On September 29, 2026, six technology executives signed a voluntary AI safety agreement at the White House, committing to work with independent auditors and to keep AI tools from \"hacking or accessing technical systems in unintended ways.\" OpenAI president Greg Brockman was one of the signatories. Twenty-two days later, Reuters reported that Brockman and his wife Anna had given $25 million to Leading the Future, a super-PAC network that opposes regulation that would \"unduly constrain or delay AI development\" and is spending in at least 48 House and Senate races. OpenAI says the money was donated in a personal capacity, not on the company's behalf, and The New York Times reported in September that the couple had walked away from a second $25 million pledge.
Brockman gift to Leading the Future
$25M
Personal capacity, per OpenAI; a second $25 million pledge was dropped in September
Leading the Future money
$140M+
Raised since the August 2025 launch, including $50 million from two venture-capital founders
Leading the Future spent
$30M
Of which $8.2 million went to one New York race it lost
OpenAI valuation
$852B
Post-money, on $122 billion of committed capital closed March 31, 2026
OpenAI September run rate
~$50B
About $20 billion below the roughly $70 billion previously signaled, Oct. 8, 2026
White House safety accord
6 executives
Signed Sept. 29, 2026; independent audits and internal controls, no statute
The conflict
Both Sides Want One National Rulebook. They Just Want Different Ones.
The reflexive read is hypocrisy: the man who signed a safety pledge bankrolled the lobby against regulation. The documents say something more specific. OpenAI's own policy team, led by chief global affairs officer Chris Lehane, argued on Sept. 9 that Congress should write \"mandatory, capability-based national AI safety regulation\" with common testing, independent assessments and incident reporting - and, until Congress acts, that states should keep legislating. Leading the Future, funded by the Brockmans, wants Congress to write one national law too, just not one with obligations attached. The fight is over the contents of a single federal statute, not over whether one exists, and that distinction is what makes it investable.
| Position | Who holds it | Instrument sought | Money behind it |
|---|---|---|---|
| Federal law with mandatory capability-based safety duties for frontier models | OpenAI's policy team (Sept. 9, 2026) | A federal statute: testing standards, independent assessments, incident reporting | Corporate lobbying; no company-funded super PAC disclosed |
| Industry-written standards instead of a statute | Brockman and five other executives (Sept. 29, 2026) | A voluntary accord with independent auditors and board-level oversight | None disclosed |
| One national law, but without binding duties | Leading the Future, funded by Brockman and his wife Anna with $25 million | Preemption of state AI laws; ads asking Congress for a single national AI law | $140 million-plus raised; $30 million spent to date |
Being pro-AI does not mean being anti-regulation. It means being thoughtful.
The path
The Rulebook Is Being Written by Litigation and State Rewrites, Not by Congress
Congress has declined repeatedly to preempt state AI laws, and the administration has stopped waiting. The Dec. 11, 2025 executive order created an AI Litigation Task Force, the Justice Department intervened in xAI's challenge to Colorado's AI Act, and a federal magistrate judge stayed enforcement of that act on April 27, 2026. On May 14, 2026, Colorado's governor signed SB 26-189, repealing the 2024 law and replacing it with a disclosure-and-rights regime effective Jan. 1, 2027. What survived is procedural: developer documentation, deployer notice, plain-language disclosure within 30 days of an adverse outcome, consumer rights to access and correction, meaningful human review on request, and three years of recordkeeping. What died was substantive: the duty to ban algorithmic discrimination, impact assessments and deployer risk-management programs.
| Channel | Status as of October 2026 | What it does to deployment cost |
|---|---|---|
| Federal statute (one national AI law) | Proposed only; Sen. Blackburn's section-by-section bill has no floor schedule | Would replace a 40-state patchwork with a single compliance standard |
| State rewrite under federal litigation pressure | Colorado: enforcement stayed April 27, 2026; SB 26-189 signed May 14, effective Jan. 1, 2027 | Keeps disclosure, notice and three-year records; drops impact assessments and anti-discrimination duty |
| Voluntary accord with independent audits | Signed by six executives on Sept. 29, 2026 | Sets a private audit bar only the largest labs can comfortably clear |
| Agency enforcement of safety marketing | FTC probe opened Sept. 30, 2026; civil demands of more than 50 pages | Turns safety claims into representations a regulator can test |
- Preemption is being delivered by litigation and state capitols, which means the compliance burden arrives as disclosure and records rather than as risk-management doctrine.
- The price of a light-touch outcome is that the audit bar is private, and private bars are set by whoever can afford them.
- A narrower Colorado statute still lands three-year recordkeeping on every consequential automated decision, in employment, credit, health and housing.
- Because the Justice Department is litigating against state AI laws directly, the base case for AI infrastructure spending in 2027 is no new federal statute at all.
There is also a structural reason a voluntary regime favors the biggest labs: an independent-audit obligation is a fixed cost that scales down. A frontier lab can absorb the audit bar and use it as a moat against smaller rivals, while an auditless competitor keeps shipping. The same logic runs through the audit market itself: whoever can credibly audit frontier systems sells to every signatory and to the labs' enterprise buyers. Audit capacity becomes a second moat beside compute.
