The Midterm Watchlist

Six months ago, in Issue #14, this newsletter led with a prediction that wasn't ours. NPR had just run a national story calling data centers "a kitchen table issue" and a potential "tipping point" for the 2026 midterms. We wrote at the time: "The 2026 midterms will be the first national test."

The test is now eleven weeks out. And the evidence that data centers have become genuine ballot-box politics is no longer a media framing, it's a money trail, a set of actual ballot questions, two governor's races functioning as referendums, a U.S. Senate race where the issue is explicit, and a documented pattern of incumbents losing their seats over a single project vote.

Nobody has compiled the full election map for this industry. So, in the tradition of the State Risk Ratings: we built it.

This is the inaugural DC Pipeline Midterm Watchlist, every race, referendum, and ballot measure in America where data centers are on the line this November, what each result means, and which State Risk Ratings move the morning after.

One promise before we start, in keeping with how this newsletter operates: this is not a partisan document. We track what outcomes mean for development risk, ratings, and the people who build, not who deserves to win. Both parties contain pro-buildout and anti-buildout factions now; that scrambling is half the story.

Let's start with the part nobody else is willing to lead with: the money.

The scale of documented political spending around data centers and AI this cycle, from FEC filings and campaign finance tracking:

More than $100 million in AI and data center industry money has been tracked into the 2026 cycle by campaign finance watchers, part of a broader crypto-and-AI super PAC surge exceeding $321 million. A data center watchdog group puts AI-related PAC and super PAC spending at $65 million or more across races nationwide.

Meta alone committed roughly $65 million to super PACs backing pro-AI candidates against Big Tech critics, a documented $45 million contribution flowing through an umbrella vehicle that funds affiliated PACs in Texas ("Forge the Future"), Illinois ("Making Our Tomorrow"), and California ("META CA," "California Leads").

The lobbying layer is scaling just as fast. The Data Center Coalition's federal lobbying spend tripled from $123,000 to $360,000 per quarter through 2025, with revenue growing from $583K (2022) to $2.5 million (2023). Amazon spent $4 million in a single recent quarter lobbying on "data centers," "grid modernization," and "energy production."

Now the two details that tell you more than the totals:

First: the money is losing some races. In Illinois' March 2026 primaries, Meta-backed PACs lost three of four state legislature races despite massive spending. Money is necessary in this fight; it is demonstrably not sufficient. Every opposition group in America noticed.

Second: the industry's own ads hide the issue. In North Carolina's most closely watched AI-money congressional race, where industry PACs are backing incumbent Rep. Valerie Foushee against a primary challenge, the pro-industry ad doesn't mention data centers at all. It frames her as a progressive fighter on immigration and Trump accountability. The AI industry is spending millions in a district where the data center debate is raging in local politics, while keeping the words "data center" out of its own advertising. When a nine-figure spender won't say its own issue's name in a swing district, it has read the same polls we have (more on those below).

And one race captures the entire dynamic in miniature, so it leads the Watchlist.

TEXAS - U.S. Senate: Paxton vs. Talarico

The newest addition to the map, and maybe the most revealing race in the country. This month, reporting confirmed that Attorney General Ken Paxton's Senate campaign has received thousands in donations from the data center industry, while Hood County awaits Paxton's own AG ruling on the legality of its data center moratorium. The conflict is structural: the official deciding whether Texas counties can pause data centers is fundraising from the industry that wants them unpaused.

His Democratic opponent, James Talarico, has gone the other way entirely, calling for data center regulation at an Arlington rally ("If data center companies can't meet our demands, they shouldn't be built in our communities") and proposing an "AI dividend" paying Americans a share of data center profits. Meanwhile a pro-Talarico super PAC received $10 million from LinkedIn cofounder Reid Hoffman, who has publicly supported data centers. Even the money is scrambled.

What it means: A statewide Texas race where one candidate proposes profit-sharing from data centers is already a landmark, that idea entering the mainstream would reshape incentive politics everywhere. And the Hood County ruling (whenever it lands) is a live Texas rating trigger regardless of the election: it decides whether the Hill County precedent (moratorium sued into rescission, Issue #21) becomes statewide doctrine. Texas remains Build (downgrade watch); this race and that ruling are both on the watch list.

