#3 of 30 · Device-first AI
Apple (Apple Intelligence)
Small on-device models by default; the cloud part is opaque.
In plain words
How scores workApple scores 70 out of 100 — leader. Discloses real numbers and runs mostly on clean power. Still has a footprint, but you can see it. Confidence in this score is good confidence (B).
What helps
- Efficiency · Most Apple Intelligence tasks run on a small on-device model; server requests go to Private Cloud Compute on Apple silicon rather than GPU clusters.
- Carbon · Corporate operations carbon neutral since 2020; 100% renewable electricity for facilities since 2018; emissions down more than 60% versus 2015.
- Water · All eight owned data centres certified to the Alliance for Water Stewardship standard; over 55% of corporate freshwater withdrawals replenished.
What hurts
- Efficiency · No per-request energy figure has been published, so the score rests on architecture, not measurement.
- Transparency · Detailed annual Environmental Progress Report, but nothing model- or request-level for Apple Intelligence, and no figure for the size of Private Cloud Compute.
What nobody knows
- Per-prompt energy disclosed
- Per-prompt water disclosed
- Training emissions in model cards
- Water-use effectiveness reported
Apple Intelligence routes most requests to a ~3B-parameter model running on the device, falling back to Private Cloud Compute on Apple silicon. Apple's corporate operations have run on 100% renewable electricity since 2018 and the company reports emissions down more than 60% from a 2015 baseline, but it publishes nothing AI-specific and began extending Private Cloud Compute to Google Cloud in 2026.
Key figures
Per-prompt and fleet metrics. Hover a title for what it means; open “How we got this number” for the working.
ecogpt.co estimate; no company figure.
How we got this number
Most Apple Intelligence requests run a ≈3B-parameter model on the device's own neural engine; measured sub-10B server models median 0.042 Wh, and a phone NPU does the same work for less. Private Cloud Compute requests are the minority. Uncertainty roughly ±3×.
On-site cooling only, so comparable with Google and OpenAI's figures.
How we got this number
0.02 Wh per prompt × 0.2 L/kWh (estimated WUE for Apple's AWS-certified water-stewardship sites). Counting the water evaporated to generate the electricity adds ≈0.04 mL.
Location-based: what the grid actually emitted. Renewable certificates would bring the market-based figure near zero.
How we got this number
0.02 Wh × 390 gCO₂/kWh (US grid, location-based).
How we got this number
Apple reports gross emissions more than 60% below its 2015 baseline of 38.4 Mt, which implies roughly 15 Mt. Almost all of it is product manufacturing, not AI.
How we got this number
Apple publishes no WUE. Its data centres hold Alliance for Water Stewardship certification and several use outside-air cooling; 0.2 L/kWh matches peers with similar designs.
Hardware & materials
The chips, power, copper and future e-waste behind the fleet — the part sustainability reports leave out
GPUs, TPUs or custom accelerators the company runs. Analyst figures are in H100-equivalents, company figures in physical chips.
How we got this number
Apple discloses nothing about PCC's size. A dedicated server plant in Houston has been shipping since October 2025 and PCC is expanding into Google Cloud, which points to a fleet in the low hundreds of thousands of nodes. Uncertainty ±5×.
Megawatts the AI fleet pulls. One megawatt runs roughly 750 US homes.
How we got this number
100,000 accelerators × 1.3 kW each, all-in. Fleet power implied by the chip count.
Terawatt-hours per year. One TWh is a year of electricity for about 95,000 US homes.
How we got this number
130 MW × 8,760 h × 80% utilisation. Compute fleet only.
CO₂e emitted making the chips and servers before they answer a single prompt.
How we got this number
100,000 accelerators × 514 kg CO₂e each (chip plus share of the host server). Emitted in manufacturing, before the first prompt is served.
Tonnes of server hardware that will be retired in three to five years.
How we got this number
100,000 accelerators × 17 kg of server hardware each. Tomorrow's e-waste unless reused; generative AI could produce 2.5 Mt of it a year by 2030.
Tonnes of copper in cabling, busbars and transformers.
How we got this number
130 MW × 27 t of copper per MW (cabling, busbars, transformers).
Diesel backup units or gas turbines at the data centres, permitted or observed.
How we got this number
130 MW ÷ 2.6 MW per generator, the average size permitted at US data centres. Permit counts are not published fleet-wide; this is the number a fleet this size normally carries.
Share of retired servers and parts given a second life.
Kilograms of antimony-equivalent, the life-cycle unit for scarce minerals consumed.
Score breakdown
Each pillar is 0–100; weights are fixed across the index
- Most Apple Intelligence tasks run on a small on-device model; server requests go to Private Cloud Compute on Apple silicon rather than GPU clusters. AppleHelp Net Security
- No per-request energy figure has been published, so the score rests on architecture, not measurement. Apple
- Corporate operations carbon neutral since 2020; 100% renewable electricity for facilities since 2018; emissions down more than 60% versus 2015. Apple
- Private Cloud Compute now also runs in Google Cloud regions, whose carbon-free share varies by location. Help Net Security
- All eight owned data centres certified to the Alliance for Water Stewardship standard; over 55% of corporate freshwater withdrawals replenished. Apple
- Detailed annual Environmental Progress Report, but nothing model- or request-level for Apple Intelligence, and no figure for the size of Private Cloud Compute. Apple
- Apple 2030: carbon neutral across the entire value chain, with a 75% absolute reduction and high-quality removals for the remainder. Apple
- No AI-infrastructure permit disputes or community water conflicts on record. Apple
Models
What this company ships, and what each prompt costs
| Model | Weights | Energy / prompt | Water | Training CO₂e |
|---|---|---|---|---|
| Apple Foundation Models (on-device) ≈3B on-device; server model on Apple silicon | Closed | 0.02 Wh Estimateecogpt.co estimate. Our own estimate from published factors. The working is shown next to the number; treat it as an order of magnitude, not a measurement. | 0.004 mL Estimateecogpt.co estimate. Our own estimate from published factors. The working is shown next to the number; treat it as an order of magnitude, not a measurement. | 1,000 t Estimateecogpt.co estimate. Our own estimate from published factors. The working is shown next to the number; treat it as an order of magnitude, not a measurement. |
Where the compute lives
Named sites from company disclosures and press reporting
- MaidenNorth Carolina, USA
- MesaArizona, USA
- RenoNevada, USA
- ViborgDenmark
- Houston server plantTexas, USA · Builds PCC servers; shipping since Oct 2025.
- Private Cloud Compute on Google CloudMultiple regions · Third-party capacity from 2026.