The White House Super Intelligence Accord: what was signed, and what it means for AI infrastructure

01 / What happened on 29 September

President Donald Trump and House Speaker Mike Johnson hosted leaders of the largest US AI developers and chipmakers at a White House lunch on 29 September 2026. Reported attendees included Mark Zuckerberg (Meta), Elon Musk (xAI), Dario Amodei (Anthropic), Sundar Pichai (Google), Jensen Huang (Nvidia), Greg Brockman (OpenAI), Jeff Bezos (Amazon), Satya Nadella (Microsoft) and Alex Karp and Shyam Sankar (Palantir). [2][7]

Six companies signed a one-page document titled “White House Accord on Super Intelligence: Joint Commitment on Frontier Responsibilities”: Google, Anthropic, Meta, OpenAI, xAI and Nvidia, represented by Pichai, Amodei, Zuckerberg, Brockman, Musk and Huang. Executives from Amazon, Microsoft and Palantir attended but were not among the reported signatories. [3][4][6]

The same day, the President signed an executive order replacing “artificial intelligence” with “Super Intelligence” in federal usage. Asked whether the accord obliges the companies to do anything, he described it as “morally binding”. He also floated a committee of about ten people to oversee the industry, without specifying its membership or authority. [1][2][3]

02 / What the accord commits companies to

The document runs to just over 300 words. It states that each company is responsible for developing its own technology safely and in a way that builds trust, and lists four layers of controls each signatory will put in place. [3][4]

The signatories also agree to meet regularly on standards and leave open the possibility of incorporating these measures into legislation or regulation. The accord creates no regulator, penalty, disclosure requirement or definition of what counts as adequate control. [4][6]

03 / Executive-branch terminology and a legislative proposal

The order, “Inaugurating the Era of Super Intelligence”, directs executive departments and agencies to use “Super Intelligence” and “SI” in official correspondence, public communications, websites, reports and policy documents. Documents issued earlier do not need to be revised. [1]

For now, “Super Intelligence” is defined by reference to the existing statutory definition of artificial intelligence. The Assistant to the President for Science and Technology has 60 days to propose legislative language for a federal definition and any conforming amendments. The order itself does not amend the statutory definition. Its implementation definition may be superseded by subsequent presidential action consistent with applicable law or by Congress; the order does not freeze existing programs. [1]

04 / How it was received

Supporters presented the accord as a starting point. White House adviser David Sacks wrote that only the President could convene the leaders of chips, data centers and frontier models, and described the AI build-out as larger than the railroads, canals and grid combined. Zuckerberg called the commitment to internal controls and multiple audit layers “a significant positive step”; Huang said the companies building the technology carry primary responsibility for deploying it safely. [8][9][11]

Critics focused on enforcement. Academic Gary Marcus read the text as an agreement not to be regulated. Senator Richard Blumenthal called the pledge “a bad deal” and argued for legal accountability through his proposed AI Risk Evaluation Act. Toby Walsh of UNSW, quoted by Al Jazeera, asked what other trillion-dollar industry “marks its own homework”. [6][12][13]

Some assessments were mixed. Helen Toner of Georgetown’s Center for Security and Emerging Technology described the commitments as “pretty minimal”, but saw meaning in more companies accepting embedded auditors under race conditions. [14]

05 / What was said at the briefing: nine speakers, and the lines worth keeping

After the lunch, the President and the executives took questions for about 32 minutes on the driveway outside the West Wing. The Factbase transcript records nine speakers besides the reporters; the President spoke 57 per cent of the words. Unless otherwise noted, quotations below are drawn from the Factbase transcripts of the driveway briefing and the earlier East Room remarks. [21][22]

President Trump. On enforcement: “there’s going to be a tremendous self-policing aspect”, plus “a committee of sorts where we put maybe 10 people on that committee”. On data centers he was specific: companies would be “forming partnerships with communities when they open up data centers”; residents could get “a dividend”; he cited Loudoun County; he wants them “built in America”, mentioning one “going up in Finland”; and, asked by Musk whether they should stay away from homes, he said “I want them to be in the right location within a community”. In the East Room earlier he had named what the companies might supply: “oil and gas … education … helping teachers along financially”. [21][22]

Speaker Johnson. He framed the midterm message as “we cannot lose the race to China”, then turned to cost: “it has to be a balance. You have to take the community concerns in hand.” He pointed to H.R. 9340, the Ratepayer Protection Act, which the House passed 417–3 on 16 September. If enacted, it would require state regulators and nonregulated utilities to consider adopting a standard under which covered data centers with peak demand of at least 100 MW pay the full incremental costs of generation, transmission and distribution upgrades needed to serve them. Adoption would remain discretionary. In his words, the companies “will produce it themselves. They will put power back onto the grid in most cases … Trust is earned. It’s not automatic.” On 30 September, cloture on the motion to proceed failed, 57–43, short of the 60 votes required. [21][23][24][25]

Jensen Huang, Greg Brockman, Shyam Sankar. Huang: “this is the largest industrial revolution in history. And we’re going through the single largest infrastructure build out in human history”, “creating a million jobs here in the United States”, and “There’s no conflict between innovation, technology and safety.” In a separate panel appearance that day, Huang said: “We used to call them data centers … But the things that we’re building now, these are really SI factories.” Brockman: “We build artificial intelligence to benefit everyone … a whole renaissance of small businesses, scientific discoveries.” Sankar (Palantir): “being the leader in America in AI is the most important thing for national security … Winning the AI race is winning.” [21][26]

