Objective, 2x by 2035, WLD
Double global compute capacity by 2035
What would have to happen for the world's AI compute capacity to double by 2035?
Current level
unavailableModeled
No stored series measures global compute or data-center power capacity yet; the catalog has no observed metric for it. The level is marked modeled and unavailable, the required pace is not computed, and the AI infrastructure corporate pulse stands in as the only response signal.
Target level in 2035
unavailableDerived
Current level unavailable.
Required annual addition
unavailableDerived
Current level unavailable, so nothing to divide.
Recent pace, trailing 5-year average change
unavailableDerived
No observations stored.
Forecast pace, Forecast Lab 5-year horizon
unavailableModeled
This metric is not a Forecast Lab target.
Required pace minus recent pace
unavailableDerived
Required pace unavailable.
Required pace minus forecast pace
unavailableModeled
Required pace unavailable.
Target metric data-center-power-capacity, entity AI compute, base year 2026, 9 years to go.
Same engine, same stored rows, no AI
Explore a different target
Backward walk over requires_input, depends_on and constrains edges, depth 4, 13 constraints
Dependency tree
Objective on the left, what it needs to the right
- Candidate
- Hypothesized
- Mechanism supported
- Causally supported
- Established relation
depth 1 AI acceleratorproduct
Established relationrequires inputstrength 1confidence 0.99relationshipwhy AI accelerator
Training and serving large models runs on accelerators. Compute capacity is the number of accelerators installed times their throughput.
- Input coefficient
- hypothesized dependence, no stored coefficient
- Bottleneck status
- no bottleneck row stored
- Lead time
- no cited lead time
- Substitutes
- none stored on this edge
- Response under way
- Semiconductor Capacity Pulse: 0.0553 fraction (2026-07-03) Derived
10 companies contributed to the stored pulse.
- Also needed by
- no other stored dependent
depth 2 High bandwidth memoryproduct
Established relationrequires inputstrength 0.9confidence 0.95relationshipwhy High bandwidth memory
Current accelerators are packaged with stacks of high bandwidth memory beside the logic die to feed it at the bandwidth large models need; the JEDEC standard defines that interface.
- Input coefficient
- hypothesized dependence, no stored coefficient
- Bottleneck status
- no bottleneck row stored
- Lead time
- no cited lead time
- Substitutes
- none stored on this edge
- Response under way
- Semiconductor Capacity Pulse: 0.0553 fraction (2026-07-03) Derived
10 companies contributed to the stored pulse.
- Also needed by
- no other stored dependent
depth 2 Semiconductor fabrication plantfacility36 to 60 months
Established relationrequires inputstrength 1confidence 0.99relationshipwhy Semiconductor fabrication plant
The logic die is manufactured on leading-edge process nodes in a small number of fabs.
- Input coefficient
- hypothesized dependence, no stored coefficient
- Bottleneck status
- no bottleneck row stored
- Lead time
- 36 to 60 months (A new 300 mm fab from groundbreaking through tool installation and qualification to volume output.) Congressional Research Service Recommended
A modern fab typically requires three to five years from groundbreaking to volume production.
- Substitutes
- none stored on this edge
- Response under way
- Semiconductor Capacity Pulse: 0.0553 fraction (2026-07-03) Derived
10 companies contributed to the stored pulse.
- Also needed by
- no other stored dependent
depth 3 Electricity systeminfrastructure system
Established relationrequires inputstrength 0.9confidence 0.98relationshipwhy Electricity system
Fabs run cleanrooms, tools and chillers continuously and are among the most electricity-intensive industrial facilities.
- Input coefficient
- hypothesized dependence, no stored coefficient
- Bottleneck status
- no bottleneck row stored
- Lead time
- no cited lead time
- Substitutes
- none stored on this edge
- Response under way
- no pipeline or pulse maps to this entity
- Also needed by
- no other stored dependent
depth 3 Industrial process watercommodity
Established relationrequires inputstrength 0.9confidence 0.95relationshipwhy Industrial process water
Wafer processing rinses with ultrapure water between steps; a fab uses millions of gallons a day. Recycling reduces intake but not the need.
