U.S.-China Relations
NextFin News - House Speaker Mike Johnson is pressing pause on federal artificial-intelligence legislation, saying any rules "shouldn't be rushed" and that he wants to "summon" AI industry leaders to "one big meeting" before Congress writes anything into law. The stance, voiced September 13, 2026, sets up a direct collision with a legislative calendar that is already running out of runway: the House has roughly one week in session before a six-week recess that extends past the November midterm elections. The question is not whether Washington will eventually regulate AI. It is whether anything binding will exist before voters head to the polls — and the answer, so far, is no.
Johnson's call for a convening arrives at the moment when pressure on Congress to act has rarely been higher. A group of House Democrats — Representatives Sam Liccardo, George Whitesides, Lori Trahan and Ted Lieu — has urged the Speaker to cancel the upcoming recess and keep the chamber in session until AI safeguards pass. Their letter cites "catastrophic risk posed by advanced artificial intelligence," including "mass cybersecurity breaches and development of biological or chemical weapons." The immediate trigger was a set of public warnings from current and former Anthropic researchers that frontier AI could pose an extinction-level risk within a decade, landing the same week Anthropic's own September threat-intelligence report disclosed that it had disrupted Claude misuse spanning bioweapons research and state-sponsored surveillance operations.
Yet the only AI-adjacent bill Johnson's conference is actually moving this month has nothing to do with model safety. The bipartisan Ratepayer Protection Act, introduced by Representative Gabe Evans of Colorado, is set for a House vote the week of September 14. It passed the Energy and Commerce Committee unanimously, 52-0, in July and carries 41 cosponsors. Its mechanism is narrow and consumer-facing: it would establish a federal standard requiring data centers to pay the full incremental cost of the grid upgrades needed to serve them, rather than shifting those expenses onto households, with each state's public-utility commission required to consider adopting the standard. A Johnson spokesperson, Taylor Haulsee, confirmed the floor plan:
"The bill will be on the floor of the House next week."
Meanwhile, the most concrete AI regulation in the United States is being written in Sacramento, not Washington. On September 9, California Governor Gavin Newsom signed Senate Bill 813 and Assembly Bill 1405, creating the nation's first third-party AI audit framework. SB 813 establishes a California Artificial Intelligence Standards and Safety Commission to recognize the independent panels — called Independent Verification Organizations — that will assess AI systems for compliance with state law. AB 1405 creates a state registry for AI auditors and sets standards for their independence, transparency and integrity. The auditor registry does not take effect until January 2029, but it is a dated, enforceable requirement in a way nothing moving through this Congress currently is. Both OpenAI and Anthropic — rivals at the frontier — endorsed the bills, and Newsom used the signing to call on the federal government to "establish robust national regulations."
That is the landscape Johnson's "one big meeting" is meant to manage: safety advocates demanding emergency action, a single cost-allocation bill reaching the floor, and states filling the vacuum with rules the White House would rather preempt.
Why a Meeting Is the Perfect Substitute for a Mandate
Johnson's sequencing is not an accident; it is consistent with a record he has been building for years. In a 2024 Capitol Hill interview, he said he was "very wary" of any "red tape" that would stifle AI development and that Congress should not create "some massive regulatory scheme." The House Bipartisan AI Taskforce, he said then, would produce not specific legislation but "an overarching set of principles that we would want to follow if legislation comes from that." By March 2026, his tone had sharpened toward speed — but only on one kind of legislation. Speaking at The Hill and Valley Forum in Washington, Johnson called on Congress to swiftly enact President Trump's AI agenda, arguing that "a single federal standard should preempt state laws governing the industry and protecting children." Then he drew the line that defines his position:
"Inaction is unacceptable, and while the window for American AI leadership is open, it will not stay open indefinitely."
Read together, the two positions are not contradictory. Swift enactment applies to preemption and competitiveness — rules that clear the path for American builders. Caution applies to safety constraints — rules that slow them down. A meeting sits perfectly between the two: it signals engagement to the safety camp, delivers no binding obligation to the industry, and consumes calendar time that a divided chamber facing midterms is happy to spend.
The mechanism is straightforward. In a Congress where the House majority is thin and the Senate's appetite for AI safety is untested, process is a low-cost substitute for policy. Convening industry leaders lets leadership say it is doing something without forcing any member to take a vote that a tech donor, a power utility, or a civil-liberties group could attack. The meeting is not a delay tactic in the crude sense. It is a sequencing device that puts industry consultation before statutory text — and in a session with one working week left, "before" can easily become "never this Congress."