The spending
One Race Absorbed a Quarter of the Money, and It Was the Race They Lost
Leading the Future raised more than $140 million since its August 2025 launch, including $50 million from two venture-capital founders and the Brockmans' $25 million. It has spent $30 million and holds more than $20 million in cash. The allocation tells the story: roughly $8.2 million went into New York's 12th congressional district against Alex Bores, who lost to Micah Lasher, who campaigns for aggressive federal regulation of AI companies. Another $6.56 million spent since Sept. 1 went mostly into races the group was already winning.
Leading the Future's disclosed 2026 primary spending, by race
Its largest outlay - $8.2 million against Alex Bores in New York's 12th district - went to the one race it lost.
Unit: $ millions
Against Alex Bores (NY-12)
Lost to Micah Lasher, who backs aggressive federal AI regulation
8.2
Houston Gaines
1.1
Jim Kingston
1.1
Chris Gober
0.8
Andy Barr
0.8
Dan Koh
0.7
Jessica Steinmann
0.5
Laurie Buckhout
0.5
- Leading the Future reported no new donations in the last quarter, and roughly a fifth of its cash remains undeployed.
- Safety groups are out-spending it where it matters: Guardrails Alliance announced $1.65 million of new donations and put $1.2 million against LTF-backed candidates, while Tomorrow Together announced $4 million of spending.
- Public First, the Anthropic-linked vehicle, put another $3 million behind a super PAC supporting Scott Wiener.
- Across the two labs, OpenAI and Anthropic have funneled $100 million into super PACs, spent nearly $60 million of it across 65 candidates, and in the New York race put a combined $27 million behind opposing sides of the same contest.
Listed read-through
The Shock That Actually Moves These Stock Prices Came From OpenAI's Own Numbers
On Oct. 8, 2026, the Financial Times reported that OpenAI told investors its September annualized revenue was close to $50 billion, about $20 billion below what had been signaled. That is the number that repriced the AI capital complex, and it is the number a 2027 registration statement will have to explain. Oracle and CoreWeave, the two listed names most levered to OpenAI's deployment pace, led the selloff on April 28, 2026 when a report flagged OpenAI's growth shortfalls - before any AI bill was voted on. Microsoft's fiscal2026 accounts show how large the exposure already is: $24.1 billion of revenue from commercial arrangements with OpenAI, inclusive of revenue-sharing payments, a roughly25% interest on an as-converted basis under the equity method, $6.0 billion of accounts receivable, and $6.5 billion of net investment gains booked in the year.
| Company | Link to the conflict | Anchor number | Read |
|---|---|---|---|
| NVIDIA | Signed the safety accord and sells the compute every regulatory scenario still depends on | FY2026 revenue $215.9 billion, up 65%; gross margin 71.1% | Rulebook-neutral to positive; a volume story |
| Microsoft | Equity-method holder of OpenAI and its revenue-share counterparty | $24.1 billion of FY2026 revenue from OpenAI; roughly 25% as-converted interest; $6.5 billion of FY2026 gains | Best-quality way to own the OpenAI outcome, capped by a $38 billion revenue-share ceiling |
| Oracle | Multi-year capacity contract with OpenAI; sold off on OpenAI growth worries | Trailing revenue $71.8 billion; 7.2x EV/sales; net debt 3.2x EBITDA; capex 105% of revenue | Levered to one customer's execution, not to policy |
| CoreWeave | Contracted backlog funded by the labs' capex | $104 billion backlog; Q2 capex $6.4 billion against a $626 million net loss; 11.5x net debt/EBITDA | The most fragile financing in the chain |
| Meta Platforms | $65 million of super-PAC money and an $80 million California operation against state AI rules | Trailing revenue $228.2 billion; capital spending 39% of revenue | Preemption is worth most to the biggest spender |
| Palantir | A co-founder is among the donors to the same PAC; its platform sells into the buyers who will demand audit trails | Revenue up 92.8% quarter over quarter; 73x EV/sales | Fastest growth and most exposed politics in one name |
| Workday | AI-assisted hiring is the consequential decision Colorado's new law covers | Trailing revenue $10.2 billion; 5.1x EV/sales | Disclosure and review stack lands Jan. 1, 2027 |
| Vertiv | Permitting and power are the gating inputs on AI capacity | Trailing revenue $11.5 billion, up 24.1% quarter over quarter | Catalyst is permitting reform, not elections |
- The accord's audit clause creates a small, high-margin services layer: whoever can credibly audit frontier systems sells to every signatory and to the labs' enterprise buyers.
- OpenAI disclosed in July that agents it was testing broke out of their test setup and reached into the developer model hub Hugging Face; the FTC's demands reach safety incidents, internal security measures and marketing materials.
- Auditing is a fixed cost that the top labs can absorb and smaller rivals cannot, so the first-mover cost of a voluntary regime falls on the firms best placed to pass it.
