Few issues in American infrastructure are as debated right now as the impact of data centers on electricity prices.
Politicians warn that ratepayers could end up subsidizing Big Tech. Developers point to jobs and tax revenue. Grid operators have raised concerns about reliability.
Neel Somani, a former quantitative researcher who covered power and gas markets at a major hedge fund – and now works in machine learning research – argues that many of the loudest claims on both sides miss the complexity of the issue.
His primer, Power 2026, published at power2026.ai, is an effort to give investors, executives, and policymakers the tools to think through the question themselves.
Why location and timing determine a data center’s power impact
Somani’s main argument is that a data center’s effect on electricity prices varies across regions. The outcome depends on where the facility is located, when it uses power, and how the local market operates.
A data center that raises costs in one region could reduce them in another.
Retail electricity bills are not simply the wholesale cost of generating power. They also include transmission, distribution, and utility costs, many of which are shared across all customers on the grid.
When a large data center connects to the system, those fixed costs are spread over a larger consumption base.
In some cases, Somani notes, ordinary ratepayers may actually see lower costs because they are sharing infrastructure expenses with a new large customer.
Timing can matter just as much as location. In regions with significant renewable generation, the gap between abundant midday solar production and higher evening demand creates what the California grid operator calls the duck curve.
Somani points out that a data center consuming power during low-demand daytime hours can help absorb excess generation, making it more economical for efficient combined-cycle gas plants to operate throughout the day.
The result can be lower evening prices for other customers. In parts of Texas, wind generation can occasionally push electricity prices below zero, meaning customers are effectively paid to consume power. Large loads such as data centers can absorb some of that excess supply and improve overall grid efficiency.
When data center growth really does raise costs
The benefits are not automatic.
A data center built in the wrong location can strain local transmission infrastructure, and congestion is often what drives regional price increases.
When a transmission line reaches its limit, cheaper electricity from distant generators may no longer reach customers. The market then relies on more expensive local generation, pushing prices higher.
The North American Electric Reliability Corporation has warned that data center growth could increase costs in capacity markets, where several US regions pay generators to remain available when needed.
This is why Somani avoids simple answers. The question is not whether data centers are good or bad for ratepayers. The answer depends on the specific market conditions surrounding each project.
Why policymakers are searching for new rules
The policy response to rising data center demand has ranged from cautious to restrictive.
The most neutral federal effort Somani discusses is the White House’s Ratepayer Protection Pledge, a non-binding statement focused on protecting consumers from higher costs.
At the more aggressive end, the Power for the People Act proposes new rate classes, interconnection rules, and federal regulatory changes, though it has not gained significant traction.
Some local governments have taken stronger action. Monterey Park, California became the first jurisdiction to ban data centers through a ballot measure, while a statewide effort in Maine was vetoed. New Hampshire has considered allowing data centers to operate entirely off-grid, physically separated from the shared electricity system.
Somani argues that these approaches can create tradeoffs. An off-grid gas-powered data center may avoid adding demand to the grid, but it also no longer shares grid costs. It still consumes natural gas, which can put upward pressure on fuel prices.
Every policy decision creates ripple effects. Addressing one visible problem without considering the downstream effects can produce unintended consequences.
AI power demand is growing faster than infrastructure can be built
For business leaders, the harder question is how AI power demand actually takes shape.
Somani reports that some major AI labs are comfortable with their expected power and compute needs several years out.
The immediate challenge is meeting demand today. That creates a financing problem: a surge of demand over the next six to twelve months is not enough to justify building a new power plant, which typically requires years of contracted revenue to support nine-figure loans.
Recent deals illustrate the range of arrangements emerging in the market.
Somani points to a SpaceX arrangement to sell power to Reflection with a 90-day exit option for either party. His estimate puts the value at roughly $5,000 per megawatt-hour for power bundled with ready-to-use GPUs.
That level of flexibility comes at a price. The short commitment also makes the revenue difficult to finance.
At the other end of the market is Anthropic’s $19 billion lease with TeraWulf, covering 400 megawatts over 20 years beginning in the second half of 2027. The deal implies roughly $271 per megawatt-hour, excluding GPUs but including supporting infrastructure such as buildings and cooling capacity.
The difference between these two arrangements shows how much the market values certainty versus flexibility.
What executives need to evaluate before building
The implications for decision-makers are fairly straightforward.
For executives choosing data center locations, the decision should involve more than land costs and tax incentives. Transmission constraints can determine whether a project becomes a valuable new customer or a burden on the local grid.
For investors, the opportunity lies with companies that can develop, finance, and operate power assets at scale. But enthusiasm alone is not enough. Power markets are volatile, and contract structures determine who absorbs that risk.
For policymakers, the challenge is avoiding one-size-fits-all rules. Data centers are neither a guaranteed burden on ratepayers nor a guaranteed benefit. The impact depends heavily on local market conditions.
The most useful part of Power 2026 is not a prediction about where electricity prices are headed. It is an explanation of how those prices are formed: through marginal costs, transmission limits, and market rules that differ from region to region.
Once those mechanics are understood, the constant stream of data center headlines becomes easier to evaluate. The better question is what conditions create those outcomes, and who has the power to change them.
