Human-only workplace allocation
vs.
2.40 gCO₂eAI-assisted, standard grid
2026 Impact Report Pre-assurance
Intelligence
for everyone.
Building advanced intelligence for broad human benefit—responsibly governed, environmentally sustainable and widely shared.
900M+weekly active users
15BAPI tokens per minute
4.0TWh annual electricity estimate
2.8MtCO₂e central footprint estimate
Leadership message01 / 06
Capability creates responsibility.
Our mission is to ensure that general-purpose artificial intelligence benefits all of humanity.
By releasing ChatGPT directly to the public, we helped establish a new expectation: advanced intelligence can be broadly usable and shareable rather than reserved for a narrow institutional elite.
We measure progress not only by how capable our systems become, but by whether those capabilities are broadly useful, responsibly governed, environmentally sustainable and translated into durable human benefit.
Scale also creates substantial physical impacts. We are advancing efficiency, additional clean power, responsible infrastructure, supplier decarbonization, water stewardship and transparent community engagement alongside our products.
Our impact thesis02 / 06
Broad access to advanced intelligence can expand human capability, accelerate discovery and create opportunity at the scale of foundational public infrastructure.
Access
Intelligence as infrastructure
Making frontier capability useful across languages, markets and institutions.
Responsibility
Safety through the lifecycle
Preparedness, model behavior, product safeguards, privacy and security.
Stewardship
Benefit per physical resource
Efficiency, clean energy, resilient infrastructure and transparent accounting.
Scale and footprint03 / 06
Growth, measured in full.
Our first enterprise-scale baseline uses transparent assumptions and wide uncertainty ranges while controlled primary data matures.
Interactive estimate
Annual electricity demand
4.02 TWh
Pre-assurance scenario estimate. The wide range reflects workload, model, hardware, utilization and training uncertainty.
Central pre-assurance estimate
Modeled Scope 1–3 footprint
2.78 MtCO₂eScope 15 kt
Scope 20.44 Mt
Scope 3 compute0.98 Mt
Scope 3 capital1.25 Mt
Other Scope 30.10 Mt
Conditional strategic pathway
The potential scale of 5 GW firm clean power
At a 90% capacity factor, 5 GW could generate 39.42 TWh annually. Any avoided emissions would be reported separately from our Scope 1–3 inventory.
Central 2026 demand4.0 TWh
5 GW annual generation39.42 TWh
Cognitive substitution04 / 06
More useful output.
Less physical intensity.
Centralized computation can augment drafting, coding, analysis and design at very low incremental emissions per task. It is a mechanism to measure—not an automatic offset.
Human-only workplace allocation
vs.
4.50 gCO₂eAI-assisted, standard grid
Conventional production scenario
vs.
0.375 kgCO₂eAI-assisted concept generation
Scenario comparisons exclude human review, devices, training, capital goods and rebound. See methodology §10.4.
2050 scenarios05 / 06
Toward a global
intelligence dividend.
Four scenarios test how access, productivity, leisure capture, system emissions and rebound could interact. They are not forecasts.
Constrained diffusion
AI-active workers3.0bn
Productivity uplift8%
Leisure / worker1.6 days
Net climate effect−46 to −11 Mt
Managed abundance
AI-active workers3.6bn
Productivity uplift18%
Leisure / worker10.0 days
Net climate effect+26 to +41 Mt
Universal intelligence dividend
AI-active workers4.1bn
Productivity uplift30%
Leisure / worker25.2 days
Net climate effect+164 to +184 Mt
Rebound-intensive downside
AI-active workers4.1bn
Productivity uplift25%
Leisure / worker2.2 days
Net climate effect−143 to −43 Mt
Risk and resilience06 / 06
Climate resilience
Test the exposure
See how electricity and carbon-price assumptions affect the central 4.0 TWh load and 1.41 MtCO₂e compute estimate.
Go deeper