Clarke Standards v1.0
Clarke Standards
The public reference for how Clarke defines and communicates AI data center capacity.
The Problem Clarke Standards Solve
AI infrastructure investment decisions are being made with incomplete and inconsistent capacity data. Site power contracts get confused with actual IT capacity. Marketing announcements don't distinguish what's live today from what's planned for 2028. Forward-looking projections—when they exist—lack uncertainty ranges or confidence assessment.
Clarke Standards provide the AI infrastructure market with a clear, shared vocabulary for capacity intelligence: precise definitions that separate IT capacity from site power, facility and market taxonomies for apples-to-apples comparison, forecast frameworks that express uncertainty through scenario ranges, and provenance standards that distinguish primary sources from estimates.
These standards underpin the Clarke Index, Regional Capacity Forecasts, Frontier Projections, and Clarke Pulse reports.
Citing and Licensing
Clarke Standards are public and free to use with attribution. You may reference, quote, or build upon these definitions in your own work.
Citation format:
Clarke Standards (v1.0), "The Intelligence Layer for AI Infrastructure" — clarkeindex.com/standards
Clarke Standards are versioned and evolving. Updates will be documented on this page.
Methodology
Clarke's capacity intelligence combines systematic primary source research, grid and utility analysis, construction timeline modeling, and operator assessment—all with explicit uncertainty quantification.
Forecasts are ranges, not guarantees. Clarke makes no guarantee that any forecast or projection will prove accurate; they reflect the best available information and assumptions as of the stated Forecast Vintage. For these forecasts, Clarke presents three scenarios:
- Base Case: Most likely outcome given current information
- Conservative: Lower bound assuming delays and constraints
- Optimistic: Upper bound assuming smooth execution
The spread between scenarios creates a confidence band—wider bands signal higher uncertainty.
All forward-looking intelligence includes confidence assessment:
- Scenario-based projections, not single-point predictions
- "Clarke Estimate" values clearly labeled when official data is unavailable
- Confidence tags indicate data quality: Decision-grade (primary sources), Probable (corroborated), or Speculative (directional)
What Clarke shares publicly: Full vocabulary, IT capacity distinctions, high-level methodology, scenario ranges, and market trends.
What Clarke keeps proprietary: Scoring algorithms, model weights, grid feasibility models, operator credibility assessments, and data pipeline architecture.
Clarke Vocabulary
The definitions that power Clarke's intelligence.
1. Facilities & Types
Clarke tracks only the part of the data center universe that materially supports AI workloads. We use AI data center as the canonical term, and then classify these facilities into AI-Native, Frontier AI, AI-Capable, and General-Purpose.
Campus
A cluster of one or more data center buildings in the same geographic area (typically within 10 km) that share power infrastructure and network connectivity.
Facility
A single data center building or development phase within a campus. This is the main unit of analysis in the Clarke Index.
AI Data Center
A data center whose primary purpose is to deliver accelerator-based AI and high-performance computing workloads using GPUs (Graphics Processing Units) or TPUs (Tensor Processing Units). In operator marketing or regulation these may be called "AI factories," "AI compute facilities," or "AI manufacturing facilities"; Clarke treats all of these as AI data centers.
AI-Native Data Center
An AI data center purpose-built for large-scale AI workloads, featuring high-density racks (often 50-100+ kW per rack), advanced liquid cooling, and AI-optimized networking fabric. AI-Native facilities are designed around accelerator clusters rather than retrofitted from general-purpose enterprise designs.
Frontier AI Data Center
A subset of AI-Native data centers specifically designed and built to train frontier AI models—the largest and most advanced models at the edge of current capabilities. Frontier AI data centers typically combine very large IT capacity (often 100+ MW), dense GPU/accelerator clusters, cutting-edge interconnect (InfiniBand or custom fabrics), and datacenter-scale liquid cooling, and are operated by leading AI labs and hyperscalers. In this context, "frontier AI models" refers broadly to the largest, most compute-intensive models deployed at a given time, rather than any specific regulatory designation.
AI-Capable Data Center
A data center that can support AI workloads but was not originally designed or fully optimized for them. These facilities may be retrofitted for higher densities or used for mixed workloads (traditional enterprise + AI).
General-Purpose Data Center
A facility designed primarily for traditional cloud, enterprise, and mixed IT workloads, with no material AI optimization. Clarke generally does not include purely general-purpose facilities in the Clarke Index unless they host significant AI clusters.
2. Capacity & Power
Site Power Contract (MW)
Total power capacity (in megawatts) secured from the utility for a facility or campus, including IT load, cooling, and infrastructure overhead.