- Input coefficient
- hypothesized dependence, no stored coefficient
- Bottleneck status
- no bottleneck row stored
- Lead time
- no cited lead time
- Substitutes
- none stored on this edge
- Response under way
- no pipeline or pulse maps to this entity
- Also needed by
- no other stored dependent
depth 4 Renewable freshwatercommodity
Established relationrequires inputstrength 0.9confidence 0.95relationshipwhy Renewable freshwater
Process water is drawn from surface water and groundwater and then purified; regional freshwater availability bounds where water-intensive plants can be sited.
- Input coefficient
- hypothesized dependence, no stored coefficient
- Bottleneck status
- no bottleneck row stored
- Lead time
- no cited lead time
- Substitutes
- none stored on this edge
- Response under way
- no pipeline or pulse maps to this entity
- Also needed by
- no other stored dependent
depth 3 Lithography toolsproduct
Established relationrequires inputstrength 1confidence 0.98relationshipwhy Lithography tools
Patterning at leading-edge nodes uses extreme ultraviolet scanners from a single supplier; fab capacity is bounded by tool deliveries.
- Input coefficient
- hypothesized dependence, no stored coefficient
- Bottleneck status
- no bottleneck row stored
- Lead time
- no cited lead time
- Substitutes
- none stored on this edge
- Response under way
- Semiconductor Capacity Pulse: 0.0553 fraction (2026-07-03) Derived
10 companies contributed to the stored pulse.
- Also needed by
- no other stored dependent
depth 1 Data centerfacility
Established relationrequires inputstrength 1confidence 0.99relationshipwhy Data center
Accelerators run in racks that need power, cooling and network at data-center scale.
- Input coefficient
- hypothesized dependence, no stored coefficient
- Bottleneck status
- no bottleneck row stored
- Lead time
- no cited lead time
- Substitutes
- none stored on this edge
- Response under way
- AI Infrastructure Pulse: 0.7014 fraction (2026-07-03) Derived
8 companies contributed to the stored pulse.
- Also needed by
- no other stored dependent
depth 2 Transmission and distribution gridinfrastructure system84 to 180 months
Established relationrequires inputstrength 1confidence 0.98relationshipwhy Transmission and distribution grid
A data center draws tens to hundreds of megawatts continuously and operates only once connected to the grid at that capacity, so siting and energization follow interconnection.
- Input coefficient
- hypothesized dependence, no stored coefficient
- Bottleneck status
- no bottleneck row stored
- Lead time
- 84 to 180 months (Interstate transmission projects from initiation through permitting and construction.) U.S. Department of Energy, Grid Deployment Office Recommended
New transmission takes on average around 10 years to complete, with some projects requiring 15 years or more; permits alone average more than four years.
- Substitutes
- none stored on this edge
- Response under way
- Grid Investment Pulse: 0.3521 fraction (2025-12-31) Derived
9 companies contributed to the stored pulse.
- Also needed by
- no other stored dependent
depth 3 Power transformerproduct24 to 60 months
Established relationrequires inputstrength 1confidence 0.99relationshipwhy Power transformer
Power is transmitted at high voltage to limit losses and stepped down for distribution. Every generator connection, substation and distribution feeder passes through transformers, so grid expansion and interconnection queues move with transformer supply and lead times.
- Input coefficient
- hypothesized dependence, no stored coefficient
- Bottleneck status
- no bottleneck row stored
- Lead time
- 24 to 60 months (Delivery of a large power transformer from order, at current order books; a new transformer plant takes longer.) U.S. Department of Energy Recommended
Lead times for large power transformers can reach as much as 60 months; the NIAC report of June 2024 puts current lead times at roughly two to four years.
- Substitutes
- none stored on this edge
- Response under way
- Grid Investment Pulse: 0.3521 fraction (2025-12-31) Derived
9 companies contributed to the stored pulse.
