Michigan regulators are reviewing a request that would make Consumers Energy's typical residential customer pay $13 more a month starting in 2027 — the utility's sixth consecutive annual push for an electric rate increase, according to Consumer Reports via GCN. That's not a one-off local story. It's the opening scene in a fight playing out from Lansing to Tallahassee to Raleigh to Sacramento, and the question underneath all of it is the one that matters most for anyone who wants the grid to get bigger, faster: who pays for the power that AI needs, and does that bill land on the companies building data centers or on the retiree down the street?
I think the backlash is real, and I think it's mostly aimed at the wrong target. The grid needs more capacity, more transmission, more generation of every kind — that's the whole thesis of this publication. But "build more" and "bill households for someone else's server farm" are not the same instruction, and 2026 is the year state regulators finally started treating them as separate questions.
Michigan's Rate Case Shows the Old Playbook, Without an AI Villain Yet
Consumers Energy's filing is a useful baseline precisely because it isn't about data centers. The utility is asking the Michigan Public Service Commission for $456 million annually under what it's calling its 2027 Reliability Action Plan, framed as "a blueprint for upgrades that will lead to fewer and shorter power outages," per WZZM, cited by GCN. That reliability promise is the utility's characterization of its own spending plan — not a verified outcome, since the upgrades haven't happened yet. Worth remembering every time a utility leads with an outage-reduction pitch: the capital gets approved first, the performance data comes years later.
What's notable is the trajectory. The prior case saw Consumers originally ask for $436 million, only to have regulators approve $276.6 million after testimony from the Michigan Attorney General's office and other parties, with that smaller increase taking effect in May 2026, according to WKAR, via GCN. Consumers is also seeking to lift its authorized return on equity to 10.25%, which the Michigan Attorney General's office says would be the highest in the state, and DTE is separately asking for the identical 10.25% figure, up from 9.9%, according to Consumer Reports via GCN. Two utilities in one state asking for the same record-setting return in the same cycle is not a coincidence — it's a coordinated test of what regulators will tolerate before the AI load shows up in the filings explicitly.
Colorado and North Carolina Show What Happens Once AI Load Shows Up by Name
Colorado is where the two threads — ordinary rate increases and data center demand — are openly tangled together, and regulators are refusing to pretend otherwise. Xcel Energy customers saw their bills rise twice this fall, with electric rates increasing in late August and gas rates on October 6, adding roughly $10 a month for a dual-fuel household, per ReEnergizeCO. Here's the part I want to be precise about, because it's exactly the kind of claim this publication has to get right: those increases were not caused by data centers, and neither Xcel nor state regulators attributed them to AI load, according to the same reporting. Data centers are a real and growing draw on Colorado's grid — Xcel itself projects that very large customers could require more than 1,900 megawatts of additional power by 2031, a 31% jump over current supply — but that's a utility's forward-looking planning estimate, not a bill driver today, and the Colorado Public Utilities Commission hasn't yet decided who pays for it, per ReEnergizeCO.
The announced-versus-operational gap in Colorado is stark enough to state plainly: market-intelligence firm Cleanview counts 13 data centers actually operating in the state today, drawing about 173 megawatts combined, while at least 10 more planned or under-construction projects could push the total above 1,400 megawatts, as reported by Axios, via ReEnergizeCO. That's an eightfold jump between what exists and what's on paper. Anyone writing — or reading — projections about data center electricity demand needs to hold that distinction constantly in view. A pipeline number and a meter reading are not the same fact.
North Carolina has moved from technical filings to open political confrontation. Governor Josh Stein is pushing the state Utilities Commission to require data center companies to pay more of their own infrastructure costs through "large load" tariffs, arguing a pending deal between tech companies and Duke Energy is too industry-friendly, according to WRAL. Stein pointed to deals in other states as the model he wants — Google committing to nearly 2,000 megawatts of wind, solar, and battery storage in Minnesota, Amazon providing 960 megawatts of clean energy in Indiana, and Meta paying for 1,200 megawatts of nuclear capacity in Ohio, per WRAL. Those are commitments the companies made elsewhere, not guarantees North Carolina has secured — which is precisely Stein's complaint. He has limited formal leverage here: after his 2024 election, the Republican-led legislature stripped the governor's control over Utilities Commission appointments, handing a majority to Republican appointees, per WRAL. That's a governance detail that matters more than most climate-adjacent political statements do, because it tells you the fight is now structural, not just rhetorical.