IT Capacity — Live (MW)
IT power currently energized and available for compute workloads. This is Clarke's primary capacity metric—what's operational today, not what's planned.
IT Capacity — Planned (MW)
Additional IT power committed or under construction but not yet operational. Includes capacity in permitting, construction, or commissioning phases.
AI IT Capacity (MW)
The portion of IT Capacity allocated to AI and HPC workloads (GPU/TPU clusters). This distinguishes AI infrastructure from general enterprise computing.
Stranded Capacity (MW)
Power contracted from the utility but unable to be deployed within the forecast period due to non-power bottlenecks—cooling delays, networking shortages, permitting challenges, or construction lags.
At-Risk Capacity (MW)
Planned capacity with elevated probability of delays or cancellation due to grid uncertainty, permitting challenges, or organizational strategy changes.
3. Status & Timeline
Entitlement Status
Where a facility sits in the regulatory permitting and land-use approval process, from early concept through fully entitled (all major permits secured).
Build Phase
Where a facility is in the construction lifecycle: Announced, Under Construction, Commissioning, Operational, or Expansion.
RFS Target Date (Ready For Service)
The projected date when a facility will be operational and ready to run production workloads. Based on operator statements, construction timelines, utility schedules, and Clarke analysis.
4. Markets, Regions & Grid
Market
A specific geographic area where AI data center capacity is concentrated, typically defined by utility service territory or grid boundaries. Examples: DFW-Oncor, Northern Virginia, Central Iowa.
Region
A broader geographic grouping of markets for high-level analysis. Examples: US Southeast, US West, US Central.
Market IT Capacity — Live / Planned (MW)
The aggregate Live and Planned IT Capacity across all tracked facilities in a given market.
Grid Constraint Type
The primary factor limiting new capacity in a market: transmission bottlenecks, substation upgrades, permitting delays, generation shortfalls, or interconnection queue backlogs.
5. Forecasts & Uncertainty
Capacity Forecast
Clarke's forward-looking projection of IT Capacity at facility, market, region, and organization levels over a 2-5 year horizon. Combines operator plans, utility filings, construction timelines, and grid analysis.
Base Case Forecast
Clarke's central scenario—the most likely outcome given current information. Assumes typical execution timelines and moderate delays.
Conservative Forecast
A lower-bound scenario assuming more severe delays and constraints. Used for stress-testing assumptions.
Optimistic Forecast
An upper-bound scenario assuming smooth execution and accelerated timelines.
Confidence Band
The range between Conservative, Base Case, and Optimistic forecasts. Clarke presents capacity forecasts as ranges, not single-point predictions. Wider bands indicate higher uncertainty.
Forecast Horizon
The time window over which Clarke projects capacity, typically 2-5 years forward.
Forecast Vintage
The date a forecast was produced or last updated. Tracking vintages shows how projections evolve and measures forecast accuracy over time.
6. Provenance & Confidence
Primary Source
Official or regulatory documents: operator filings, utility dockets, SEC/EDGAR disclosures, interconnection queue reports. The gold standard for capacity intelligence.
Secondary Source
Credible third-party material: trade press, vendor announcements, analyst reports, construction databases. Evaluated for reliability and cross-referenced when possible.
Clarke Estimate
A value Clarke has modeled or inferred when official data is missing or inconsistent. Always labeled as such and paired with confidence assessment.
Confidence Score / Tag
Clarke's data quality assessment using tags like "Decision-grade" (primary sources, high confidence), "Probable" (multiple sources, moderate confidence), or "Speculative" (limited information, directional only).
7. Clarke Products & Views
Clarke Index
Clarke's curated index of AI data centers, with high-impact facilities covered through comprehensive dossiers and capacity forecasts. Prioritizes frontier training clusters, major hyperscaler buildouts, and strategic markets.
Regional Capacity Forecasts
Market- and region-level capacity intelligence showing what's live today and what's coming over multi-year horizons. Combines interactive mapping, scenario projections with confidence bands, grid constraint analysis, and timeline visualization.
Frontier Projections
Comparative intelligence on major AI actors—labs, hyperscalers, and operators. Aggregates facility-level data into per-organization capacity totals and forecasts for competitive benchmarking.
Clarke Pulse
Executive reports and public blog translating capacity data into strategic narratives: market commentary, trend analysis, facility deep-dives, and forward-looking intelligence.
If you'd like to reference or build on Clarke Standards, you can cite:
Clarke Standards, Public v1.0 — "The Intelligence Layer for AI Infrastructure."