The Current State of U.S. Data Center Infrastructure

Power Scarcity

Goldman Sachs Research forecasts that U.S. data center demand will reach 45 GW by 2030, driven by a 15% Compound Annual Growth Rate (CAGR) in power demand specifically for AI. To put this in perspective, the entire current installed base of U.S. data centers is being tasked with doubling or tripling its capacity within a decade. The bottleneck is physical: transformers, substations, and high-voltage transmission lines have lead times ranging from 3 to 7 years. Consequently, existing energized sites, even those originally built for lower-tier workloads, have accrued a massive "time-to-market" premium.

The fundamental constraint is no longer silicon availability, but power availability. While NVIDIA has ramped up production of H100 and Blackwell GPUs, the timeline to energize a new data center has lengthened significantly. In major markets like Northern Virginia, vacancy rates have plummeted to near-zero levels (1.6% in H1 2025), and power utilities are imposing multi-year wait times for new transmission connections.

A Shift to Larger Sites

To meet the skyrocketing demand for AI compute, we'll need to build a new class of data centers that is fundamentally different from what exists today. The massive computational loads and high power density required to train modern AI models are forcing companies to abandon traditional, smaller server rooms in favor of colossal "hyperscale" facilities—sites with 500 MW to 1 GW of capacity—that can economically support the specialized cooling and gigawatt-scale energy infrastructure needed. A single training run for a GPT-5 class model may require tens of thousands of GPUs operating in a low-latency cluster, consuming hundreds of megawatts continuously.

There are very few sites in North America with this level of energized capacity available today. Most sites of this size are in the "planned" or "permitted" stage, meaning they are years away from drawing power. With 1 GW of total capacity at its Corsicana facility, Riot possesses one of the most valuable undeveloped power assets in North America.

According to Altman Solon, the volume of generative AI tasks is expected to reach 80-130 million per hour within the next year across key enterprise functions in the US alone. This explosion in inference workloads (using the models) will require a distributed network of data centers. While training (creating the models) happens in massive centralized clusters (100 MW or more), inference can happen in slightly smaller, more distributed facilities. Riot’s portfolio, including Rockdale and potentially other sites, could play a role in this tiered architecture, but the immediate opportunity is in Corsicana.

Data Center Fuel Sources

Rather than attempting to summarize this myself, I've pasted the original data from the International Energy Agency (IEA) below.

The United States and China are by far the largest data centre markets today. In both countries, most of the electricity consumed by data centres is produced from fossil fuels, which also meet most of the increase to 2030. However, rising deployment of renewables and later nuclear power is expected to slow the growth of fossil fuel power generation after 2030.

With a share of over 40%, natural gas is currently the biggest source of electricity for data centres in the United States, followed by renewables – mostly solar PV and wind – with 24%, as well as nuclear and coal power with shares of around 20% and 15%, respectively. As demand growth is particularly rapid over the next five years, natural gas is the largest source of additional supply, adding over 130 TWh of annual generation until 2030. Renewables are the second-largest source of additional electricity supply, adding 110 TWh to the data centre electricity supply between 2024 and 2030. This is mainly due to the continuing increase in the share of wind and solar PV in the electricity mix of most states, as well as some data centre operators investing in co-located renewables.

Nuclear power plays a significant role in meeting data centre electricity demand in the United States, particularly after 2030 when the first SMRs are expected to be commissioned. Technology companies have plans to finance more than 20 GW of SMRs to date, though successful development of the technology could open up even larger opportunities. Together with the ongoing increase in renewable electricity generation, the expansions of SMRs reduces the need for additional natural gas-fired generation so that by 2035 low-emissions sources account for over half of the United States' data centre electricity supply mix.