Everyone is still covering data centers like it’s an electrons problem — will the grid supply enough power for the next model-training run. That story is over: the sharper fight breaking out across three states this month is a data center financial assurance requirement, regulators forcing hyperscalers to post cash, bonds, or letters of credit before they’re allowed to build. The reason is blunt — nobody fully believes the AI-demand forecasts hyperscalers are using to justify gigawatt-scale buildouts anymore. I’ve watched this exact kind of over-provisioning bet before, first around CERN’s collision infrastructure and later in data-center land, and this is a credit-risk story wearing an energy-policy costume.
Three states, three totally different wagers on the same underlying risk. Michigan and Wisconsin are treating hyperscalers like subprime borrowers, demanding collateral against the chance a gigawatt campus gets cancelled halfway through construction. Virginia picked a toll instead of a deposit — a straight consumption tax collected only on power actually used. Missouri picked neither, and it just pulled in roughly $25 billion in new projects precisely because it hasn’t asked anyone for money upfront.
Why Virginia’s Tax Isn’t a Data Center Financial Assurance Requirement
Virginia became the first U.S. state to pass a direct data center electricity levy when Governor Abigail Spanberger signed the bill on June 30, 2026, with the tax taking effect the very next day. The mechanics are simple: $0.011 per kilowatt-hour on qualifying data center load, revenue capped at $600 million a year with anything above the cap refunded, and the whole thing sunsets on June 30, 2028. It’s worth reading in full at Kiplinger’s rundown of the new law, but the short version is that it functions exactly like any other consumption tax — you pay for what you actually draw from the grid, nothing more. That distinction matters more than most coverage has given it credit for.
A data center financial assurance requirement is a completely different animal — it isn’t collected on power used, it’s posted against power that might never get used at all. Virginia’s tax still bites hard at scale, which is exactly why it’s useful as a baseline before we get to the collateral states. According to mgrid.org’s analysis, a continuously running 500MW facility owes roughly $48 million a year under the new tax, and a 1GW campus is close to $100 million — about a 10% bump in effective electricity rate.
A toll only fires when the meter is actually running. A deposit fires the moment the shovel goes in the ground, whether the demand ever shows up or not. That’s the whole difference between “we’ll tax your success” and “we’re hedging against your failure,” and it’s why Michigan and Wisconsin’s approach is the one that should actually worry hyperscalers’ finance teams.
Source: mgrid.org analysis of Virginia data center electricity tax, effective July 1, 2026
Inside the Data Center Financial Assurance Requirement: Michigan and Wisconsin
Here’s the reframe nobody’s saying out loud: this stopped being a megawatts story and became a credit-underwriting story. Utility regulators aren’t asking “does the power exist” anymore — they’re asking “who eats the loss if this hyperscaler’s demand forecast is wrong.” That’s the same question a bank asks before it lends to a risky borrower, and states are starting to answer it the same way a bank would: collateral first, construction second.
Think of it like a landlord demanding a security deposit from a tenant with no credit history. Before a hyperscaler can move forward, the utility regulator requires it to post cash, a bond, or a letter of credit covering part of the cost of new transmission and generation capacity built specifically to serve that one customer. If the AI company’s demand forecast doesn’t pan out — the project gets cancelled, the capacity sits overbuilt, the parent company hits a rough patch — the collateral pays down the stranded asset instead of ordinary ratepayers eating it through higher bills. That’s the whole innovation: it turns a forecasting error into the hyperscaler’s balance-sheet problem, not the neighborhood’s.
Wisconsin regulators at the Public Service Commission are already requiring some hyperscale developers to post hundreds of millions of dollars in financial security before construction begins. Oracle didn’t take that quietly — it has asked a Wisconsin court to overturn the requirement, and that case is shaping up as the first real legal test of whether regulators can make this kind of collateral demand stick. However that ruling lands, every other state weighing a similar rule is watching it closely.
Michigan is moving in the same direction from the top down: Governor Gretchen Whitmer’s “Michigan Affordable and Responsible Growth Action Plan” requires data center operators to pay the full cost of their own operations, post financial assurances upfront against future utility demand, and lock in community-benefit commitments before breaking ground. It’s the same instinct that’s reshaping the broader hardware supply chain right now — we wrote about the physical side of that squeeze in our piece on the data center memory shortage, where fab capacity, not policy, is the bottleneck. Policy bottlenecks and physical bottlenecks are starting to rhyme.
Source: Data Center Knowledge, “New Data Center Developments: July 2026”
Missouri’s $25 Billion Bet Against a Deposit
While Michigan and Wisconsin write collateral requirements into law, Missouri is doing the opposite: rolling out the welcome mat. Amazon is planning roughly $10 billion in new investment there, and Google is set to put in about $15 billion on a separate project — neither has been asked to post a dime of financial assurance against the risk that any of it goes sideways. The same week, Microsoft broke ground on a new data center in La Porte, Indiana, a reminder that hyperscalers are still moving fast wherever the friction is lowest. That’s the counterintuitive part: this isn’t really a story about grid capacity anymore, it’s a story about which states are underwriting AI-demand risk like a bank and which ones are still extending unsecured credit.
⚡ PHOTON’S TAKE
Financial assurance requirements are the smartest policy idea to hit data centers in two years, and almost nobody outside utility law is talking about them. Virginia’s tax is a toll — annoying, but it doesn’t stop a bad bet from happening. Wisconsin and Michigan are pricing the actual risk: what happens when a hyperscaler’s demand forecast is simply wrong. Missouri’s $25 billion haul isn’t proof its approach works — it’s proof the deposit hasn’t come due yet. Watch the Oracle case. It decides who pays when AI’s growth curve breaks.
What This Means for the Next Data Center Boom
Expect the collateral model to spread, not stay contained to two states. Once one court upholds a data center financial assurance requirement — and Wisconsin’s Oracle case is the one to watch — every utility commission with a hyperscale campus in its territory gets legal cover to demand the same thing. Missouri’s $25 billion advantage is almost certainly temporary; it’s a first-mover discount, not a permanent policy stance, and I’d bet it closes within two or three legislative sessions.
The long-run winners won’t be the states that stayed cheapest the longest — they’ll be the ones that priced AI-demand risk correctly before a stranded 1GW campus forced the question. Big physics collaborations learn this lesson every time a detector upgrade gets sized for a physics case that shifts underneath it, the same over-provisioning problem we wrote about in our piece on the PLATON particle detector. AI infrastructure is about to get the same discipline the grid already has — the only question is how many stranded gigawatts it takes to get there.