The political arithmetic explains why. With the House returning November 9, days after the midterm elections, any AI regulation vote before then carries risk for incumbents who rely on tech-industry campaign support. A letter from four Democrats asking a Republican Speaker to cancel recess is, on its own, a press release rather than a legislative signal. The one development that would change that calculus — a Republican co-signer — has not materialized.
The Only Bill With Momentum Is About Electricity, Not Intelligence
It is revealing that the legislation most likely to clear the House before the recess is the Ratepayer Protection Act rather than any safety bill. The reason is not technical; it is political. A data center's effect on a household electricity bill is visible, monthly and blameable. An extinction-level AI risk is abstract, probabilistic and temporally distant. One of these gets votes in the weeks before an election; the other gets a hearing.
The bill also lands on fertile ground because the industry's own promises have not convinced the public. In March 2026, seven major AI companies — Amazon, Google, Meta, Microsoft, OpenAI, Oracle and xAI — signed a White House Ratepayer Protection Pledge committing to cover their facilities' energy costs. Two months later, a nationally representative Consumer Reports survey of 2,082 US adults found that a large majority remained skeptical that large data-center developers would actually pay the full cost of the power their facilities require. Chris Harto of Consumer Reports called the committee's unanimous passage a signal of "strong bipartisan support for holding massive tech companies accountable for the costs associated with AI data centers," while noting the legislation alone is "insufficient" to ensure companies cover their full costs.
Representative Evans framed the choice in explicitly competitive terms:
"As the United States competes to win the global AI race and outcompete adversaries like Communist China, we must ensure that growth does not come at the expense of hardworking families."
That sentence captures the entire political sweet spot: growth is good, but not if voters see it on their electric bill. It is the one AI position that is safe to vote for in a midterm year.
The economics behind the politics are real. Data-center electricity consumption is set to more than double to around 945 terawatt-hours by 2030, according to the International Energy Agency, slightly more than Japan's total consumption today, with AI the dominant driver. In the United States, data centers account for nearly half of electricity-demand growth between now and 2030. When a region's grid needs billions in upgrades to serve a new hyperscale campus, someone pays — and the Ratepayer Protection Act is Congress's attempt to make sure that someone is the data-center operator, not the household downstream.
States Are Writing the Rules Congress Won't — and Industry Knows It
California's audit laws are the clearest evidence that the regulatory center of gravity has shifted to the states. What makes SB 813 and AB 1405 politically significant is not just their scope but their backing: OpenAI and Anthropic, whose rivalry defines the frontier, endorsed the same state rules. That convergence is a signal that parts of the industry now prefer a known state compliance regime to an open-ended federal debate.
The federal government has tried to reclaim the field. On March 20, 2026, the White House released its National Policy Framework for Artificial Intelligence, urging Congress to establish a federal framework that would preempt the emerging patchwork of state AI laws. A companion House bill, the American Artificial Intelligence Leadership and Uniformity Act (H.R.5388), calls for federal preemption that is "precise, time limited and clear in scope." President Trump also pushed to include state-AI-law preemption in the fiscal 2026 National Defense Authorization Act — a proposal that was quickly opposed by House Armed Services Committee Chair Mike Rogers, Republican of Alabama, and Senate Armed Services Ranking Member Jack Reed, Democrat of Rhode Island, and was ultimately excluded from the final package.
That failure is instructive. Preemption is where the pro-innovation coalition fractures: national-security hawks and state attorneys general do not want Washington stripping states of AI oversight, even when the White House asks. The result is a stalemate that benefits the states. Beyond California, Pennsylvania's legislature passed data-center energy and water reporting requirements in July 2026, Governor Josh Shapiro signed a data-center executive order in August, and Massachusetts, New York and Oklahoma have moved on clean-power mandates for data centers.
The transmission mechanism runs in both directions. State rules create the compliance patchwork that industry says it wants Washington to fix — but only after the states have already set the baseline. By the time a federal standard arrives, it will be negotiating upward from California's floor, not downward from a regulatory blank slate. Johnson's meeting is where that negotiation begins informally: industry leaders will not be coming to discuss whether rules should exist. They will be coming to discuss which rules, and on what timeline.
