Pricing comparison

    What the cloud really costs

    List prices of 14 providers brought to the same units: one vCPU hour, one GB of RAM, one GB of block or object storage, one Kubernetes control plane, one H100 hour, one million tokens. Net, Frankfurt or EU region, checked daily against the public price lists – and as a calculator for your workload.

    Workload calculator

    One workload, 14 monthly prices

    Enter instances, storage, network, Kubernetes, database, GPU and tokens – the ranking uses the matrix list prices and shows per provider what is included and what has no public price.

    Net list prices, pay-as-you-go, Frankfurt or the provider's EU default region, 730 hours per month, USD/CHF at the matrix reference rate. No savings plans, reservations, support plans or egress tiers. If a provider has no list price for a component, it is named and not counted as 0 – the total is then a lower bound.

    Price matrix

    Unit prices compared

    Every row is a comparable unit, every cell a list price in euros. This matrix is where the "Cost Indication" rows and the cost rating in the provider profiles and the comparison table come from.

    Colour = the row's score: green cheapest, red most expensive provider (log-linear 10 → 3). Columns sorted by cost rating.

    Price itemInfomaniak Logo9.3/10Hetzner Logo8.9/10OVHcloud Logo8.2/10IONOS Cloud Logo6.7/10Oracle Cloud Logo6.6/10Hikube Logo6.4/10Scaleway Logo6.2/10Google Cloud Logo5.9/10IBM Cloud Logo5.8/10Microsoft Azure Logo5.7/10AWS Logo5.4/10STACKIT Logo5.4/10secunet cloud Logo5.2/10T Cloud Public Logo4.7/10
    Compute
    CPU per vCPUper hour€0.002205€0.0120€0.0190€0.0120€0.0129€0.0119€0.0187€0.0350€0.0390€0.0365€0.0357€0.0352€0.0399€0.0514
    RAM per GBper hour€0.000710€0.001390€0.001738€0.002000€0.001720€0.0119€0.004262€0.004694€0.003348€0.004105€0.004126€0.004080€0.002386€0.001442
    Windows licence per vCPUper hour€0.006780–€0.0347€0.0149€0.0396€0.0149€0.0544–€0.0396€0.0396–€0.0548–€0.0342
    GPU
    GPU H100 (1 GPU)per hour––€2.80–€8.60€2.98€2.87€9.51€10.71€7.80€7.40€8.04€5.66€5.22
    GPU L40S (1 GPU)per hour€0.6347–€1.40–€3.01€1.13€1.47–€3.22–€2.00€1.74€1.86–
    Storage
    Block storage per GB (≥ 3000 IOPS)per hour€0.000110€0.00007836€0.000119€0.00009720€0.00003004€0.000268€0.000130€0.000111€0.000120€0.000112€0.000112€0.001095€0.00008900€0.00009041
    Object storage standard per GBper hour€0.00001300€0.00000858€0.00000972€0.00000959€0.00003004€0.00002980€0.00002200€0.00002710€0.00002979€0.00002309€0.00002886€0.00003644€0.00003397€0.00003315
    Archive storage per GBper hour––€0.00000228–€0.00001178–€0.00001100€0.00000707€0.00000813€0.00000530€0.00000477–€0.00002740€0.00001808
    Deep archive / tape per GBper hour––€0.00000176–€0.00000306–€0.00000348€0.00000295€0.00000169€0.00000212€0.00000212––€0.00000633
    Platform
    Kubernetes control plane (SLA plan)per hour€0.0360–€0.0900€0.00€0.0860€0.0477€0.1100€0.0860€0.00€0.0860€0.0860€0.0996€0.1000€0.1334
    Managed database (1 vCPU / 4 GB per node)per hour€0.0267–€0.0373€0.0880€0.0315€0.0745€0.0259€0.0716€0.0930€0.0912€0.0860€0.0592€0.0748€0.0882
    Generative AI
    LLM input tokens Qwen3.8-27Bper 1M tokens–€0.00€0.4000€0.4000––€0.6000––––€0.4500–€0.5000
    LLM output tokens Qwen3.8-27Bper 1M tokens–€0.00€2.70€2.70––€3.30––––€0.6500–€2.50
    LLM input tokens gpt-oss-120bper 1M tokens––€0.0800€0.1500€0.1290–€0.1500€0.0774€0.1367€0.1290€0.1720€0.4500–€0.2000
    LLM output tokens gpt-oss-120bper 1M tokens––€0.4000€0.6500€0.5160–€0.6000€0.3096€0.5470€0.5160€0.6794€0.6500–€0.6500
    Network
    Public IPv4per hour€0.004110€0.000800€0.002700€0.006944€0.00€0.005961€0.005000€0.004300€0.005160€0.004300€0.004300€0.004056€0.004000€0.004400
    Egress per GBper GB€0.00€0.001000€0.00€0.0300€0.007310€0.00€0.00€0.1032€0.0871€0.0748€0.0774€0.00€0.007623€0.0745

