Brief: $500B Stargate Project

A closer look at the $500 billion Stargate Project venture announced by OpenAI, NVIDIA, Oracle, and Softbank to build multiple data centers and new AI infrastructure across the US.

January 22, 2025 • Est. Reading Time: 5 minutes

Mobius Intel Brief:

Yesterday, President Trump announced a new $500 billion AI infrastructure project dubbed the ‘Stargate Project’ — a joint venture between OpenAI, Oracle, Softbank, and several of the US’ largest tech players.

  • Stargate Project will commit an initial $100 billion to build its first hyperscale data centers and associated infrastructure in Texas before expanding to other states over the next four years. As outlined below, the project’s short timeline and material resource demands further reinforce upside skew to price and reliability risks.

The Early Details

  • According to Oracle co-founder, Larry Ellison: “The data centers are actually under construction. The first of them are under construction in Texas. Each building is half a million square feet. There are ten buildings currently being built, but that will expand to 20 and other locations beyond the Abilene location, which is our first location.”

  • The scale: The Information previously reported negotiations between OpenAI and Oracle for a lease on an Abilene, Texas hyperscale data center that will expand to 1 gigawatt (GW) by mid-2026. Yesterday’s Stargate Project announcement follows a separate report from The Information last year about a potential venture between OpenAI and Microsoft for a series of data centers, including a 5 GW facility called ‘Stargate’ that was planned to come online at the end of the decade.

  • Context: A 1-GW data center like the facilities proposed by the Stargate Project are approximately 4-5x as large as current hyperscale data centers serving ‘Fortune 100’ tenants.

Why It Matters: Gigawatt-Scale & Implied Heat Rates

Initiating the Stargate Project in Texas will provide developers with a friendly regulatory environment and less bureaucratic friction, to be sure. However, proximity to Texas’ abundant energy resources eases tech giants’ primary bottleneck: finding enough compute capacity to train and deploy new AI models.

  • For example, OpenAI’s Sam Altman pointed to this compute constraint during a Reddit AMA held late last October:

As detailed extensively in our research archive, power price and reliability risks from gigawatt-scale data centers and other AI infrastructure skew to the upside.

  • For (simplified) context: a 1-GW data center operating at full capacity for a year consumes 8,766 gigawatt-hours or 8.77 billion kilowatt-hours (kWh). In other words, a single 5-GW facility, such as the one proposed by OpenAI and Microsoft last year, would increase the United States’ total electricity consumption by 1%.

Even when accounting for typical operational load, the resources required to meet this demand are frequently misunderstood or underappreciated. This is more apparent considering Stargate Project’s plans to multiply gigawatt-scale consumption across several campuses in fewer than four years.

Furthermore, a short time horizon of four years will complicate tech giants’ aspirations to power facilities with new, co-located nuclear generation. The list of probable generation resources narrows again when we consider hyperscale data centers’ 24/7/365 reliability requirements — power must be available when called upon.

  • While wind generation can supplement reliable generation from thermal units, claims that a North Texas hyperscale data center can reliably supply high-stakes compute capacity with wind generation fail to recognize the severity of daily variability and long-term seasonality across the Texas wind fleet. Texas wind generation hits its lowest during peak demand months in the summer and winter, often falling below 20% of capacity for consecutive days.

As a result, the lion’s share of large load growth will likely turn to reliable thermal generation via natural gas and semi-predictable solar. Still, questions about the velocity of generation installations remain, and a look at North Texas implied heat rates shows the extent of supply concerns for peak demand months in 2025, 2026, and 2027.

  • According to yesterday’s ERCOT North 345 kV Hub 5 MW Peak and Houston Ship Channel, average Jun-Sep implied heat rates range between 20 and 21 MMBtu/MWh for the next three years. An implied heat rate of nearly 30 MMBtu/MWh for August ‘25 compares to a an average (real) heat rate of 7.7 MMBtu/MWh for US natural gas units and is roughly triple the real heat rate of ERCOT’s average gas peaker.

Takeaway

We reiterate that price and reliability risks skew to the upside.

Please reach out to the Mobius team to explore additional downstream scenarios, distinct regional impacts, and potential portfolio-specific opportunities.

This commentary contains our views and opinions and is based on information from sources we believe are reliable. This commentary is for informational purposes, should not be considered investment advice, and is not intended as an offer or solicitation with respect to the purchase and sale of commodity interests or to serve as the basis for one to decide to execute derivatives or other transactions. This commentary is exclusively intended for Mobius clients and is not considered promotional material.