Grok 2 Environmental Impact

Energy Consumption

Grok consumes approximately 0.5 Wh per query, making it relatively efficient for a full-size model. xAI's Colossus supercomputer in Memphis houses approximately 150,000 H100 GPUs, 50,000 H200 GPUs, and 30,000 GB200 GPUs — one of the largest GPU clusters in the world — but the energy per individual query remains modest.

Power Source & Carbon

Grok runs exclusively at xAI's Colossus facility in Memphis, Tennessee, built in just 122 days starting September 2024. The facility consumes nearly 2 gigawatts of power. Memphis's grid is served by the Tennessee Valley Authority (TVA), which generates power from a mix of nuclear (~40%), natural gas (~25%), hydro (~10%), coal (~15%), and renewables (~10%). The facility has faced scrutiny for initially running on diesel generators and for air quality concerns from local residents.

Water Usage

Grok uses approximately 3 mL of water per query. The Memphis facility uses evaporative cooling towers, drawing from the local water supply. Memphis sits above the Memphis Sand aquifer, one of the largest artesian aquifer systems in the world, which provides the city's water. The rapid scale-up of xAI's facility has raised local concerns about water resource impact.

How much energy does Grok 2 use per query?

Each Grok 2 query consumes approximately 0.50 Wh of energy. This is 2x more than a traditional Google search (~0.3 Wh).

What is Grok 2's carbon footprint?

Based on the carbon intensity of Memphis, Tennessee (Colossus), each query produces approximately 0.17 g of CO2. The grid in this region has a carbon intensity of 420 g CO2/kWh with 30% renewable energy.

How much water does Grok 2 use?

Each query consumes approximately 3 mL of water, primarily used for cooling the data centers that process the request.

How does Grok 2 compare to a Google search?

A Grok 2 query uses 2x more than a Google search in terms of energy. A Google search uses approximately 0.3 Wh, while Grok 2 uses 0.50 Wh.

Architecture

Transformer Mixture-of-Experts (decoder-only)

Parameters

314B

Context window

128,000 tokens

Release date

2024-08-13

Open source

No

Training data cutoff

2024-07