- Also needed by
- no other stored dependent
depth 4 Refined coppercommodity120 to 240 monthsbottleneck row
Established relationrequires inputstrength 0.8confidence 0.95relationshipwhy Refined copper
Transformer windings are conductors carrying the full load current. Copper is the standard winding material for its conductivity and mechanical strength; aluminum windings are used in many distribution transformers at a larger cross-section.
- Input coefficient
- hypothesized dependence, no stored coefficient
- Bottleneck status
- Copper: insufficient data, severity 85.48 (2026-07-01) Recommended
- Lead time
- 120 to 240 months (expansion lead time, bottlenecks table) Humanity Vector Recommended
- Substitutes
- Primary aluminum: Aluminum windings, larger cross-section for the same current; common in distribution transformers.
- Response under way
- Heavy Industry Pulse: no value stored Derived
No pulse stored yet (corporate.recompute_pulses has not run, or no facts are stored).
- Also needed by
- Residential construction
depth 4 Primary aluminumcommodity
Established relationrequires inputstrength 0.4confidence 0.9relationshipwhy Primary aluminum
Aluminum is the alternative winding conductor and is used in tank and structural parts; it is lighter and cheaper per unit of conductance but needs more volume.
- Input coefficient
- hypothesized dependence, no stored coefficient
- Bottleneck status
- no bottleneck row stored
- Lead time
- no cited lead time
- Substitutes
- Refined copper: Copper windings where space and strength matter.
- Response under way
- Heavy Industry Pulse: no value stored Derived
No pulse stored yet (corporate.recompute_pulses has not run, or no facts are stored).
- Also needed by
- no other stored dependent
depth 4 Grain-oriented electrical steelmaterial
Established relationrequires inputstrength 0.9confidence 0.95relationshipwhy Grain-oriented electrical steel
Transformer cores are stacked laminations of grain-oriented electrical steel chosen for low core loss. Amorphous alloy cores are a partial alternative in distribution units; there is no drop-in core material for large power transformers.
- Input coefficient
- hypothesized dependence, no stored coefficient
- Bottleneck status
- no bottleneck row stored
- Lead time
- no cited lead time
- Substitutes
- none stored on this edge
- Response under way
- Heavy Industry Pulse: no value stored Derived
No pulse stored yet (corporate.recompute_pulses has not run, or no facts are stored).
- Also needed by
- no other stored dependent
Arithmetic on stored values; ranges follow the inputs
Required additions
| Quantity | Value | Class | Basis |
|---|---|---|---|
| Required annual addition Current level unavailable, so nothing to divide. | unavailable | Derived | no ids |
| AI accelerator needed per year | not estimated | hypothesized dependence | rel:08f70642-c436-4e33-adc2-2680aa46efe1 |
| Data center needed per year | not estimated | hypothesized dependence | rel:24d7b94a-cd57-4c73-8cc8-524f588ee6b4 |
| High bandwidth memory needed per year | not estimated | hypothesized dependence | rel:7456677a-bd22-4daf-9258-a2e3a5f6ab74 |
| Semiconductor fabrication plant needed per year | not estimated | hypothesized dependence | rel:521b668e-efb1-4c6f-8d93-8db729546bd8 |
| Transmission and distribution grid needed per year | not estimated | hypothesized dependence | rel:d7726fc3-c40d-4559-9ddd-2aa045ad8aac |
| Electricity system needed per year | not estimated | hypothesized dependence | rel:5e087bc1-ba28-40d7-9a68-f1b7f9cb4460 |
| Industrial process water needed per year | not estimated | hypothesized dependence | rel:eda83651-a77e-4333-8e06-fa9192e885a9 |
| Lithography tools needed per year | not estimated | hypothesized dependence | rel:a7d21a38-73e2-4415-a681-94d5894b9e09 |
| Power transformer needed per year | not estimated | hypothesized dependence | rel:0f9233ff-917a-41b9-bf03-ffb05c829ced |
| Renewable freshwater needed per year | not estimated | hypothesized dependence | rel:3cf7af04-3b0e-4638-bd91-8379b0f4dfa1 |
| Refined copper needed per year | not estimated | hypothesized dependence | rel:57478abe-2355-4de1-8b4d-be19c58b7045 |
| Primary aluminum needed per year | not estimated | hypothesized dependence | rel:aa5d07e3-221c-4748-b5bc-d3b871fcb961 |
| Grain-oriented electrical steel needed per year | not estimated | hypothesized dependence | rel:361814d4-9451-4107-b575-92a3073b6956 |
Capacity pipeline and corporate pulses that map to the walk
Current pipeline and responses
AI Infrastructure Pulse0.7014fractionDerived2026-07-03
8 companies contributed to the stored pulse.