    vCPU and RAM prices are derived from flavour prices (difference or least-squares method over 2c/8g, 4c/8g and 4c/16g sizes); "–" means no public list price in the comparison region. Token rows compare strictly the same model per provider.

    Cost rating

    Who is cheap overall – and at what?

    Per row the cheapest provider gets 10 points, the most expensive 3, log-linear in between. The cost rating is the mean of all priced rows – a transparent heuristic, not an editorial grade. The cards show where a provider wins or loses its score.

    1. 19.3/10

      InfomaniakEU / CH

      Cheap at
      CPU per vCPU (10.0) · RAM per GB (10.0)
      Expensive at
      Public IPv4 (4.7) · Block storage per GB (≥ 3000 IOPS) (7.5)

      10 of 17 rows priced · as of 3 Oct 2026 · Provider profile

    2. 28.9/10

      HetznerEU / CH

      Cheap at
      Public IPv4 (10.0) · Object storage standard per GB (10.0)
      Expensive at
      CPU per vCPU (6.2) · Block storage per GB (≥ 3000 IOPS) (8.1)

      8 of 17 rows priced · as of 3 Oct 2026 · Provider profile

    3. 38.2/10

      OVHcloudEU / CH

      Cheap at
      GPU H100 (1 GPU) (10.0) · Archive storage per GB (10.0)
      Expensive at
      LLM output tokens Qwen3.8-27B (3.9) · Windows licence per vCPU (4.5)

      17 of 17 rows priced · as of 3 Oct 2026 · Provider profile

    4. 46.7/10

      IONOS CloudEU / CH

      Cheap at
      Kubernetes control plane (SLA plan) (10.0) · LLM input tokens Qwen3.8-27B (10.0)
      Expensive at
      Public IPv4 (3.0) · Managed database (1 vCPU / 4 GB per node) (3.3)

      13 of 17 rows priced · as of 3 Oct 2026 · Provider profile

    5. 56.6/10

      Oracle CloudHyperscaler

      Cheap at
      Public IPv4 (10.0) · Block storage per GB (≥ 3000 IOPS) (10.0)
      Expensive at
      GPU L40S (1 GPU) (3.3) · Object storage standard per GB (3.9)

      15 of 17 rows priced · as of 3 Oct 2026 · Provider profile

    6. 66.4/10

      HikubeEU / CH

      Cheap at
      Egress per GB (10.0) · GPU H100 (1 GPU) (9.7)
      Expensive at
      RAM per GB (3.0) · Public IPv4 (3.5)

      11 of 17 rows priced · as of 7 Nov 2026 · Provider profile

    7. 76.2/10

      ScalewayEU / CH

      Cheap at
      Managed database (1 vCPU / 4 GB per node) (10.0) · Egress per GB (10.0)
      Expensive at
      LLM output tokens Qwen3.8-27B (3.0) · LLM input tokens Qwen3.8-27B (3.0)

      17 of 17 rows priced · as of 3 Oct 2026 · Provider profile

    8. 85.9/10

      Google CloudHyperscaler

      Cheap at
      LLM input tokens gpt-oss-120b (10.0) · LLM output tokens gpt-oss-120b (10.0)
      Expensive at
      Egress per GB (3.0) · GPU H100 (1 GPU) (3.6)

      13 of 17 rows priced · as of 3 Oct 2026 · Provider profile

    9. 95.8/10

      IBM CloudHyperscaler

      Cheap at
      Deep archive / tape per GB (10.0) · Kubernetes control plane (SLA plan) (10.0)
      Expensive at
      Managed database (1 vCPU / 4 GB per node) (3.0) · GPU L40S (1 GPU) (3.0)

