Price grid access honestly. Data centers should not pay ordinary average rates if they are causing extraordinary new costs for wires, transformers, substations, and generation. Otherwise, the result is a quiet cost shift from hyperscalers to legacy customers.

Treat interconnection like scarce capacity. The current queue can reward whoever files first, whether the project is real or just a paper claim on future power. Kavulla’s better model is an “open season” for grid access, borrowing from gas pipelines, where providers define available capacity and data centers bid for firm withdrawal rights.

Let serious projects pay for certainty. If a data center truly values power at a particular site, it should be willing to back that demand with real money. That would help separate actual projects from queue confetti and give utilities a clearer signal for what infrastructure is worth building.

Return surplus value to ratepayers. If data centers bid for scarce grid capacity and those bids exceed the cost of needed upgrades, the upside should not simply disappear into the utility fog bank. Done right, surplus revenues could flow back to existing customers.

Make “bring your own generation” mean something. BYOG cannot just be a press release about clean energy while the legacy grid quietly absorbs the reliability risk. A serious version would require data centers to bring new generation, free up unused capacity, or operate flexibly as a condition of service.

Reward flexibility, not just size. Data centers that can curtail, ramp, shift computing, or operate with backup power during stress should be more valuable to the grid than loads that demand perfect service at all hours and then send everyone else the bill.

E3: Understanding the Drivers of Rising Electricity Rates and the Role of Data Centers
A useful corrective to the cartoon version of the rate debate. E3 finds that many forces are pushing bills up, including inflation, gas volatility, resilience spending, grid modernization, market design, retirements, and load growth. In PJM, E3 estimates load growth accounted for about half the capacity price increase, with the rest driven by market design changes, retirements, accreditation changes, and other supply-side factors.

VICE: The AI Boom Is Taking Over the U.S. Power Grid, and This Map Shows Where It’s Happening
A more consumer-facing look at the national data center map, with 4,000-plus facilities spread across the country and heavy concentrations in Virginia, California, and Texas. Not the wonkiest piece in the stack, but useful for seeing how this issue is starting to reach normal people, which is where politics usually begins sharpening its little teeth.

Financial Times: Is Gas Making a Comeback on the U.S. Power Grid?
The FT looks at the surge in gas capacity in interconnection queues, with gas back in the conversation because data centers and industrial load need firm power faster than the existing clean-energy buildout can always deliver it. The uncomfortable reality: the AI boom is reviving arguments that many policymakers thought they had already retired, laminated, and placed gently in a drawer.

Scientific American: NASA’s Plan for a Nuclear Reactor on the Moon
A fun one for the nuclear file. NASA is looking at lunar nuclear power as part of the infrastructure needed for longer-term moon operations, with China and Russia also eyeing lunar reactors. The lesson travels back to Earth nicely: wherever humans want durable activity beyond the easy reach of the grid, firm power becomes the main character.

The basic idea: Instead of plugging a data center into a stressed land grid, the platform generates power from ocean-wave motion and runs compute directly onboard.

The shape: Panthalassa has described large autonomous offshore structures, roughly “lollipop”-like, that convert wave motion into electricity through internal systems.

The cooling advantage: The ocean is a giant heat sink, which helps with one of the least romantic but most important parts of AI infrastructure: keeping machines from cooking themselves.

The connectivity problem: Offshore compute still has to move data. Reports say Panthalassa’s approach relies on satellite connectivity, including Starlink, which makes the project part energy concept, part telecom concept, part naval weirdness.

The real question: Floating data centers avoid some land, water, and grid fights, but they create new ones around reliability, maintenance, storms, permitting, marine impacts, latency, and whether anyone actually wants their AI infrastructure bobbing around like a sovereign Roomba.

PJM got permission to pull data centers off the grid before blackouts. DOE granted PJM emergency authority to curtail data centers and other large loads with backup generation during this week’s hot-weather stress, but only as a last resort before rolling blackouts. That is a giant little sentence: data centers are no longer just “new load” in PJM. They are now a controllable emergency resource, which is what happens when the grid runs out of polite fiction.

PJM moved the backstop auction up because March is too late now. PJM now plans to hold its backstop reliability auction in September instead of March to address rising data-center demand. The unresolved problem is cost allocation: PJM wants states to protect existing customers from data-center-driven auction costs, but nobody has fully solved how to keep hyperscaler capacity needs from washing through ordinary bills.

PJM’s winter bill came due at almost $1 billion. Reuters found PJM paid a record $990 million in uplift payments to money-losing plants in Q1, more than the $764 million paid in all of 2025 and far above the $270 million paid in 2024. This is the grid’s most expensive kind of confession: when the system gets tight, old plants with ugly economics suddenly become priceless.

NERC says the summer looks manageable, but the shoulders are getting weird. NERC says the U.S. power grid should have enough resources for typical summer demand, helped by new additions, but risk is growing in shoulder seasons as demand patterns, maintenance schedules, and data center interconnection delays make forecasting harder. The old calendar assumed spring and fall were the quiet rooms of the grid. AI load is dragging a drum kit in there.

South Carolina’s data center politics are turning Republican-on-Republican. Rep. Nancy Mace is pushing for a statewide data center moratorium, joining a growing bipartisan pause-the-buildout movement driven by utility-cost concerns. That matters because data center skepticism is no longer just progressive environmentalism or rural NIMBYism. It is becoming right-flank ratepayer politics with a hard hat and a microphone

Texas may be planning water as if AI never showed up. E&E reports that Texas officials are scrambling to prepare for AI-serving projects, but the state’s draft water outlook skips the data center surge. That is the Texas growth model’s blind spot in miniature: the power story gets all the attention, but the water ledger is sitting quietly in the corner with a knife.

Storage just had a monster quarter. U.S. developers installed 9.7 GWh of energy storage in Q1, a first-quarter record and up 32% year over year, with SEIA pointing to data centers, volatile power prices, and gas-market disruption as demand drivers. Storage is not firm power, but it is becoming the grid’s shock absorber, and the AI buildout is buying every cushion it can find.

California may fumble one of its best flexibility tools. Newsom’s revised budget funds California’s Demand Side Grid Support program through this year but zeroes it out in 2027, moving participants toward a utility-run framework that clean-energy groups oppose. The program dispatched more than 539 MW during a test last July, which is exactly the sort of flexible capacity everyone says they want until the budget scissors come out.