Most of the noise about AI regulation is about chatbots. What they can say, who they can talk to, and which level of government gets to police them. It makes for loud headlines. But if you own AI stocks, or an index fund that is quietly full of them, the rules worth watching may be the dull ones. Who pays for the power lines. How fast a substation gets permitted. Whether a data center can break ground at all.
That distinction matters more than usual right now. The fight over who writes the rules for AI has flared up again, and it is landing on top of a build-out that is already stretching the balance sheets of the biggest companies in the market.
The federal push for a single rulebook
On December 11, 2025, President Trump signed an executive order titled “Ensuring a National Policy Framework for Artificial Intelligence,” according to a summary from the law firm Latham & Watkins. Its stated goal was a “minimally burdensome national policy framework” for AI, meant to replace what supporters call a patchwork of state laws. The order created an AI Litigation Task Force to challenge state AI laws, told the Commerce Department to identify conflicting ones, and pointed agencies toward conditioning some federal grant money on states holding back. The Colorado AI Act was named as an example of what the administration had in mind.
Then came a second step. On March 20, 2026, the White House released a National Policy Framework with legislative recommendations, as described in an alert from Ropes & Gray. It asks Congress to broadly preempt state AI laws that impose “undue burdens.” The framework itself is not binding law, and the same alert notes that Congress has previously declined to adopt broad preemption, including in the One Big Beautiful Bill Act and the National Defense Authorization Act. Until Congress acts or courts side with the administration, state laws stay on the books.
So the headline is simple. Washington wants one rulebook, and it has not yet gotten Congress to write it.
The states are not folding
Resistance has been steady and, notably, bipartisan. In November 2025, 36 state and territorial attorneys general signed a letter opposing a federal ban on state AI laws, according to the National Association of Attorneys General. On September 23, 2026, a coalition of 26 attorneys general led by New York Attorney General Letitia James wrote to congressional leadership, as summarized by the law firm Troutman Pepper Locke. Their ask is worth reading carefully. They are not saying Washington should stay out. They want Congress to set comprehensive federal rules for frontier AI, and they want those rules to include an explicit prohibition on preempting state AI laws.
That is a different argument than the one you usually hear. It is not regulation versus no regulation. It is a floor versus a ceiling. Washington’s proposal reads like a ceiling, and the states are asking for a floor they can build on.
For investors, the practical read is uncertainty with no obvious expiration date. Nobody can say from the outside how this resolves, and anyone who claims to know is guessing.
The carve-outs tell you where the real friction is
Here is the detail that deserves more attention than it gets. Even the administration’s own proposals leave room for local control in the places where AI touches the physical world. The December executive order, per Latham & Watkins, carves out child safety, state government use of AI, and AI compute and data center infrastructure, other than generally applicable permitting reforms, from the preemption legislation it calls for. The March framework, per Ropes & Gray, goes further and calls for protecting residential ratepayers from data center costs while streamlining permitting.
Read that twice. The people arguing for a single national standard are, in the same breath, acknowledging that data centers and the electricity bills attached to them are a local political matter. That is not an accident. A chatbot rule is abstract. A neighbor’s power bill is not.
Why the physical build is the exposed flank
Now put those carve-outs next to the numbers. Gartner projects that global data center electricity consumption will grow 26% in 2026, according to a July 7 report in Network World, with the United States accounting for roughly 36% of the global total. Gartner economist Linglan Wang is quoted saying AI capacity is now constrained by power availability, and that grid supply is expected to fall short of what future data center construction will need. Gartner’s own forecast, it should be said, already factors in parts shortages, delayed or canceled projects, and the conflict with Iran.
Meanwhile the money is enormous. A recent HB Capital note pulls together several estimates. J.P. Morgan puts 2026 hyperscaler capital spending at $697 billion, while CreditSights puts the top five spenders near $602 billion, a 36% increase over 2025. The gap comes from who is counted and how, which is a reminder that these figures are estimates, not audited totals. Back in May, Yahoo Finance reported a Reuters podcast host saying hyperscaler spending had topped $700 billion, and the host added that the number was rising all the time. Different counts, same order of magnitude.
What matters more than the total is how it is being paid for. According to that same HB Capital note, Epoch AI, working from SEC filings, projects that the combined cash capital spending of Microsoft, Amazon, Alphabet, Meta, and Oracle will overtake their combined operating cash flow around the third quarter of 2026, with the timing plausibly landing anywhere from the second to the fourth quarter. CreditSights data cited there says new debt financing rose from 9% of hyperscaler capex in fiscal 2024 to 32% on a trailing basis by mid-2026.
This is an analytical framing, not a forecast, but follow the chain. Spending that is increasingly financed with borrowed money needs the data centers to come online on schedule. Data centers coming online on schedule need power. Power needs interconnection, permits, and public consent, and those are decided in county boards, state utility commissions, and statehouses. A delay does not pause the interest meter. It only pauses the revenue.
What a thoughtful investor can watch
None of this says the AI build-out is a mistake, and nothing here is a call on any stock. The demand side of AI is real enough that companies are borrowing against it. The point is narrower. The risks that get discussed most, such as model safety rules and federal preemption of chatbot laws, may matter less to cash flows than the quiet local decisions about who bears the cost of power.
So what are the useful signals? One is whether Congress actually moves on preemption, given its past reluctance. Another is whether the states that host the most data center capacity change how they allocate grid costs, since the White House framework itself flags ratepayer protection. A third is whether financing conditions shift for the companies leaning hardest on debt, because the HB Capital note observes that credit agencies have already flagged pressure at Oracle, one of the heaviest borrowers in the group.
Each of those is observable in public filings, regulatory dockets, and company disclosures. Each is more concrete than a prediction about which AI model wins next quarter.
The loudest part of the AI regulation story is the part about who gets to write rules for software. The part that touches returns may be the part about who gets to say yes to a transmission line. Keep an eye on both, but know which one has a meter running.
This article is written for educational and informational purposes only and does not constitute financial or legal advice. The views and analytical frameworks presented draw on publicly available information and reported commentary from industry participants. Readers are encouraged to consult primary sources and form their own informed views on these complex topics.
