Fueling the Growth in AI
Fueling the Growth in AI
The stock performance of NVIDIA Corp. (NASDAQ:NVDA) illustrates the fast-growing demand for artificial intelligence (AI) hardware and software products. This stock has appreciated over 200% in the last year, but its growth projects a significant developing need for additional electricity to power data centers, required to both train the next generation of AI products and to run the models for the growing number of users.
NVIDIA Corp.’s One Year Stock Performance1
Powering AI Data Centers
The power requirements for AI data centers vary, given the number and types of servers within, but a rough estimate is around 10-20 megawatts (MW) of power required per data center. For this article, I will assume an average power requirement of 15 MW per data center. Note that 15 MW reflects the demand for electricity, multiplying this by the hours that the center is run results in daily electricity consumption of 360 megawatt hours (MWH), or approximately 130,000 MWH per year.
At the same time that this new demand for electricity is growing, in the U.S. oil and gas operators are flaring or venting (i.e. releasing into the atmosphere) large quantities of natural gas.
Natural Gas Being Flared or Vented
According to the U.S. Energy Information Administration (EIA), in 2022, the U.S. vented and flared approximately 270.9 billion cubic feet (BCF) of natural gas.2 This amount, although significant, is down from a peak volume of 539.5 gas vented and flared in 2019, as environmental concerns and legislation have impacted this practice.
What if we could capture all of this flared gas and use it to generate electricity in a modern combined cycle gas turbine power plant? How many new AI data centers could be powered?
Converting Gas to Electricity
Modern combined cycle gas turbine (CCGT) generating plants have an efficiency rating of about 55%.3
Natural gas being flared or vented has a heat content around 1,036 BTU/cubic foot (or 1.036 MMBTU/MCF). Therefore, assuming that U.S. operators are flaring or venting a total of 270.9 BCF each year, this means that they are also releasing 280,652,400 MMBTU’s of natural gas heating capacity each year that could be used to generate electricity if captured.
At the assumed efficiency rating above, this amount of gas currently being flared or vented could produce approximately 45.2 million MWH of electricity in a year.
Number of Data Centers
Now, let’s calculate how many data centers we could power with the captured natural gas currently being flared or vented in the U.S. Assuming an average power consumption per data center of 15 MW, this implies that facility would require 131,400 MWH/year of electricity.4
So, if all the natural gas being flared at wells in the U.S. were someday captured and used to generate electricity through an efficient CCGT generating station, it could potentially power around 344 new AI data centers.
Key Takeaways
Keep in mind that this is a rough estimate, as actual energy requirements and efficiencies may vary. Also, capturing and utilizing flared gas is a complex process with infrastructure needs. However, this example illustrates how capturing natural gas at well sites, currently being flared or vented, and using it to produce electricity could help offset the expected future electricity demand being driven by new and existing data centers, as well as other power demands resulting from the growth in AI vendors, programs, and users.
WTI Strip Prices Increase
Spot prices and futures prices for the West Texas Intermediate (WTI) contract remained relatively unchanged in the near term and increased approximately $1.00 over the longer term.
WTI Strip Prices - One Month Change
As shown, the oil price curve remains in a state of “backwardation,” reflecting the market’s expectation of lower future spot prices.
Oil Price Outlook
The price distribution below shows the crude oil spot price on June 12, 2024, as well as the predicted crude oil prices based on options and futures markets. Light blue lines are within one standard deviation (σ) of the mean, and dark blue lines are within two standard deviations.
WTI Crude Oil $/BBL
Based on these current prices, the markets indicate there is a 68% chance oil prices will range from $67.50 and $87.00 per barrel in mid-September 2024. Likewise, there is roughly a 95% chance that prices will be between $53.50 and $103.00. By mid-November 2024, the one-standard deviation (1σ) price range is $64.50 to $89.50 per barrel, and the two-standard deviation (2σ) range is $47.50 to $113.50 per barrel.
Key Takeaways
Remember that option prices and models reflect expected probabilities, not certain outcomes, but that does not make them any less useful. Throughout most of 2023 and 2024, crude oil spot prices have primarily fluctuated within the range of $70 to $90 per barrel. During that time, we observed general increases in futures price volatilities as prices neared the upper bound of that range, as evidenced by the futures price ranges observed. For mid-November 2024 pricing as of June 12, 2024, the 1σ range had a spread of $25.00 per barrel, and the 2σ range has a spread of $66.00 per barrel. For comparison, in 2022 we observed 1σ and 2σ price ranges in excess of $65.00 and $150.00, respectively.
- "NVIDIA Corp," CNBC.
- "Natural Gas Gross Withdrawals and Production," U.S. Energy Information Administration.
- "Combined-cycle gas turbines," Ipieca.
- 131,400 MWH = 15 MW * 24 hours per day * 365 days per year.