The Large-Load Tariff Is Becoming the Grid's New Default Settlement Mechanism
This is where I get genuinely excited, because the policy machinery catching up to AI demand is more interesting than any single rate case. States and grid operators are converging on a specific tool — the large-load tariff — to answer the "who pays" question before it becomes a crisis rather than after.
Florida built this into statute first. Under SB 484, signed in May 2026 per Utility Dive, via DataCenterFeed, every investor-owned utility had to file a compliant large-load tariff — defined as covering customers with an anticipated peak load of 50 megawatts or more — by October 1, 2026. Duke Energy Florida was first out of the gate, but Florida Public Service Commission staff recommended on October 22 that the commission deny Duke's proposal and send it back for revision, on the grounds that it sets contract terms for these mega-customers without actually setting a rate before 2028, according to MGrid, via DataCenterFeed. Commissioners vote on it November 3 in Tallahassee. If staff's recommendation holds, it's a real signal that regulators are willing to reject the first drafts utilities put in front of them rather than rubber-stamp something that defers the actual pricing fight.
The same architecture is now moving at the grid-operator level, which is arguably more consequential because it sets rules across entire regions rather than one utility at a time. CAISO's Board of Governors approved its own large-load interconnection framework on October 28, also built around a 50-megawatt threshold, and will file the tariff language with FERC by November 16 to answer a federal show-cause order that told six regional grid operators to prove their interconnection rules don't shift AI-driven costs onto everyone else, per MGrid. CAISO's centerpiece is a "Flexible Interim Load Interconnection" service that lets large loads connect quickly in exchange for accepting curtailment during grid stress — trading speed for flexibility rather than forcing new transmission to be built before any megawatt flows. Notably, PG&E pushed for a lower 20-megawatt threshold and CAISO declined, for now, while flagging that MISO has already proposed exactly that split — a 50 MW generic class and a separate 20 MW class for computational loads, per MGrid. Underneath all of this sits a DOE proposal asking FERC to assert direct federal jurisdiction over large-load interconnection above 20 megawatts, a docket that's been sitting at the proposed-rule stage since last fall, according to InfraCompute's regulatory tracker. Different thresholds, different clocks, same underlying problem: nobody wants to be the state or grid operator that let AI load degrade service for everyone else, and nobody wants to be the one that scared off $70 billion in data center investment either.
Backlash Can Kill Projects Just as Effectively as a Denied Tariff
It's worth sitting with what happens when the cost-allocation fight goes unresolved for too long, because Australia is showing the downside case in real time. More than three-quarters of the 40-plus large data centers planned around Sydney and Melbourne — representing more than $70 billion in projected investment — are now at risk from strained grids and local opposition, with only 23% of those projects having publicly confirmed they've actually secured power supply or grid connections, according to a Bloomberg review of company disclosures. That 23% figure is the cleanest announced-versus-operational gap in this entire story: four out of five of these "planned" projects haven't cleared the most basic test of whether they'll ever get built.
And the stakes aren't only about who pays — they're about what gets built to supply the power in the first place. Developers facing slow interconnection queues and uncertain cost allocation are increasingly turning to on-site natural gas to meet AI's electricity demand, a shift Bloomberg's climate team argues could erase over a decade of progress in cutting power-sector emissions, in an analysis highlighted in their newsletter. That's the honest tradeoff electricity maximalists have to sit with: when the permitting and pricing fights drag on, the fallback isn't "no power gets built." It's "the fastest, dirtiest power gets built," because gas turbines can be sited and financed without waiting for a state commission to decide who's holding the bag.
What to Watch
The next month settles more of this than the last six have. Florida's commission votes November 3 on whether to send Duke's large-load tariff back to the drawing board. CAISO's tariff language hits FERC by November 16, which will show whether a 50-megawatt threshold and curtailable "flexible interim" service becomes the template other regional operators copy or a cautionary tale PG&E and others point to when pushing for stricter rules. And Michigan's Public Service Commission will eventually rule on Consumers Energy's $456 million case — the sixth round of a fight that, so far, has nothing to do with data centers at all. That last point might be the most important one. The rate case backlash didn't start with AI, and it won't end there either. AI is just the loudest, largest new customer walking into a cost-allocation fight that utilities and regulators have been having for years. Getting the tariff structure right now is what lets the next ten gigawatts get built without turning every new data center into a referendum on someone's power bill.