      15 of 17 rows priced · as of 7 Nov 2026 · Provider profile

    10. 105.7/10

      Microsoft AzureHyperscaler

      Cheap at
      Deep archive / tape per GB (8.8) · LLM input tokens gpt-oss-120b (8.0)
      Expensive at
      Managed database (1 vCPU / 4 GB per node) (3.1) · Egress per GB (3.5)

      14 of 17 rows priced · as of 7 Nov 2026 · Provider profile

    11. 115.4/10

      AWSHyperscaler

      Cheap at
      Deep archive / tape per GB (8.8) · Archive storage per GB (7.9)
      Expensive at
      LLM output tokens gpt-oss-120b (3.0) · Egress per GB (3.4)

      14 of 17 rows priced · as of 7 Nov 2026 · Provider profile

    12. 125.4/10

      STACKITEU / CH

      Cheap at
      LLM output tokens Qwen3.8-27B (10.0) · Egress per GB (10.0)
      Expensive at
      LLM input tokens gpt-oss-120b (3.0) · Object storage standard per GB (3.0)

      15 of 17 rows priced · as of 3 Oct 2026 · Provider profile

    13. 135.2/10

      secunet cloudEU / CH

      Cheap at
      Block storage per GB (≥ 3000 IOPS) (7.9) · RAM per GB (7.0)
      Expensive at
      Archive storage per GB (3.0) · Object storage standard per GB (3.3)

      11 of 17 rows priced · as of 3 Oct 2026 · Provider profile

    14. 144.7/10

      T Cloud PublicEU / CH

      Cheap at
      RAM per GB (8.2) · Block storage per GB (≥ 3000 IOPS) (7.9)
      Expensive at
      Kubernetes control plane (SLA plan) (3.0) · Deep archive / tape per GB (3.0)

      16 of 17 rows priced · as of 3 Oct 2026 · Provider profile

    GPU & AI

    GPU and LLM prices of every provider

    Beyond the benchmark rows of the matrix: every GPU from T4 to B300 and every token-priced model the providers publish – per GPU-hour or per million tokens, directly comparable.

    182 GPU offers · 197 token prices · as of Nov 7, 2026

    GPU prices per GPU-hour

    Cheapest on-demand instance per GPU model and provider, normalised to one GPU. Green = cheapest provider in the row. Instance name and GPU count in the tooltip.

    GPUAWSMicrosoft AzureGoogle CloudOVHcloudsecunet cloudSTACKITIBM CloudIONOS CloudT Cloud PublicOracle CloudScalewayInfomaniakHetznerHikube
    GB300288 GB–––––––––€15.48––––
    B300288 GB–––––––––€12.90€7.50–––
    GB200192 GB–––––––––€13.76––––
    B200180 GB–––––––––€12.04––––
    MI355X288 GB–––––––––€7.40––––
    H200141 GB––€9.12€5.25–––€3.00–€8.60–––€3.72
    H200 NVL141 GB––––€8.94–––––––––
    MI300X192 GB–––––––––€5.16––––
    H10080 GB€7.40€13.74€9.51€2.80–€8.04€10.71–€3.41€8.60€2.87––€2.98
    H100 NVL94 GB–€7.80––€5.66–––€5.80–––––
    A10040/80 GB€2.95€3.80€3.16€2.75–€2.66––€3.90€3.44–€1.31–€2.23
    RTX PRO 600096 GB–––––––––€3.87–––€2.31
    RTX PRO 6000 BLACKWELL MAX Q96 GB––––––––––––€1.92–
    L40S48 GB€2.00––€1.40€1.86€1.74€3.22––€3.01€1.47€0.635–€1.13
    A4048 GB––––––––€3.90–––––
    A3024 GB––––€0.999–––––––––
    A1024 GB–€3.58–€0.760–––––€1.72––––
    A10G24 GB€1.08–––––––––––––
    RTX PRO 4000 BLACKWELL SFF24 GB––––––––––––€0.340–
    L424 GB€0.866–€0.716€0.750––––€0.748–€0.788€0.244–€0.246
    QUADRO RTX 500016 GB–––€0.360––––––––––
    V100S32 GB–––€0.800––––––––––
    V10016/32 GB€3.29––€0.700––––€3.11–––––
    T416 GB€0.566€0.808––––––€0.720––€0.213––
    T4G16 GB€0.452–––––––––––––
    A216 GB–––––––––––€0.146––
    P10016 GB––––––––––€1.22–––
    K8012 GB€1.14–––––––––––––
    RADEON PRO V5208 GB€0.407–––––––––––––

    LLM API prices per million tokens

    Blended price of 3 parts input and 1 part output per 1M tokens; input/output separately in the tooltip. Same model = same row, even when providers name it differently.

