensions. Deal-breakers are optional, set during the buyer intake, and documented in the delivered shortlist alongside the weights.
Archetypes pre-fill these tiers for common buyer profiles, so the buyer does not start from a blank form. Every buyer can adjust the pre-fill before submission.
Archetype rankings
Archetypes apply a predefined ranking order. The weights shown above stay the same, only the order changes.
Each directional archetype pre-fills one criterion as Critical, two as Important, and two as Nice to have, producing weights of 40% / 20% / 20% / 10% / 10% applied in the order shown. The Balanced archetype pre-fills all five as Important, producing 20% weight on each criterion. Custom buyers set their own tier choices in the intake form.
Cost-led
Total cost, Deployment risk, Workload fit, Operational maturity, Compliance and data residency.
Teams where budget is the primary constraint and cost is the first filter.
Compliance-led
Compliance and data residency, Operational maturity, Deployment risk, Workload fit, Total cost.
Teams where data residency, certifications, and governance are non-negotiable.
Performance-led
Workload fit, Deployment risk, Operational maturity, Compliance and data residency, Total cost.
Teams where GPU type, throughput, latency, or framework compatibility are non-negotiable.
Deployment-speed-led
Deployment risk, Workload fit, Total cost, Operational maturity, Compliance and data residency.
Teams where confirming capacity quickly is the top priority.
Operational-maturity-led
Operational maturity, Compliance and data residency, Deployment risk, Workload fit, Total cost.
Teams where support, SLA, reliability, and tooling matter most.
Balanced
Workload fit, Total cost, Deployment risk, Compliance and data residency, Operational maturity.
Teams that want a balanced view across all five criteria.
Custom ranking available. Set your own priority order in the intake form.
Boundaries
- — We structure provider comparisons around buyer workload requirements and disclose the score breakdown behind every shortlist.
- — We use GDPR-aware and EU data-residency focused comparison inputs.
- — We do not certify legal compliance or guarantee GDPR or AI Act outcomes for any service.
- — We do not present paid relationships as ranking signals.
Framework status
A living framework.
The framework reflects current best practice for EU GPU provider comparison. It will evolve as Compute Compass completes more deployment cycles and incorporates buyer and provider feedback. Updates are versioned and dated.