Skip to content
Get my AI grade

Rank.ai Index

Infrastructure

The compute, data centers, power, chips and clouds the AI build out runs on, priced and counted from public filings and official releases.

42

Boards

7

Categories

Oct 26, 2026

Latest snapshot

Featured view

US AI data centre map

Planned capacity, sites, owners and power prices, state by state.

Leaders

The leaders on the first 3 infrastructure boards

Each board names its source and the date of its snapshot, and every row opens the evidence behind it.

Open the first board
Estimated AI compute ownership by companyEstimated owned compute (median)
  1. Alphabet5,453,215 H100e
  2. Microsoft3,574,563 H100e
  3. Amazon2,508,034 H100e
Operational AI data centersFacility power
  1. Anthropic-Amazon New Carlisle1,037 MW
  2. Colossus 2946 MW
  3. Microsoft Fairwater Atlanta636 MW
Planned AI data center capacityFacility power
  1. Microsoft Fairwater Wisconsin2,715 MW
  2. OpenAI Stargate New Mexico2,450 MW
  3. Anthropic-Amazon New Carlisle2,310 MW

Boards

Every infrastructure board

GPUs and compute · How the boards are measured

All 42 boards are free to read and link to the evidence behind every row. Each board shows the date of its snapshot.

Compute

What an hour of GPU time costs, and who owns the most of it.

Azure GPU VM retail prices
  1. 1Azure NC24ads A100 v4$4 / GPU-hr
  2. 2Azure ND96amsr A100 v4$4 / GPU-hr
  3. 3Azure NC40ads H100 v5$7 / GPU-hr
The top three sit within 3.3 of each other.6 rows
AWS EC2 GPU instance prices
  1. 1AWS EC2 g6.xlarge$0.805 / GPU-hr
  2. 2AWS EC2 g6.12xlarge$1 / GPU-hr
  3. 3AWS EC2 g6e.xlarge$2 / GPU-hr
The top three sit within 1.1 of each other.9 rows
Akash GPU market
  1. 1NVIDIA M4000 8 GB PCIe$0.040 / GPU-hr
  2. 2NVIDIA P4 8 GB PCIe$0.060 / GPU-hr
  3. 3NVIDIA RTX 3060TI 8 GB PCIe$0.140 / GPU-hr
The top three sit within 0.1 of each other.15 rows

Datacenters

Where capacity is being built, and how much power it will need.

Operational AI data centers
  1. 1Anthropic-Amazon New Carlisle1,037 MW
  2. 2Colossus 2946 MW
  3. 3Microsoft Fairwater Atlanta636 MW
Anthropic-Amazon New Carlisle leads by 91.74 rows
Planned AI data center capacity
  1. 1Microsoft Fairwater Wisconsin2,715 MW
  2. 2OpenAI Stargate New Mexico2,450 MW
  3. 3Anthropic-Amazon New Carlisle2,310 MW
Microsoft Fairwater Wisconsin leads by 265.54 rows

Power

What electricity costs by state, and how clean each grid is.

State power economics
  1. 1New Mexico5.08¢/kWh
  2. 2Louisiana6.10¢/kWh
  3. 3Tennessee6.92¢/kWh
The top three sit within 1.8 of each other.51 rows
State grid carbon intensity
  1. 1Vermont52 pounds per megawatt hour
  2. 2Washington267 pounds per megawatt hour
  3. 3New Hampshire276 pounds per megawatt hour
The top three sit within 224 of each other.51 rows

Chips & memory

Memory, bandwidth and supply across the accelerators everyone is waiting on.

Vector databases

Who AI recommends when a buyer asks about Vector DBs, and which pages those answers cite.