PENNSYLVANIA - Governor: Shapiro vs. Garrity

The referendum disguised as a race, fully documented in Issue #23. Shapiro: build, with mandatory community benefit agreements and 32% clean energy conditions (his GRID framework, already House-passed). Garrity: a Republican attacking "sweetheart tax deals" from the populist flank, running listening sessions across the state. $90 billion in announced projects, $20 billion from Amazon, 64% public concern, a legislature that punted everything (one disclosure rule), and SB 1345's local-moratorium-authority model waiting in the wings.

What it means: A Shapiro win validates the conditioned-build model, Cedar Rapids economics enforced at state scale, and likely moves Pennsylvania from Contested toward Build with Conditions as GRID codifies. A Garrity win validates subsidy rollback as a winning message for either party, keeps the exemption fight open, and holds Pennsylvania at Contested with downside risk. Either way: the pre-2025 status quo isn't on the ballot. It no longer exists in Pennsylvania politics.

NEW YORK - Governor: Hochul's Re-election

Everything from Issues #21-22 converges here. Hochul's executive order (50 MW+ pause) is a revocable instrument that lives exactly as long as her administration wants it to. The legislature's harsher 20 MW statutory bill still sits unsigned, she has until year-end. The $19.4 billion Genesee County project hangs between her labor base (IBEW: build it) and her environmental base (sign the bill). Her June polling showed the moratorium is a 46%-good / 21%-bad issue statewide, the rare position backed by Democrats +37 and Republicans +13.

What it means: Hochul winning locks in the EO-plus-leverage structure through the GEIS process; the July 2027 moratorium expiration becomes the next battlefield. A loss puts the entire framework, EO, unsigned bill, community-benefits guidance, in a successor's hands with no guarantees in either direction. New York stays Frozen (50 MW+) through the election; the post-election review is already scheduled.

MAINE - Governor: The Open Seat

Mills vetoed the first statewide moratorium in April and is now running for Senate. Her successor inherits the rematch: the legislature is drafting a modified moratorium for the January session, LD 713 already stripped data centers from state incentive programs, and the veto override math failed by a margin a new governor could erase, or entrench. Small market, national bellwether. Caution holds; the January session is the trigger.

JANESVILLE, WISCONSIN - The First Project on a Ballot

The purest test in America: voters directly approving or rejecting a specific data center project. We've tracked this referendum since Issue #14. Wisconsin's statewide climate is the harshest documented anywhere, 70% of residents told the Marquette Law School Poll that data center costs outweigh benefits, but a specific project with specific local benefits is a different question than an abstract industry.

What it means: Approval gives the industry its most valuable talking point of the cycle: "when actual voters decide on an actual project, they say yes." Rejection is a first-in-the-nation event, a project killed by popular vote, and the template gets photocopied by opposition groups in all fifty states within the month. A rejection likely moves Wisconsin from Contested to Caution.

COLUMBIA COUNTY, GEORGIA - The Counterweight Nobody Covers

The other side of the coin, covered here since Issue #24: a referendum on a funding structure where Google's confirmed 8-million-square-foot, 23-building campus could eventually eliminate residents' homestead property taxes. It is the single most pro-data-center ballot framing in the country, a community voting on whether to accept a deal where the data center pays their taxes for them.

What it means: Passage creates the industry's best-ever community economics case study, the anti-Janesville, and every developer's government affairs team will carry it into every negotiation in 2027. Failure, in a county that confirmed the project enthusiastically, would signal that even maximal benefits can't buy consent anymore. Either result feeds Georgia's Contested rating, the most two-directional state on the board.

The kitchen-table districts. The NPR thesis gets its test in congressional races sitting in data center corridors. The one to watch: North Carolina's Foushee-Allam primary fight, where AI industry money is flooding a district with a live local data center debate, and where a progressive counter-ad now attacks the incumbent for taking the industry's support. When "funded by AI PACs" becomes an attack line that moves votes, the industry's political cover changes permanently. Also on the board: industry PACs placing $500K each behind candidates in a North Carolina swing district and an open Texas seat.

The down-ballot wave watch. The pattern this newsletter documented first: David Batts unseating a four-term North Carolina incumbent after "We will primary you." Festus, Missouri firing four incumbents over one project. Edgecombe County. The Watchlist question for November isn't whether this happens again, it's whether 2026 produces the first coordinated wave of data-center-issue victories in county commissions, town councils, and state legislative seats across multiple states. We'll count them. If the count crosses roughly a dozen documented cases across three or more states, the political risk premium on every contested-market project reprices, because at that point, no local official anywhere can vote on a data center without thinking about their own race.