    ModelAWSMicrosoft AzureGoogle CloudOVHcloudSTACKITIBM CloudIONOS CloudT Cloud PublicOracle CloudScalewayInfomaniakHetzner
    gpt-oss-120b€0.149€0.226€0.135€0.160€0.500€0.239€0.275€0.313€0.226€0.262––
    Llama 3.3 70B–€0.611€0.619€0.670€0.500–€0.650––€0.900––
    Qwen3.8-27B–––€0.975€0.500–€0.975€1.00–€1.28–€0.0000
    gpt-oss-20b€0.144–€0.0989€0.0675€0.175–––€0.110–––
    Qwen3.5-397B-A17B–––€1.35––€1.35––€1.35€1.50–
    Claude Haiku 4.5€1.72–€1.72––––€1.98––––
    gemma-4-31B-it––––€0.500––€0.750––€0.250–
    Qwen3 235B A22B 2507€0.218–€0.331––––––€1.13––
    Qwen3-32B€0.149€0.452–€0.118––––––––
    Qwen3 Coder 30B A3B€0.299––€0.100–––––€0.350––
    Claude Opus 4.6€9.46––––––€9.90––––
    Claude Opus 4.8––€8.60––––€9.90––––
    Claude Sonnet 4.6€5.68––––––€5.94––––
    Claude Sonnet 5––€3.44––––€5.94––––
    DeepSeek-R1–€2.03€2.03–––––––––
    DeepSeek-V3.1€0.735–€0.753–––––––––
    DeepSeek-V3.2–€0.735€0.722–––––––––
    Gemini 2.5 Flash––€0.731–––––€0.731–––
    Gemini 2.5 Flash Lite––€0.151–––––€0.151–––
    Gemini 2.5 Pro––€2.96–––––€2.96–––
    Gemini 3.5 Flash––€2.90––––€3.34––––
    GLM-5.2–––––––€2.00–€2.73––
    GPT-5–€2.96–––––€3.29––––
    GPT-5 mini–€0.591–––––€0.658––––
    GPT-5 nano–€0.118–––––€0.135––––
    Llama 4 Maverick 17B-128E–€0.376–––€0.558––––––
    Mistral Medium 3––€0.688––––€0.720––––
    Mistral-Small-3.2-24B-Instruct-2506–––€0.138–––––€0.200––
    Mistral-Small-4-119B-2603–––––––€0.750––€0.338–
    Qwen3.5-9B–––€0.113––€0.113–––––
    Qwen3.6-27B–––€0.975€0.500–––––––
    qwen3.6-35b-a3b–––––––––€0.563–€0.0000

    Net list prices, USD/CHF converted to euro at a reference rate. Sources are the providers’ price lists: curated entries of the price matrix plus every further GPU instance and model the daily crawler finds in the public catalogues. GPU models are derived from the provider’s naming scheme; anything that cannot be identified unambiguously is left out rather than guessed. No discounts, reservations or spot prices. catalog.json

    Cloud Price Index

    How cloud list prices move

    Every daily check writes price changes into a time series. From it we derive a price index for all providers, for EU/CH providers and for hyperscalers, sub-indices per price axis, the monthly price of a fixed reference workload and a public changelog.

    Overall

    100.0

    0 % since 2 Nov 2026

    EU/CH providers

    100.0

    0 % since 2 Nov 2026

    Hyperscalers

    100.0

    0 % since 2 Nov 2026

    Reference workload, median / month

    EU/CH providers
    €276
    Hyperscalers
    €409

    EU/CH median 32 % cheaper than hyperscalers

    OverallEU/CH providersHyperscalersBase 100 = 2 Nov 2026
    98.7100.0101.3Nov 02Nov 07

    Compute

    100.0

    0 %

    Storage

    100.0

    0 %

    Network

    100.0

    0 %

    Database

    100.0

    0 %

    AI & GPU

    100.0

    0 %

    The series started on 2 Nov 2026 and grows with every daily check; a point is only added when a list price actually changes.

