# Cloud Region Capacity Planning

Compare regions for the next cloud capacity expansion using power availability, latency constraints, GPU supply, and timing assumptions before quotas, routing constraints, or utility gates force a poor placement decision.

For cloud infrastructure and capacity planning teams deciding where to expand next.

Which regions can support the next capacity expansion without power, latency, or GPU bottlenecks?

## Sample region comparison

Review one region comparison view showing power timing, latency fit, supply tightness, and operating cost tradeoffs across candidate regions.

### Illustrative region comparison

Example view for a next-region expansion decision.

Example question

Which region gives the best balance of power timing, GPU access, latency fit, and operating cost for the next expansion move?

| Region               | Power lead time | Latency fit                      | GPU supply | Egress cost | Watchout                                                   |
|---------------------|-----------------|----------------------------------|------------|-------------|-----------------------------------------------------------|
| Northern Virginia    | 9-12 months     | Strong for East Coast demand     | Moderate   | $0.026/GB  | Utility power timing is the main gating factor.           |
| Dallas               | 4-6 months      | Moderate for national mix        | Tight      | $0.020/GB  | GPU supply alignment is weaker than power timing.        |
| Columbus             | 6-8 months      | Strong for Midwest and East      | Available   | $0.018/GB  | Best balance, but resilience needs a second-region pair.  |

The full scorecard expands this comparison with ranking logic, constraint notes, and the assumptions behind each region recommendation.

## What we test

- region-level power availability
- latency constraints
- GPU availability and supply constraints
- demand and expansion timing assumptions

## What the scorecard includes

- a ranked comparison of which regions are safer or riskier for the next expansion move
- the main capacity constraints behind each region score
- a scorecard that can be circulated to infrastructure and planning teams

## Region Capacity Scorecard

Which regions can support the next capacity expansion without power, latency, or GPU bottlenecks?

## Preview the variables behind the scorecard

These cards show the outcome measures, conditions, and levers tracked in the region capacity scorecard.

## Key outcome measures

**Latency SLO Attainment Share**

Share of traffic or minutes meeting the latency SLO across the region mix.

percent

**Capacity Deficit Index**

Normalized shortfall of available capacity vs. demand forecast plus provisioning buffer.

index (0–1)

**Effective Capacity Cost per GPU‑hour**

Blended $/GPU‑hour from on‑demand, reserved, spot, and unused commitment effects.

USD/gpu hour

**Multi‑region Resilience Index**

Probability‑scaled index of serving demand under AZ/region failure scenarios and observed failover success.

index (0–1)

**Compliance Coverage Share**

Portion of demand served from regions meeting data‑residency/regulatory constraints for the workload.

percent

**Egress Cost per GB**

Effective $/GB for inter‑region/Internet data transfer under the current region plan.

USD/gb

**Power Availability Lead Time (months)**

Expected months to secure incremental MW capacity in target regions (provider + interconnect milestones).

months

**Chip Supply Alignment Index**

Fit between GPU supply (deliveries/allocations) and planned demand across regions and priority tiers.

index (0–1)

## Key conditions behind the comparison

**GPU Allocation Inventory**

GPUs available for allocation (delivered and commissioned but not yet assigned).

gpu

**Capacity Request Backlog (GPU)**

Outstanding GPU capacity requests awaiting provider approval/quota increase.

gpu

**Power‑permit Queue Backlog (MW)**

MW awaiting permitting/interconnect approval in targeted regions/campuses.

MW

**Savings Plan Unused Commitment (USD)**

Cumulative unused Savings Plan commitment within the current accrual period.

USD (millions)

**RI Unused GPU‑hours**

Purchased reserved GPU‑hours not applied to usage in the period.

gpu hour

**Workload Placement Backlog (requests)**

Pending placement requests waiting for region assignment given constraints.

requests

## Levers that can change the scorecard

**RI/Savings Plan Coverage Target Share**

Target fraction of compute cost/hours to cover with RIs or Savings Plans.

0–1 ratio

**Redundant Regions Count**

Number of active regions provisioned for failover (N‑way).

count

**Traffic Steering Aggressiveness Index**

Degree to which routing favors best‑latency/lowest‑cost regions among eligible choices.

index (0–1)

**Data Locality Enforcement Index**

Strictness of data‑residency enforcement in placement and routing.

index (0–1)

**Provisioning Buffer Target Share**

Planned headroom over forecast to absorb demand and lead‑time variance.

0–1 ratio