Control of Congress decides which of the bills we've tracked live or die:

The public opinion record, assembled from six months of coverage, clip this:

Read those five lines together and the industry's strategic position is clear: historically unpopular, structurally unstoppable, and spending nine figures to manage the gap. That gap, between what the public wants and what the capital is doing, is what this election measures.

For subscribers who want to follow the results like this newsletter will, the order the verdicts arrive on November 3:

7:00 PM ET - Georgia polls close. Columbia County's referendum is the night's first data center verdict. An early, clean signal on whether maximal community benefits buy consent.

8:00 PM ET - Pennsylvania and Texas close. The big ones. PA's governor race won't be called early if it's close, but county-level returns in the data center corridors (Northeast PA, Susquehanna Valley, Lackawanna) will show whether the issue moved votes where the projects actually land. Texas Senate returns test whether "AI dividend" populism plays statewide.

9:00 PM ET - New York and Wisconsin close. Hochul's margin, and Janesville, the first American project decided by popular vote, likely called before midnight.

Through the week - the down-ballot count: county commissions and councils in the corridors. This is where the wave, if there is one, shows up, and where we'll be counting.

The post-election issue will score every Watchlist entry and announce any rating actions. Consider it pre-committed.

Floating data centers left the concept phase. Samsung Heavy Industries signed an engineering agreement with Mousterian Corporation, a Texas-based developer, to advance factory-built, moored floating data centers, 50 MW of critical IT capacity per unit, with initial deployments planned in Texas and other U.S. markets. File under: the industry engineering its way around the land-use fight entirely. A data center that arrives by water answers a zoning board with a mooring line. Watch this one, it's either a curiosity or the beginning of an entirely new siting category.

The financing layer is tightening. Nvidia moved to help secure capital "at attractive rates" for its own customers, the chipmaker becoming a financing arranger for the buildout it supplies. Meanwhile Arista's multi-year purchase commitments tripled as networking supply tightens. When the component suppliers start financing and rationing simultaneously, the supply chain is the constraint speaking.

The moratorium count ticked again. Another city adopted a 120-day pause this week to re-evaluate its data center stance, and a 500-acre rezoning proposal hit the usual resistance. The background hum continues.

AWS committed $48 billion to India, $21 billion of it for cloud and AI infrastructure through 2030, a reminder that every domestic constraint we track has an international escape valve, exactly as the Issue #17 analysis predicted.

The Watchlist's Tier 3 is only as good as its coverage, and this readership lives in the corridors. If there's a county commission, town council, or state legislative race where data centers are an explicit issue in your area, reply with "BALLOT" in the subject line and two sentences on the race. Every verified submission goes into the down-ballot tracker, and the post-election issue will credit the network that built it.

That's three reader networks now: RATING (challenge the index), MAP (join the supplier list), BALLOT (build the election tracker). This is what "the hub for data center data" looks like as it assembles, the readership is the sensor network.

Six months ago, "data centers could decide elections" was a media thesis. Today it's a nine-figure spending war, two ballot referendums, a Senate race with an AI dividend proposal, three governor's races carrying regulatory frameworks on their backs, and an industry so aware of its own polling that its advertisements won't say its own name.

Here's the analysis you won't find anywhere else, because it comes from watching both sides all year: the money and the votes are pulling in opposite directions, and November 3 is the first national measurement of which force is stronger. The industry has the capital, the lobbying infrastructure, and $3 trillion in committed projects. The opposition has the polling, the down-ballot energy, and a proven playbook for firing incumbents. The Illinois primaries suggested money alone doesn't close the gap. The Ohio signature failure suggested energy alone doesn't either.

Whichever way the races break, the morning after November 3, this newsletter will publish the scorecard, move the ratings, and count the wave. The Watchlist is now standing infrastructure, like the index, like the Map.

Where the risk sits. What the machines do. Who decides. That's the hub.

Eleven weeks.

The DC Pipeline tracks data center construction, policy, and market intelligence across North America. Home of the State Risk Ratings and the Midterm Watchlist.

Know a local race where data centers are on the line? Reply with "BALLOT" and two sentences.

Think a State Risk Rating is wrong? Reply with "RATING" bring receipts.

To join the Power Equipment Map early access list, reply with "MAP."

Forward this to someone who votes in a corridor. They can subscribe at thedcpipeline.com.