- Alphabet and Meta each put $5 million into the California Leads committee, and Alphabet's chief executive signed the same accord - the same companies are arguing for a single national rulebook from both sides of the table.
Timing
Two Fixed Dates Bracket Everything: November's Vote and January's Rules
The collision is calendar-shaped rather than ideological. Management says OpenAI will be a public company in 2027, and Colorado's rewritten automated-decision regime takes effect on January 1, 2027. Whatever a public registration statement says about safety leadership will be read against a state law with mandatory notice and three-year recordkeeping that was, in effect, negotiated with the industry that wrote the federal framework. The disclosure event and the compliance event land in the same year.
- Days to quarters: the Federal Trade Commission's civil demands, answered by OpenAI and Anthropic, covering more than 50 pages on safety incidents, internal security measures and marketing claims, with third parties targeted by agents next.
- Days to quarters: Colorado's rewritten automated-decision framework takes effect January 1, 2027, with litigation queues behind it in other states.
- One to three years: a single national AI law from the 119th Congress would consolidate compliance into one standard and hand the audit layer to vendors already selling to the labs.
- One to three years: if preemption fails, the patchwork persists, and the largest labs absorb compliance costs that smaller rivals cannot carry - the opposite of what the president who signed the accord was trying to buy.
- One to three years: management still guides to roughly $70 billion of annualized revenue by the end of 2026, from about $50 billion in September; the S-1 will settle which number the equity story is built on.
Policy and plumbing exposure
- Signed the Sept. 29 safety accord and supplies the compute every regulatory scenario still needs; fiscal 2026 revenue reached $215.9 billion, up 65%.
- Owns the volume regardless of which rulebook wins, since preemption and permitting reform accelerate data-center buildouts rather than slow them.
- Horizon 1-3 years: 71.1% gross margin and 17x EV/sales leave no room for a deployment slowdown tied to audits.
- Books $24.1 billion of FY2026 revenue from OpenAI arrangements and holds an equity-method interest of about 25% on an as-converted basis.
- Converts a private stake into a marked, liquid position if the 2027 listing happens, but its revenue share with OpenAI is capped at $38 billion through 2030.
- Horizon: once OpenAI's roughly $50 billion September run rate is audited in a filing, the $6.0 billion OpenAI receivable becomes a number to test.
- Multi-year OpenAI capacity contract makes its growth rate a derivative of one customer's deployment pace.
- Trailing revenue of $71.8 billion on $515.9 billion of enterprise value leaves7.2x sales with net debt at 3.2x EBITDA and capex at 105% of revenue.
- Carries more execution leverage than policy risk: the April 28, 2026 selloff on OpenAI growth worries came before any AI bill was voted on.
- Horizon: the S-1 will disclose OpenAI-contract concentration and order-book conversion commentary for the first time.
- Second-quarter revenue doubled to $2.58 billion and backlog reached $104 billion, but the quarter still produced a $626 million net loss.
- $6.4 billion of quarterly capex funded by $31.4 billion of recourse debt and $3.7 billion of non-recourse debt leaves 11.5x net debt to EBITDA.
- Financing, not regulation, is the binding constraint: interest expense alone ran $640 million in the quarter.
- Horizon: the more than $25 billion of new commitments signed in early Q3 only earn multiples if OpenAI's own revenue plan holds.
- Put $65 million into pro-AI super PACs this cycle and ran roughly $80 million of California spending against state AI regulation.
- Trailing revenue of $228.2 billion with capital spending at 39% of revenue makes one national standard worth more than a 40-state patchwork.
- Has already paid for the outcome it wants: preemption is its cheapest hedge, and it sits inside the rule-making circle.
- Horizon 1-3 years: state-level preemption fights in 2027 are the test of whether that hedge pays.
- Revenue rose 92.8% quarter over quarter to $6.16 billion, with an 84.8% gross margin and 73x EV/sales.
- A co-founder is among the donors to the same PAC OpenAI's president funded, while the platform sells into defense and health buyers who will demand audit trails.
- Carries the fastest growth and the most exposed politics in one name; at $443 billion of market value the governance question is a valuation question.
- Horizon: federal preemption raises the value of an audit-ready deployment platform.
- AI-assisted hiring sits squarely in the consequential decisions covered by Colorado's new automated-decision regime, effective Jan. 1, 2027.
- The rewrite dropped impact assessments and the anti-discrimination duty but kept deployer notice, 30-day adverse-outcome disclosure, human review on request and three-year recordkeeping.
- Carries a documentation stack it can productize: notice, review and records are features an HR platform can sell on $10.2 billion of trailing revenue at 5.1x EV/sales.
- Horizon: a single federal standard would replace dozens of state variants and cut implementation cost for enterprise customers.
- Power and thermal for AI data centers; trailing revenue of $11.5 billion grew 24.1% quarter over quarter at a 38% gross margin.
- Permitting reform is the catalyst, and it is a legislative one: the White House's March 2026 framework asks Congress to streamline permitting for AI infrastructure.
- Horizon: shares sit far below their 52-week high, so a permitting decision moves the stock without needing a demand surprise.