Dario Amodei, Mark Zuckerberg, Sundar Pichai. Amodei: “AI has incredible benefits … but I think the technology has very real risks. And … the mechanism, how we address those risks is still under discussion”; “we can win and we can win safely.” Zuckerberg: the purpose is to “give the American people and our customers confidence that the technology works in the way that we intend”; “every company is building something that’s a little bit different”; “The idea isn’t that this is the only thing that we will ever do. It is that this is a start.” Pichai: “we are signing up to a set of processes and controls like we do in other areas like financial controls”; “You can see it in the … GDP growth we see.” [21]

Elon Musk. “I think by far the most likely outcome is an age of abundance … where we don’t have just a universal basic income, we have universal high income”; with robots and superintelligence “everyone in the world can have better medical care than anyone here, including me.” A reporter interjected: “Not if they haven’t got a job.” Musk: “jobs are going to change. … jobs have always changed … being a computer used to be a job. There used to be buildings full of people doing calculations … I don’t think anyone is … pining to go back and do calculations all day”; he expects “a tremendous demand for people”. He has used the phrase since at least a 2023 conversation with then UK Prime Minister Rishi Sunak and did not say how a universal high income would be funded; Gulf News listed the forecast’s preconditions, including “plentiful energy and computing capacity”. The IMF estimated in January 2024 that almost 40 per cent of global employment is exposed to AI. In advanced economies, about 60 per cent of jobs are exposed; roughly half of those exposed jobs may benefit from AI integration. It also judged that AI would likely worsen overall inequality in most scenarios. [16][17][18][19][21]

The remarks travelled widely in translation. One Chinese-language subtitled version of Musk’s answer had about 120,000 views by 1 October; its commentary adds claims that are not in the remarks, such as a medical robot costing a few thousand dollars and wealth being redefined by energy and compute. We quote the remarks themselves. [20]

Our analysis. Read across the nine speakers, the briefing was about three questions: who polices the models, who pays for the power and the grid, and who benefits locally. The accord answers the first with internal controls and external review. On power costs, Johnson pointed to the pending Ratepayer Protection Act; the broader promises about self-generation and community benefits were not commitments set out in the accord. Our reading is that developers should assess these expectations against each project’s utility terms, applicable rules and local agreements. [21][24][25] Musk’s abundance thesis points the same way when read as an infrastructure statement: if robots and superintelligence do the work, the scarce inputs become electricity, compute, sites and materials. Two cautions remain. Universal high income is a distribution claim whose mechanism was not stated; productivity gains could benefit both workers and capital owners, with their distribution depending on adoption, competition, ownership and policy. [19] And the “computer” analogy describes a destination, not the transition that the reporter asked about, which the accord does not address. For INTERTIE.ai the practical reading is narrow: inference serving healthcare, education and robotics would be demand for capacity near people, and we treat it as a scenario to test with customers, not as a forecast.

06 / What it means for AI infrastructure

This section, like section 05’s closing paragraph, is our interpretation. The accord governs how model developers supervise their own systems. It does not limit training runs, chip purchases or facility construction, and the executive order does not change any permitting, energy or procurement rule. At the federal level, the posture toward building compute remains supportive. Pichai, for example, cited Google investment of hundreds of billions of dollars over the past two years. [1][4][10]

The binding constraints on delivery remain physical and local: firm power, grid interconnection, cooling, permitting and community acceptance. NBC reported that the President pledged data centers would benefit local communities, while expansion remains unpopular with some voters ahead of the 3 November midterm elections. The accord does not address these questions. [5]

We expect some indirect effects on infrastructure providers, although none are stated in the accord. Signatories that commit to monitoring, external evaluation and board-level reporting may ask compute and hosting partners for stronger access controls, logging, incident cooperation and auditor access. Microsoft and Amazon were not among the reported signatories. We expect any requirements for infrastructure partners to vary by customer. [3][4][6]

For INTERTIE.ai, this research informs how we assess AI-ready infrastructure. Our adopted model integrates campus development, company-owned GPUs and cloud operations; GPU deployment and cloud services are in development. The analysis does not change the commercial status of any project or establish existing ownership or available capacity.

07 / What would change the picture?

Six developments would tell us whether the accord becomes more than a statement of principles. First, the proposed legislative language that the President’s science and technology adviser must submit to the President within 60 days of the order, around late November 2026. Second, the membership and authority of any oversight committee. Third, whether non-signatories such as Microsoft and Amazon join. Fourth, congressional action after the midterm elections. Fifth, international follow-through, including the AI communication channel with China that the White House announced in September. Sixth, the Ratepayer Protection Act, passed by the House 417–3 and stopped at a Senate cloture vote on 30 September, or any state rule that moves large-load costs onto the loads themselves. [1][2][15][24][25]

For infrastructure, the more useful evidence will be in contracts rather than communiqués: security and audit clauses in compute agreements, interconnection queues, state and local data-center rules, and the price of firm power. Publishing evaluation results could influence customer expectations. Binding requirements for infrastructure providers would depend on the scope of enacted legislation or the terms of individual contracts.