Can a Solar and Battery Microgrid Power Elon Musk’s Gigafactory of Compute?

Can a Solar and Battery Microgrid Power Elon Musk’s Gigafactory of Compute?

In this article, we estimate the required capacity of a solar and battery microgrid capable of powering the Gigafactory of Compute. We also estimate its upfront cost and payback period.

‍

The AI industry has created a growing demand for data center infrastructure. Advanced language models like ChatGPT and Gemini are trained in data processing “clusters” with thousands of graphics processing units (GPUs). Elon Musk plans to build a 100,000-GPU cluster for his company xAI in Memphis, Tennessee, which will be used to train the next generation of AI models. Specifically, Musk plans to use 100,000 H100 GPUs from Nvidia. This project has been dubbed the “Gigafactory of Compute” due to its scale.

‍

Gigafactory of Compute: Understanding the Energy Needs

Each Nvidia H100 GPU consumes up to 700 watts, leading to an astounding 70 MW when 100,000 units operate at full capacity. With an 80% utilization rate, the GPU cluster would require 490,560 megawatt-hours (MWh) annually, equivalent to the energy consumption of over 46,400 American homes. However, GPUs only represent part of the power input of an AI data center. Considering additional equipment like networking devices, storage servers, CPUs, and cooling systems, total consumption skyrockets to 1.59 terawatt-hours (TWh) annually, akin to the energy demand of 150,000 homes in the US (SemiAnalysis, US Energy Information Administration).

‍

How Many Solar Panels Are Needed for an AI Data Center with 100,000 GPUs?

To power this facility with solar energy, we must consider Memphis’ solar output, which averages 1,500 kWh annually per kW of installed capacity. Thus, generating 1.59 TWh/year necessitates a 1,060 MW solar array. At $980,000 per MW, this setup costs approximately $1,038.8 million. Assuming 500 W solar panels, which are common in large-scale applications, the array would consist of 2.12 million panels (World Bank Global Solar Atlas, Solar Market Insight Report).

Energy Storage Needs of the 100,000-GPU Data Center

Energy storage is critical for continuous operation. An AI data center like the Gigafactory of Compute requires a minimum sustained power of around 150 MW (SemiAnalysis). The required storage capacity in kWh depends on the number of hours of backup power the owner wants to achieve.

‍

Many of the megawatt-scale batteries currently in operation are designed as one- or two-hour systems. Here, we will also consider a scenario with 24-hour backup power. Assuming a battery cost of $200/kWh, here is the upfront cost of each option:

• 1-hour backup: 150,000 kWh, costing $30 million.
• 2-hour backup: 300,000 kWh, costing $60 million.
• 24-hour backup: 3,600,000 kWh, costing $720 million.

Now we add these values to the cost of the 1.06-GW solar array, which is $1,038.8 million.

Total Cost of the Solar Array and Energy Storage System

• 1.06 GW solar array with 1-hour backup: $1,068.8 million
• 1.06 GW solar array with 2-hour backup: $1,098.8 million
• 1.06 GW solar array with 24-hour backup: $1,758.8 million

Here we compare the annual electricity cost of a 100,000-GPU AI data center in Tennessee and California, considering the annual consumption of 1.59 TWh and the average industrial tariffs in each state:

• Electricity cost in Tennessee: $98.1 million per year
• Electricity cost in California: $306.87 million per year

Based on these figures, we can estimate the payback period of the solar and battery microgrids in each state. We consider the three configurations discussed in the previous section:

‍

In conclusion, even with the sheer size of the gigafactory, a solar and battery microgrid could be designed to supply the power needed, and with the right prices, the buyback period could be within a reasonable time frame.

‍