Vector database engine breadth
  1. 1Weaviate3 engines
  2. 2Zilliz Cloud / Milvus3 engines
  3. 3Qdrant3 engines
Zilliz Cloud / Milvus climbed 2 places on the last run.16 brands
Vector database buying prompt leaders
  1. 1Recommend an open-source vector database that an engineering team can self-host in its own cloud account.100%
  2. 2Recommend a vector retrieval platform for searching text, images, audio, and other multimodal embeddings.67%
  3. 3What is the best serverless or usage-based vector database for a startup with an unpredictable retrieval workload?67%
This question climbed 2 places on the last run.12 questions
Vector database engine disagreement
  1. 1When should a team buy a dedicated vector database instead of adding embedding search to its existing operational database?100%
  2. 2Which enterprise vector retrieval services offer private networking, access controls, compliance, and regional deployment?100%
  3. 3How should an engineering team evaluate vector search benchmark claims for recall, tail latency, throughput, scale, and total cost?100%
This question climbed 8 places on the last run.12 questions

Cloud GPU platforms

Who AI recommends when a buyer asks about GPU clouds, and which pages those answers cite.

Cloud GPU platform rank
  1. 1Runpod58%
  2. 2Lambda Cloud50%
  3. 3Google Cloud42%
Lambda Cloud climbed 2 places on the last run.16 brands
Cloud GPU buying prompt leaders
  1. 1What GPU cloud should a startup use for fast self-service access, clear hourly pricing, persistent storage, and simple deployment?100%
  2. 2Recommend cloud GPU providers with transparent on-demand access to H100 or H200 instances and no long-term commitment.100%
  3. 3Which GPU cloud platforms are best for an enterprise requiring private networking, compliance controls, auditability, and support?67%
This question climbed 6 places on the last run.12 questions
Cloud GPU engine disagreement
  1. 1Which GPU clouds offer strong managed Kubernetes or Slurm support for distributed AI training and inference?100%
  2. 2Which GPU cloud platforms are best for an enterprise requiring private networking, compliance controls, auditability, and support?89%
  3. 3Which cloud GPU platforms are best for reliable multi-node training of large foundation models?81%
This question climbed 7 places on the last run.12 questions

Cloud platforms

Who AI recommends when a buyer asks about Cloud, and which pages those answers cite.

Cloud platform rank
  1. 1Google Cloud71%
  2. 2Amazon Web Services67%
  3. 3Microsoft Azure54%
Google Cloud leads by 4.2 points.12 brands
Cloud platform presence
  1. 1Google Cloud54%
  2. 2Amazon Web Services54%
  3. 3Microsoft Azure48%
Google Cloud leads by 0 points.12 brands
Cloud platform engine breadth
  1. 1Google Cloud3 engines
  2. 2Amazon Web Services3 engines
  3. 3Microsoft Azure3 engines
The top three are level at 3 engines.12 brands
Cloud buying prompt leaders
  1. 1Which cloud platforms are strongest for a large enterprise running a mixed portfolio of modern and legacy applications?100%
  2. 2What cloud platform should a startup evaluate for predictable costs, simple deployment, developer support, and room to scale?100%
  3. 3Recommend cloud providers for organizations that need sovereign deployment options, broad regional coverage, and data-residency controls.67%
This question climbed 9 places on the last run.16 questions
Cloud platform engine disagreement
  1. 1Recommend cloud providers for organizations that need sovereign deployment options, broad regional coverage, and data-residency controls.94%
  2. 2Which cloud platform is best for a governed lakehouse, streaming analytics, business intelligence, and machine learning stack?92%
  3. 3Which cloud platforms are strongest for regulated workloads that require private networking, auditability, compliance, and key control?83%
The leading row leads by 2.8 points.16 questions

See how often AI names you.

Enter your website. In about two minutes, Rank.ai asks ChatGPT, Claude and Gemini 12 questions your customers ask and grades how often they name you.

  • Your grade out of 100How often AI names you, cites your site, and how high it ranks you.
  • Who gets namedEvery competitor in the answers, most named first.
  • The pages AI readsThe sources behind each answer.
  • Three fixesWhat to fix first, with a brief for the first article.