The 2025 PJM annual meeting occurred on May 12 through May 14 this year. In the members committee meeting panels of experts, PJM officials, and PJM members discussed AI integration at PJM and the 5-year plan for PJM, and members voted for or against the approval of the nominated board members. The following day in the general session, panels discussed large load and challenges in building new generation.
PJM Board Election Results
The headline of the event was the vote for the approval of board members. Each year the nominating committee puts forward three nominees to serve on the board for the next three years, if approved by the membership. There are 9 total board members. The three nominees this year were the current board member Terry Blackwell, new board member Matt Nelson, and the current chair of the board, Mark Takahashi. Terry Blackwell and Mark Takahashi both failed to get approval by the PJM members; however, Matt Nelson was accepted with over 90% of the sector-weighted vote. Some members were concerned that the board would be less stable without reappointing the current members, that it would be a bad idea to drastically change the board when they will be appointing a new CEO this year, and that it would be difficult to find other qualified and independent candidates for the role. However, their efforts were not enough to get approval for a revote. Some members have voiced concerns with the state of PJM recently. This included the long queue process for generation, the state of the capacity market, unfair treatment from the board to certain PJM members or groups of members, and the board making decisions that went against the will of the membership. At this point the nominating committee is looking for new candidates for the board, to present to the members at the meeting next month.
See a Related Article
Impacts of Data Centers and Other Large Loads
In the panel on large load, the energy needs of data centers for AI was a major concern. Currently, PJM expects to see 20% more electricity demand in the next 10 to 15 years. However, the efficiency of data storage, the number of users, how much data storage each user needs, and the expansion of AI are all rapidly changing which makes it difficult to predict what the true energy consumption levels will be in a few years. Another concern is that there can be multiple companies who put in interconnection requests for a data center, but they are both seeking the same client. This leads to problems where only one of the data centers will get built, but PJM and the utilities were predicting that two or three data centers were being built.
Experts also discussed what would happen if large loads like data centers come offline suddenly. If a data center disconnects from the grid without warning, it will have a major destabilizing impact on the grid. Data centers are able to reduce their consumption much more quickly than most generators can change their production, which will make it very difficult to balance electricity supply and demand in real time, since a large portion of the demand can change more quickly than the supply can. Some standards or regulations may be required to make sure that this does not pose a threat to the reliability of the electric system. Currently, NERC is studying the impacts of large load on the grid with their large load task force.
Building Generation
Building Generation must be considered at the same time as building load since it is the load growth that drives the need for additional generation. Data centers can be built in about 2 years, while it takes 5 years for most large generators to be built, and 10 years for nuclear plants. The biggest challenges for building energy infrastructure at the moment are the uncertainties in policies (see above), interconnection timelines, supply chain issues for key components, and a global lack of manufacturing compared to demand for equipment.
The uncertainty in government programs is an issue at every level of government. The federal government’s programs for carbon capture and investment tax credits could change dramatically in the next few months. Renewable energy goals and incentive programs at the state level are facing pressure from rising electricity prices. Siting and permitting has been an ongoing challenge in many local jurisdictions. All of these issues are made worse by the interconnection timelines. Since the average project in PJM takes 5 years to get through the queue, it is very challenging to finance projects with such uncertainty about how policies will look when the plants go into service.
There are supply chain issues with a lot of key components for large generators such as transformers and turbines that can set projects back several years. The recent tariffs have made things worse, especially for large power plants which need specialty equipment that can only be manufactured in a few factories in the world.
Lastly, PJM is not unique in having load increases. There is a high demand for electricity all around the world which has made the supply chain issues much more of a challenge to overcome. Some companies such as GE are looking to outsource some of their manufacturing to other factories to try to catch up with demand.
PJM 5-year plan
In 2020 PJM set goals for the next 5 years which included facilitating decarbonization, getting the grid ready for the future of renewable energy, and encouraging innovation through stakeholders. The strategy is determined by balancing reliability of the system, stakeholder engagement, risk management, the PJM workforce, and using the efficiencies of scale. At the end of this year, PJM will release their plan for the next 5 years. They are taking stakeholder input now through June on what they should prioritize. Some topics mentioned are balancing plant retirements and load growth with state policies and bringing on new generation in a way that keeps reliability high and prices low.
AI integration at PJM
PJM has partnered with Google and Tapestry to improve the interconnection process. They have not decided on a long-term strategy yet, but their current plan is to use it to improve data accessibility and front-end data processing. They are also hoping to use machine learning to improve their modelling processes such as day-ahead market forecasting or expected renewable energy plants outputs based on weather conditions. Currently, many models are very time consuming, so a faster and more accurate model would enhance gird operations and planning.