    Reference workload per provider

    2 VMs with 4 vCPU/16 GB, 1 IPv4, 500 GB block, 1 TB object storage, 500 GB egress · 730 h · net list price

    1. Infomaniak€82.1
    2. Hetzner€138
    3. Oracle Cloud€152
    4. IONOS Cloud€179
    5. OVHcloud€204
    6. Scaleway€276
    7. secunet cloud€356
    8. Microsoft Azure€407
    9. AWS€409
    10. IBM Cloud€419
    11. Google Cloud€429
    12. T Cloud Public€431
    13. Hikube€472
    14. STACKIT€730

    Price changes (changelog)

    history.json
    DateProviderUnitBeforeNewΔ
    7 Nov 2026IBM CloudRAM per GB€0.003448€0.003348-2.9 %
    7 Nov 2026IBM CloudCPU per vCPU€0.0377€0.0390+3.5 %
    7 Nov 2026HikubeBlock storage per GB (≥ 3000 IOPS)€0.000272€0.000268-1.4 %

    Method: the index is a Jevons index – the geometric mean of price relatives per provider and unit against the first recorded price, every provider weighted equally. Base 100 is the first day of the series; a provider added later starts from its own first price. If a benchmark model changes (e.g. the reference LLM), its series restarts instead of counting as a price change. Inputs are the same net list prices as in the matrix above.

    Data status & sources

    Where the numbers come from

    A daily job reads the providers' public price sources – price APIs, catalogues embedded in price pages, HTML tables and PDF price lists – and reconciles them with the matrix. Only prices are taken over; which products are compared stays editorial. Where a provider offers no machine-readable source, it says so here.

    ProviderSource / methodLast checkLast price changeReconciledSources
    AWS
    eu-central-1 · USD
    Price API
    AWS price list API (eu-central-1): EC2 meteredUnitMaps + bulk offers S3/EKS/data transfer/RDS
    partial
    9 Nov 2026
    7 Nov 2026
    32
    3 not in source
    Microsoft Azure
    germanywestcentral · USD
    Price API
    Azure Retail Prices API (germanywestcentral, pay-as-you-go, USD)
    partial
    9 Nov 2026
    7 Nov 2026
    22
    18 not in source
    Google Cloud
    europe-west3 · USD
    editorial
    Google Cloud pricing pages (JavaScript-rendered)
    Google publishes machine-readable prices only through the Cloud Billing Catalog API, which needs an API key; the public pricing pages render the region tables client-side. Prices are checked by hand against the listed sources.
    manual
    3 Oct 2026
    3 Oct 2026
    –
    OVHcloud
    DE1 · EUR
    Catalogue JSON of the price page
    OVHcloud price page — embedded product catalog (DE, net)
    partial
    9 Nov 2026
    3 Oct 2026
    60
    9 not in source
    secunet cloud
    HAM1 · EUR
    PDF price list
    secunet cloud (SysEleven) Preisliste PDF (EUR net, HAM1/DUS2)
    partial
    9 Nov 2026
    3 Oct 2026
    14
    1 not in source
    STACKIT
    EU01 · EUR
    PDF price list
    STACKIT price list PDF (linked from the price page; EUR net, EU01)
    confirmed
    9 Nov 2026
    3 Oct 2026
    29
    IBM Cloud
    eu-de · USD
    Price API
    IBM Cloud Global Catalog pricing (eu-de, USD, per VPC profile)
    confirmed
    9 Nov 2026
    7 Nov 2026
    4
    IONOS Cloud
    DE · EUR
    HTML table
    IONOS Cloud price page — HTML price tables (EUR net)
    IONOS price page: vCPU/RAM table not found
    source changed
    8 Nov 2026
    3 Oct 2026
    0
    6 not in source
    T Cloud Public
    eu-de · EUR
    PDF price list
    T Cloud Public Leistungsbeschreibung PDF (EU-DE, EUR net, hourly on-demand column)
    partial
    9 Nov 2026
    3 Oct 2026
    16
    11 not in source
    Oracle Cloud
    eu-frankfurt-1 · USD
    Price API
    Oracle Cloud cetools price list API (USD, all regions)
    partial
    9 Nov 2026
    3 Oct 2026
    16
    20 not in source
    Scaleway
    fr-par · EUR
    editorial
    Scaleway price pages (bot shield) / instance products API (prices only with an API key)
    The price pages sit behind a bot shield (HTTP 403 challenge), which we do not work around, and the public instance products API returns hourly_price 0 without authentication. Prices are checked by hand against the listed sources; with SCW_SECRET_KEY the products API could be read.
    manual
    3 Oct 2026
    3 Oct 2026
    –
    Infomaniak
    Schweiz · EUR
    Catalogue JSON of the price page
    Infomaniak price page data (Gatsby static query, EUR net per hour)
    partial
    9 Nov 2026
    3 Oct 2026
    17
    2 not in source
    Hetzner
    eu-central · EUR
    Price API
    Hetzner website price API (live_data_prices.json, EUR net, FSN1)
    confirmed
    9 Nov 2026
    3 Oct 2026
    20
    Hikube
    Schweiz · CHF
    Catalogue JSON of the price page
    Hikube pricing page — SKU catalog in the page payload (CHF/month, net)
    confirmed
    9 Nov 2026
    7 Nov 2026
    52

