The digital merit order
Historically, hyperscalers positioned themselves as clean energy leaders by signing large renewable power purchase agreements (PPAs) and pursuing aggressive decarbonization targets. They are promoting AI as a vital tool to help develop the next generation of sustainable technologies and solutions. However, the race to build AI infrastructure and ensure reliable, high-quality power has shifted short-term priorities from sustainability to speed and reliability.
This creates a profound paradox: The infrastructure meant to develop our low-carbon future is being built with carbon-intensive solutions. This is part of our series on data centers and powering AI. Future blogs will highlight power quality challenges ranging from machine learning load dynamics to water consumption and much more.
The traditional merit order
In most electricity markets, the merit order ranks generation resources by short-run marginal cost, dispatching the cheapest first—typically renewables and nuclear—followed by coal, gas, and oil. This approach minimizes system costs and often supports decarbonization because low-carbon resources tend to have low operating costs. However, the actual dispatch order set by the grid system operator is not purely merit-based; it reflects the merit order modified by real-world power system constraints such as transmission congestion, reliability requirements, and electrical operational limits. To keep the grid working, these practical constraints override cost rankings, especially during peak demand, causing higher-cost and higher-emission plants to be dispatched ahead of cheaper resources.
The emerging digital dispatch order
Hyperscalers would be thrilled to get a GW interconnection in under two years with a grid that offers them five-nines (99.999%) reliability. Apart from a handful of sites repurposed from shuttered coal plants, most grids lack the transmission and generation capacity to support this scale of demand, and none provide that extreme level of reliability.
Building new transmission lines alone can take 5–10 years, so operators are increasingly turning to behind-the-meter generation as an interim solution. Hydro and nuclear power projects have decade-long timelines and challenging citing requirements.
The five-nines reliability standard for hyperscale data centers is achieved through redundant systems, UPS (Uninterruptable Power Supply), and backup generation. Co-located renewables would require massive battery storage to meet these standards, making them cost-prohibitive in the near term. This challenge is compounded by current limitations in grid-forming capabilities of inverter-based resources, which makes achieving hyperscale reliability even harder.
The result: Developers are deploying modular gas turbines and reciprocating engines behind the meter, often in clusters that can scale from tens to hundreds of megawatts within 12–24 months. To meet the demand of the next 2-3 years, we’re seeing power generation technologies being dispatched in the following order:
Digital Dispatch Order (Next 2–3 Years)
Power generation technologies being dispatched in the following order.
-
Simple-cycle gas turbines and reciprocating engines
These machines dominate the near-term strategy because they can be deployed quickly and scaled modularly, offering proven reliability for hyperscale uptime requirements. Their downside is a heavy reliance on natural gas and exposure to fuel price volatility, which raises emissions concerns. While carbon capture or hydrogen blending could help mitigate these emissions, in most cases connecting the site to the grid and wheeling renewable power would be a more cost-effective and sustainable solution. The demand for these machines has put major pressure on the supply chain, driving up costs and delivery times; leading developers to consider options further down the dispatch order.
-
Gas-fired boilers
Boilers are often chosen for their speed because they can be installed rapidly and scaled to meet immediate thermal and backup needs; however, they are inefficient and carbon-intensive, making them a short-term fix rather than a sustainable solution. In the future, these systems could be retrofitted or replaced by small modular reactors (SMRs) to provide zero-carbon steam generation. Rising carbon costs and stricter emissions standards could accelerate their obsolescence.
-
Combined-cycle gas turbines
Combined-cycle plants offer higher efficiency and are well-suited for large, steady loads, making them attractive for hyperscale campuses. However, a clogged supply chain for large scale machines that can deliver 100s of megawatts has pushed this solution down the digital dispatch order with wait times in excess of 5 years. While more efficient than simple cycle or gas engines, these units are less flexibility for the rapid ramping AI data centers require.
-
Fuel cells (natural gas)
Fuel cells provide cleaner power than combustion systems and deliver high-quality, steady electricity. Their Achilles’ heel is cost and complexity, and certain technologies like solid oxide fuel cells struggle with load-following, which limits their role in hyperscale environments. If low-carbon hydrogen (green or turquoise) becomes viable, fuel cells could evolve into a cornerstone of sustainable strategies. For now, technology risk and supply chain maturity remain barriers.
