# Siting, regional infrastructure and access to compute

Canonical: https://semiaifoundry.com/learn/atlas/regional-siting-access/
Revision: 2026-10-07
Rights: https://semiaifoundry.com/ai-atlas/licenses.html

Global AI infrastructure · Atlas v1.3.0

# Siting, regional infrastructure and access to compute

Connects location choices to service paths, resources and who can actually use capacity.

[Explore in Atlas](https://semiaifoundry.com/ai-atlas/#regional-siting-access)Facility phases to region and world; geography schematic

Learning objectiveExplain siting and access tradeoffs without equating investment, geography and available service.

## The mechanism

Data-center siting combines electricity readiness, fiber routes, latency, climate, water, staffing and the needs of its users. A project can improve connectivity or electrical infrastructure while remaining under construction. Google’s India hub announcement explicitly links the site with surrounding infrastructure work, while EuroHPC’s gigafactory call describes a public access and procurement mechanism. These are different evidence categories from private installed hardware. Owner, operator and access model are separate attributes. Announced investment does not equal completed equipment, and nearby compute is not automatically available to every local user.

1.  Define the users, data paths and service objectives that location must support.
2.  Compare actual electrical, network and cooling readiness for the same time horizon.
3.  Identify physical operators and the mechanism through which users receive compute access.
4.  Model resource and latency tradeoffs with explicitly chosen scenario weights.
5.  Keep construction announcements, funding commitments and operational service evidence separate.

## Science and assumptions

### Conditional site score

$$
\mathrm{Score}=\sum_i w_i z_i
$$

The scenario score sums normalized criterion values multiplied by declared preference weights.

This is a preference-dependent decision aid, not an objective scientific ranking. Weight normalization is a chosen convention, not implied by the sum.

Symbols and units

Score

Conditional scenario preference score

dimensionless

wᵢ

Declared preference weight for criterion i

dimensionless

zᵢ

Normalized metric value under the stated transformation

dimensionless

i

Criterion index

integer index

zi is a normalized scenario metric and wi a declared preference weight; the score is dimensionless.

Assumptions A teaching decision aid, not an objective scientific ranking; normalization and uncertain inputs must be visible.

## Engineering example

Synthetic choice: Site A offers lower network delay but opens a year later; Site B opens sooner with a longer path. A batch-training user and an interactive-serving user can reasonably choose differently.

## Distinctions and tradeoffs

Investment dollars are not operating megawatts, and public access programs do not imply every listed facility is already commissioned.

Proximity can reduce network delay while a different region may have earlier power availability; access rules and reservations can dominate physical nearness.

Useful compute capacity depends on admission and data paths, so geographical expansion matters through more than building counts.

## Specifications and interfaces

Access states

Owned, reserved, allocated and available

Physical capacity and user access are separate ledgers.

Location objective

Latency, readiness, resilience and resources

Weights belong to an explicit scenario, not a universal ranking.

### Interfaces

Workload and regional requirements Input

User locations, resource limits and access needs

Siting/access decision Output

A bounded choice with evidence-backed tradeoffs

## Check your understanding

### Which question distinguishes accessible capacity from physical capacity?

1.  How many countries appear in the press release?
2.  Who can obtain an allocation, under what conditions and date?
3.  How large is the site’s sign?

Answer and reasoning

Who can obtain an allocation, under what conditions and date?

Access depends on reservations, procurement and admission policy beyond the presence of equipment.

## Evidence and scope

No jurisdiction-specific legal advice or exhaustive regional census is provided. Public project and procurement statements support dated examples only.

Browse the numerical library; no current calculator directly models this lesson’s central mechanism.

## Sources

-   [Google AI Hub construction milestone in India](https://blog.google/intl/en-in/google-ai-hub-scaling-indias-ai-infrastructure-and-community-impact/)
-   [The EuroHPC Joint Undertaking launches the AI Gigafactories Call](https://www.eurohpc-ju.europa.eu/eurohpc-joint-undertaking-launches-ai-gigafactories-call-2026-07-30_en)

[Published lesson data](https://semiaifoundry.com/ai-atlas/lesson-data/regional-siting-access-797dbbfb173b8938.json) · [Atlas releases and licensing](https://semiaifoundry.com/ai-atlas/licenses.html)

## Continue exploring

### Foundations

-   [Facility phases and geographical evidence](https://semiaifoundry.com/learn/atlas/capacity-phase-geography/)
-   [Cluster scheduler and resource allocation](https://semiaifoundry.com/learn/atlas/scheduler/)

### Related lessons

-   [Grid interconnection and deliverable electrical capacity](https://semiaifoundry.com/learn/atlas/grid-interconnection-readiness/)
-   [Job admission, topology and resource fragmentation](https://semiaifoundry.com/learn/atlas/topology-job-admission/)