Scheduling for carbon, not just cost

WORKLOAD: dev environment scheduled off
COST SAVING:  ~64% × dev compute spend
CARBON SAVING: ~64% × dev compute carbon

The same percentage applies to both. Scheduling is mathematically
a "double-lever": it cuts both compute hours (cost) and compute
hours (carbon) proportionally.

SCENARIO 1: Internal tool with low absolute cost but high compute
  Cost: small (let it run; not worth the operational change)
  Carbon: significant in the company's annual footprint
  → Schedule for carbon, even if cost case is weak

SCENARIO 2: ESG-driven team or org
  Reporting requires carbon footprint reductions
  Scheduling is a measurable, defensible lever
  → Apply scheduling for carbon as a primary motivation

SCENARIO 3: Investor / partner sustainability commitments
  Net-zero or reduction targets in stockholder messaging
  Scheduling contributes to those targets
  → Scheduling decisions are carbon-driven

SCENARIO 4: Public-sector or regulated industry
  Mandatory carbon reporting
  Scheduling reduces measured emissions
  → Required, not optional

EVALUATION ORDER:
  1. Cost savings (primary driver in most orgs)
  2. Carbon savings (secondary, increasingly important)
  3. Latency / availability (constraints: never violate)
  4. Engineering effort (consider operational complexity)

A schedule that saves BOTH is the strongest case.
A schedule that saves cost ONLY is acceptable for non-ESG-driven orgs.
A schedule that saves CARBON ONLY is justifiable if ESG matters AND
the operational overhead is minimal.
A schedule that hurts user experience is wrong regardless.

PATTERN A: Time-shift to renewable peaks
  Workload runs at clean-grid hours
  Example: ML training at 11 AM (solar peak)
  Cost impact: usually neutral
  Carbon impact: 20-40% reduction

PATTERN B: Geographic shift
  Workload runs in cleaner region
  Example: us-east-1 → eu-north-1
  Cost impact: minor variance
  Carbon impact: 70-90% reduction (huge)

PATTERN C: Throttle / batch during dirty hours
  Reduce parallel processing during high-carbon periods
  Example: scale down to 50% capacity during evening peaker hours
  Cost impact: usually neutral (same total compute over time)
  Carbon impact: 10-25% reduction

PER-WORKLOAD CARBON METRIC:
  Compute hours × power draw × carbon intensity = kg CO2

COMPARING:
  Workload A baseline (no schedule): 100 hr/wk × 1.5 kW × 0.35 = 52.5 kg/wk
  Workload A with schedule (60% off): 40 hr × 1.5 kW × 0.35 = 21 kg/wk
  Difference: 31.5 kg/wk saved (60% reduction)
  Annual: 1.6 tons CO2/year per workload

COST:                                CARBON:
─────────────────────────────────────────────────────
Monthly review                       Quarterly review
Quarterly with finance               Annual with leadership
Annual planning                       Annual ESG/sustainability
                                    Industry benchmark annually

DECISION: schedule the dev environment

COST CASE:
  Current cost: $12K/month
  Schedule 60% of hours off → save 60%
  Cost savings: $7.2K/month, $86K/year

CARBON CASE:
  Current carbon: 8 tons CO2/year (us-east-1)
  Schedule reduces by 60% → save 60%
  Carbon savings: 4.8 tons CO2/year

COMBINED CASE (presented to leadership):
  Annual cost savings: $86K
  Annual carbon savings: 4.8 tons CO2
  ESG narrative: contributes to net-zero commitment
  Operational effort: low (schedule + override mechanism)

DECISION: approved. Combined cost+carbon case is strongest.