    Method in five sentences

    We compare net list prices (pay-as-you-go) in the Frankfurt region or the provider's EU default region; USD and CHF are converted to euros at an editorially maintained reference rate.

    Hardly any provider quotes a price per vCPU or per GB of RAM, so they are derived from flavour prices: same RAM and a different vCPU count gives the vCPU price, same vCPU count and different RAM the RAM price, the remainder is a per-instance base. One instance series is pinned per provider so generations do not mix.

    Storage rows take the cheapest class with at least 3000 IOPS and the standard, archive and deep-archive classes; Kubernetes the paid SLA plan, not the free tier; managed database a MySQL or PostgreSQL plan normalised to 1 vCPU and 4 GB per node; token rows strictly the same model per provider.

    A month is 730 hours. Discounts, savings plans, reservations, support plans, minimum storage durations and egress tiers are not included – the calculator delivers the list price as a comparison basis, not a quote.

    Every number carries a source; the provider blocks are open YAML in the repository, the matrix is JSON. matrix.json · Full methodology

    Frequently asked questions about the cloud pricing comparison

    Are the prices net or gross?

    Net. All rows are list prices without VAT, pay-as-you-go, in the Frankfurt region or the provider's EU default region. US dollar and Swiss franc prices are converted to euros at an editorially maintained reference rate shown in the matrix.

    Why does the matrix compare prices per vCPU and per GB of RAM instead of per instance?

    Because instance sizes never line up across providers. From flavour prices with the same RAM and a different vCPU count we derive the vCPU price, from the same vCPU count and different RAM the RAM price; the remainder is a per-instance base. A 2-vCPU, 8-GB instance is then built from the same bricks at every provider.

    How current are the prices?

    A daily job reads the public price sources – price APIs at AWS, Azure, Oracle, Hetzner, Scaleway and IBM, catalogues embedded in the price page at OVHcloud, Hikube and Infomaniak, HTML tables at IONOS, PDF price lists at STACKIT, T Cloud Public and secunet cloud – and updates prices that changed. The "Data status & sources" table shows the last check and the last price change per provider. Google Cloud publishes machine-readable prices only through a keyed API and is checked editorially.

    What does the cost rating mean?

    Per price row the cheapest provider gets 10 points and the most expensive 3, all others sit log-linearly in between; a row a provider offers for free counts 10. The cost rating is the mean over all priced rows of a provider. It is a transparent heuristic for orientation, not an editorial grade and no substitute for a quote.

    Why is a row missing for some providers?

    Because there is no public list price for that unit in the comparison region – for example H100 GPUs at AWS and Google in Frankfurt, or the benchmark model Qwen3.8-27B at Azure and Infomaniak. Missing rows are neither estimated nor counted as zero; the calculator names them as missing and the total becomes a lower bound.

    What does the calculator not tell you?

    It uses list prices without discounts: no savings plans, reservations or committed-use discounts, no support plans, no minimum storage durations, no free egress volumes or volume tiers, no licence special cases. It is the comparison basis before quotes are requested – for a robust calculation add the Cloud Compass or an assessment.