The Market Has Priced a Light-Touch Regime — and That Is the Story
The equity market's read on all of this is embedded in capital-expenditure expectations, not in legislative calendars. NVIDIA has guided that capital spending from the five largest hyperscalers will reach nearly $800 billion in 2026 and $1.3 trillion in 2027. NVIDIA's own market capitalization stands at roughly $5.3 trillion. Those figures are not pricing a safety crackdown. They are pricing a multiyear buildout in which the binding constraint is power and chips, not permits.
The second-order implication is what matters. Investors are not underwriting the risk that Washington shuts the AI buildout down. They are underwriting a narrower, more manageable risk: that the cost of the buildout gets re-allocated. Ratepayer-protection bills, state clean-power mandates and local moratoria do not stop data centers; they change who pays for them and where they can be sited. That is a margin issue, not a thesis issue — which is exactly why a "one big meeting" with no deadline is the outcome the market prefers. It preserves optionality.
The asymmetry cuts the other way for a specific set of players. States that set standards early — California first among them — gain a first-mover advantage in shaping the compliance industry: auditors, verification bodies, legal and consulting capacity will cluster around the strictest regime. Companies that have already signed long-term power-purchase agreements in friendly jurisdictions are insulated. The exposed parties are households in states without cost-allocation rules, smaller AI labs that cannot absorb a state-by-state audit burden, and House incumbents in competitive districts if power prices stay elevated through the fall campaign.
The Counter-Thesis, the Signals, and the Scenarios Ahead
The strongest case against reading Johnson's pause as obstruction is also the simplest: AI moves faster than legislation, and bad rules are worse than no rules. A chamber that has spent most of 2026 unable to pass anything comprehensive may do more damage by rushing a framework that hardens the wrong assumptions, locks in compliance costs for startups, or hands a competitive advantage to China. The House Bipartisan AI Taskforce's original timeline — a report by November — reflects exactly this caution. From this view, a meeting that gathers the builders before writing the rules is not delay; it is due diligence. The counter-thesis is backed by the same instinct that animated the taskforce's design: regulate the outcomes, not the technology, and do it once you understand what you are regulating.
That argument is coherent as far as it goes. But it does not survive contact with the calendar. The risk of acting too fast is real; the risk of not acting at all is being externalized onto ratepayers and onto a public that a majority already distrusts the industry's cost promises. A meeting with no date, no agenda and no deadline is not due diligence. It is a holding pattern. And a holding pattern in a six-week recess, with states writing enforceable rules in the interval, is itself a policy choice — one that defaults to the industry's preferred timeline.
The base case is clear: no comprehensive AI safety bill reaches the House floor before the midterm elections. The Ratepayer Protection Act vote next week is the concrete federal action to watch, and its passage would confirm that cost allocation, not model safety, is the only AI policy with bipartisan legs in 2026. After the recess, the lame-duck session could revive preemption language if Republicans retain control, but that is a 2027 story at the earliest.
Two signals would falsify that base case. First, if a Republican member of the House adds their name to the Liccardo letter, the safety push becomes bipartisan rather than partisan, and leadership can no longer treat it as a press release. Second, if a high-profile AI incident — a bioweapons-adjacent misuse, a large-scale cybersecurity breach traced to a frontier model, or a visible grid failure tied to data-center load — occurs before the recess ends, the political cost of inaction flips overnight. Either one would turn a meeting request into a vote.
Split by horizon, the picture is mixed. In the short term, sentiment favors the buildout: the meeting announcement removes the near-term risk of a surprise safety bill, which is mildly positive for AI infrastructure names and the utilities that have signed long-term data-center power deals. Over the medium term, the compliance burden migrates to the states, and California's January 2029 auditor registry becomes the de facto national standard that multi-state operators must design around. Over the long term, the structural question remains unresolved: the United States still has no federal AI statute, and every month without one increases the odds that the eventual law is written in reaction to a crisis rather than in anticipation of one.
Scenarios, from most to least likely: the base case is a convened meeting, a principles document, and no safety legislation before the midterms. The upside case for industry is a lame-duck preemption bill that locks out stricter state rules and gives builders a single federal standard. The downside case is an incident-driven emergency session that produces rushed, poorly targeted constraints — the very outcome Johnson's pause is meant to avoid, and the one his delay makes more likely.
Johnson is right that AI legislation should not be rushed. But a meeting is not a strategy, and a calendar is not an excuse. The window for American AI leadership may not stay open indefinitely — but the window for Congress to shape it on its own terms is closing faster, and the rules that fill the gap will be written in Sacramento first.
Explore more exclusive insights at nextfin.ai.