- Doesn't replace renewable energy investments
  (Carbon-aware is supplementary; not a substitute for
  cleaner energy generation)

- Doesn't eliminate carbon entirely
  (Compute still has some footprint; the goal is reduction,
  not elimination)

- Doesn't justify ignoring cost
  (Still need cost discipline; carbon doesn't excuse waste)

- Doesn't pre-empt the "right-size first, schedule second"
  ordering
  (Right-sized resources matter more than scheduling small
  resources)

PRESENTING carbon savings to leadership:
  - Annual carbon reduction (tons CO2)
  - Equivalent in commonly-understood terms:
    "Saved X tons of CO2 = N cars off the road for a year"
  - Per-team contribution
  - Trend over 3-5 years
  - Comparison to industry benchmarks
  - Specific initiatives that contributed

This makes the savings legible to:
  CEO / board
  Investors
  Customers (some require sustainability disclosure)
  Press / industry coverage

MISTAKE                                   FIX
──────────────────────────────────────────────────────────────────
Quantifying cost but not carbon            Track both side-by-side
                                          in the same report

Carbon math wrong (forgetting              Use full compute-power-
power draw assumptions)                   intensity formula; reference
                                          published instance specs

Reporting carbon without context           Pair with industry / peer
                                          benchmarks for meaning

Treating carbon as separate from cost     They're the same scheduling
                                          decision; report jointly

Over-claiming carbon savings               Use conservative estimates;
                                          audit annually against
                                          actual emissions data

ZOPNIGHT future:
  Per-workload carbon (paralleling per-workload cost)
  Per-team carbon (paralleling per-team showback)
  Scheduling impact on carbon (paralleling cost impact)
  ESG report exports

COMPANY: 250-employee SaaS
INITIATIVE: reduce cloud carbon 25% in 2026

CURRENT STATE:
  Annual cloud spend: $4M
  Annual carbon: 580 tons CO2
  Target: 435 tons (-25%, ~145 tons reduction)

ANALYSIS: Three levers:

  LEVER 1: Schedule all non-prod environments
    Affected workloads: dev, staging, sandbox
    Effective compute hours reduction: ~65%
    Estimated cost impact: -$50K/year
    Estimated carbon impact: -120 tons CO2/year
    Operational effort: 2 weeks one-time setup

  LEVER 2: Region migration for batch (us-east-1 → eu-north-1)
    Affected workloads: ML training, data pipelines
    Cost impact: roughly neutral
    Estimated carbon impact: -85 tons CO2/year
    Operational effort: 4-6 weeks of migration work

  LEVER 3: Graviton migration (eligible workloads)
    Affected: ~30% of x86 workloads
    Cost impact: -$30K/year (10-20% cheaper)
    Estimated carbon impact: -40 tons CO2/year
    Operational effort: 6-8 weeks (testing + migration)

TOTAL POTENTIAL: 245 tons CO2 reduction (over target)
TOTAL COST DELTA: -$80K (some increase) to -$50K (some savings)

PHASING:
  Phase 1 (Q1): Schedule non-prod (quick win, no cost impact)
  Phase 2 (Q2): Migrate batch ML to eu-north-1 (carbon-focused)
  Phase 3 (Q3): Graviton migration (cost + carbon)
  Phase 4 (Q4): Quarterly reporting + ESG presentation

QUARTERLY REPORTING includes:
  Cost savings achieved
  Carbon reduction achieved
  Progress vs target
  Lessons learned + next-quarter plan

WORKLOAD candidate:    __________
  Cost: $______/month
  Carbon (estimated): __________ kg CO2/month

WHY CARBON-MOTIVATED (not cost-motivated):
  □ Cost is too small to justify operational change
  □ ESG / sustainability narrative needs it
  □ Investor / customer requires it
  □ Mandatory carbon reporting

SCHEDULE CANDIDATE:
  Off hours: __________ (e.g., nights + weekends)
  Reduction: __________ % of hours
  Carbon savings: __________ kg/month (= __________ tons/year)
  Cost savings: __________ $/month

DECISION:
  □ Approve (carbon case strong enough)
  □ Defer (operational complexity too high for the carbon benefit)
  □ Combine with another lever (region + schedule)

OPERATIONAL EFFORT estimate:    __________
NEXT STEP:    __________