-
Renewable microgrids with battery storage
Microgrids paired with storage offer resilience and sustainability benefits, and can be deployed relatively quickly for supplemental power. They cannot provide hyperscale baseload alone, and intermittency remains a challenge. Hybrid configurations that combine renewables with gas engines can offset fuel costs and improve sustainability, especially in regions with high gas prices. Land use constraints and variability in renewable resources are ongoing concerns.
-
Coal plant extensions
Extending coal plants offers can add immediate capacity by leveraging existing infrastructure, but it entails significant emissions and reputational risk. While these extensions can help support generation in constrained grids, regulatory restrictions and investor pressure make them a last-resort option for new developments.
-
Nuclear refurbishments
Refurbishing existing nuclear plants offers zero-carbon baseload and proven reliability, but timelines and costs are significant, limiting near-term impact. These projects are strategic for long-term decarbonization and energy security, though public opposition and regulatory hurdles remain persistent threats.
-
Small modular reactors (SMRs)
SMRs promise scalability, safety, and carbon-free baseload power, and they can be sited closer to demand centers. The success of the demonstration projects underway will dictate the pace of adoption. If licensing and supply chain barriers fall, SMRs could redefine hyperscale power strategies in the next decade.
-
Large scale nuclear
Large nuclear plants deliver unmatched reliability and zero-carbon output, but their multi-decade timelines and massive capital requirements make them a long-term solution only. Cost overruns and political risk continue to challenge their viability of new large scale nuclear development.
-
Fusion
Fusion represents the ultimate game-changer for zero-carbon baseload power, but it remains experimental, with no commercial deployments expected this decade. For fusion to become commercial, several demonstration projects will need to be built, and lessons learned incorporated into commercial facilities a process likely to take ten years. Given that there are now a meaningful number of well-funded private groups pushing ahead to build pilot plants, fusion is no longer “10 years away from being 10 years away.” Start the clock when ground is broken on more than one protype plant.
The path ahead: Grid integration
While this appears to be a short-term retreat from sustainability, these assets could eventually be connected to the grid, improving flexibility and resilience. If connected to the grid, AI facilities could access renewables when available, monetize grid ancillary services, and improve system-wide resilience. Ultimately, this would lead to a more reliable and cost-effective power system for all consumers.
A more nuanced conversation on reliability is also warranted. While highly reliable, North American grids can’t meet five-nines reliability when measured by SAIDI/SAIFI metrics. However, four-nines reliability with two hours of battery storage and redundant UPS architectures can achieve reliability levels suitable for most datacenter operations, with generators reserved for exceptional circumstances.
The emerging digital dispatch order prioritizes technologies that can be deployed at scale quickly and deliver the high reliability hyperscale operations demand. In the near term this doesn’t help strengthen the grid and the pressure on supply chains makes most major power infrastructure more expensive. However, if these projects maintain a commitment to the long-term benefits of grid connectivity, their assets can become grid-tied resources. This creates an opportunity for the fossil-fueled generators currently providing AI with baseload power to shift into peaking roles, facilitating a greater deployment of intermittent renewables and aligning the system more closely with the merit order envisioned for the future.
Over the next decade, through strategic grid integration, policy innovation, and maturation of SMRs and long-duration storage, we can realign the Digital Dispatch Order with the Digital Merit Order. The energy challenge is just one part of the sustainability equation. In our next article, we’ll dive into AI’s hidden water costs and the solutions ready to address them.
About the author
Kerry McKenna
Senior Project Manager , Power
Driving complex projects at the intersection of energy and innovation, Kerry McKenna focuses on Power-to-X solutions, where “X” can be hydrogen, alternative fuels, chemicals, or even data. His role involves enabling large-scale infrastructure by integrating diverse power generation technologies—renewables, thermal, hydro, nuclear, and hybrid systems—to deliver sustainable, reliable, and future-ready solutions. He is passionately committed to bringing positive change to the ever-growing energy sector.