Semiconductor Capacity Pulse0.0553fractionDerived2026-07-03
10 companies contributed to the stored pulse.
Grid Investment Pulse0.3521fractionDerived2025-12-31
9 companies contributed to the stored pulse.
Heavy Industry Pulseno valueDerived
No pulse stored yet (corporate.recompute_pulses has not run, or no facts are stored).
Only where a coefficient and a sourced unit cost both exist
Capital
Not estimated: no coefficient plus unit cost pair is stored for this objective.
Downstream systems that compete for the same inputs; the engine does not rank them
Tradeoffs
Refined copper depth 4
Also needed by Residential construction (requires input). Pursuing this objective draws on an input those systems also need; which should have it is a choice this site does not make.
What is unavailable, hypothesized or modeled
Uncertainty
- Current level: No stored series measures global compute or data-center power capacity yet; the catalog has no observed metric for it. The level is marked modeled and unavailable, the required pace is not computed, and the AI infrastructure corporate pulse stands in as the only response signal.
- Recent pace: No observations stored.
- Forecast pace: This metric is not a Forecast Lab target.
- AI accelerator: hypothesized dependence, no stored coefficient, so no quantity of it is computed.
- Data center: hypothesized dependence, no stored coefficient, so no quantity of it is computed.
- High bandwidth memory: hypothesized dependence, no stored coefficient, so no quantity of it is computed.
- Semiconductor fabrication plant: hypothesized dependence, no stored coefficient, so no quantity of it is computed.
- Transmission and distribution grid: hypothesized dependence, no stored coefficient, so no quantity of it is computed.
- Electricity system: hypothesized dependence, no stored coefficient, so no quantity of it is computed.
- Industrial process water: hypothesized dependence, no stored coefficient, so no quantity of it is computed.
- Lithography tools: hypothesized dependence, no stored coefficient, so no quantity of it is computed.
- Power transformer: hypothesized dependence, no stored coefficient, so no quantity of it is computed.
- Renewable freshwater: hypothesized dependence, no stored coefficient, so no quantity of it is computed.
- Refined copper: hypothesized dependence, no stored coefficient, so no quantity of it is computed.
- Primary aluminum: hypothesized dependence, no stored coefficient, so no quantity of it is computed.
- Grain-oriented electrical steel: hypothesized dependence, no stored coefficient, so no quantity of it is computed.
- Capital not estimated for AI accelerator, Data center, High bandwidth memory, Semiconductor fabrication plant, Transmission and distribution grid, Electricity system, Industrial process water, Lithography tools, Power transformer, Renewable freshwater, Refined copper, Primary aluminum, Grain-oriented electrical steel: no coefficient plus unit cost pair is stored for them.
Stated, versioned, and shown before the numbers are trusted
Assumptions
- Base year 2026: the required pace divides the whole gap evenly over 9 years.
- The constraint walk follows requires_input, depends_on and constrains edges to depth 4; an entity reached twice is listed once at its shortest depth. Anything the curated graph does not store is not listed.
- Lead times are order-to-delivery or discovery-to-production ranges from the cited documents; they do not include financing or permitting unless the source says so.