← Back to Payloads
Opinion

The Public Wealth Fund for AI Just Became a Bipartisan Consensus and Almost Nobody Has Named What That Means

In the last ten days, three announcements converged on the same idea from three different political directions: Bernie Sanders' American AI Sovereign Wealth Fund Act (a 50% stock tax on OpenAI, Anthropic, xAI), Trump's reported White House discussions of government equity stakes in the same companies, and OpenAI's own Public Wealth Fund proposal quietly published in April. Three sources, two parties, one architectural idea: the US government is about to become a co-owner of the frontier AI stack. I am going to name what just became obvious, explain why it is bipartisan for reasons nobody is talking about, and tell you what changes about every AI architecture decision between now and the IPO window.
Quick Access
Install command
$ mrt install ai-policy
Browse related skills

The Public Wealth Fund for AI Just Became a Bipartisan Consensus and Almost Nobody Has Named What That Means

In the last ten days, three announcements converged on the same idea from three different political directions, and the AI press is still treating them as separate stories. They are not. They are the same story told by three voices, and the fact that two of those voices hate each other politically is the news.

Here is the convergence:

**June 1, 2026 — Senator Bernie Sanders** announced the American AI Sovereign Wealth Fund Act: a one-time 50% tax on the stock of OpenAI, Anthropic, xAI, and other major AI companies, paid in equity, with the proceeds funding a sovereign wealth fund for the American public. The tax is on stock, not profits. Half of every major AI lab, by share count, would transfer to public ownership. Sanders framed it as "giving the public a direct role in determining the future of this technology" and "guaranteeing that the trillions of dollars potentially generated by AI are used to improve the lives of all of us."

**June 5-6, 2026 — President Trump's White House** publicly confirmed, via Air Force One press gaggle and CNBC reporting, that the administration is actively discussing government equity stakes in major AI companies, framed as deals "where the American people can benefit from the success of AI." Trump referenced a 10% Intel stake from August 2025 as the template. Bloomberg reported the discussions include OpenAI. Trump's AI and crypto czar (now former) David Sacks, who now co-chairs the President's Council of Advisors on Science and Technology, publicly said he can "see why Sanders' idea resonates, including with many on the right," but warned it would "accelerate the corporate-government fusion we're already sliding toward." The right and the left are arguing about a shared premise: the government should own a piece of the AI stack.

**April 6, 2026 — OpenAI's own policy paper** "Industrial Policy for the Intelligence Age" proposed a Public Wealth Fund, seeded by AI companies, invested in diversified assets, with every American getting a stake in AI growth. OpenAI's framing is closer to Alaska's Permanent Fund than to Sanders' wealth-redistribution framing, but the mechanism is the same: the public becomes a co-owner of the AI economy by way of government equity.

Three announcements. Three voices. Two parties. One architectural idea. The US government is about to become a co-owner of the frontier AI stack, and the only question is the mechanism — Sanders' 50% tax, Trump's negotiated stakes, OpenAI's seed contribution, or some hybrid that satisfies all three. The policy debate is now about *how*, not *whether*.

This is the most underreported structural shift in the AI industry since the original Microsoft-OpenAI partnership. Let me show you why it is going to change every architecture decision your team makes between now and the IPO window, why it is bipartisan for reasons almost nobody is naming, and what the next eighteen months actually look like under a Public Wealth Fund regime.

Why The Political Convergence Is Real And Not A Coincidence

The conventional reading is that Sanders and Trump disagree on everything, so a shared AI policy idea must be a coincidence or a fluke. The conventional reading is wrong. The convergence is forced by three structural facts that both parties have noticed.

**Fact 1: Frontier AI is concentrated in three to five private companies whose combined equity value is already larger than the GDP of most US states.** OpenAI, Anthropic, xAI, Google DeepMind (inside Alphabet), and Microsoft (via its OpenAI stake and Maia) own essentially all of the world's frontier model capacity. The combined enterprise value of those companies is between $5 trillion and $10 trillion depending on which 2026 round you cite. The equity is held by a few thousand employees, a few dozen venture firms, and the founding tech giants. The political class — both parties — has noticed that the AI revolution is creating the most concentrated private wealth event in human history, and that the public is going to want a cut.

**Fact 2: The IPO window is open and closing in 2026-2027.** OpenAI is widely reported to be filing for IPO this year. Anthropic filed confidentially for IPO in May. SpaceX (parent of xAI) filed for IPO in May. The market window for liquid public trading of frontier AI equity is 2026 through 2028. Any policy that intends to take public stakes in these companies has to land before the IPOs price, because once the shares trade freely, the public has to buy them at market like everyone else. Sanders' 50% tax on stock is timed for the pre-IPO window for exactly this reason. Trump's equity-stake discussions are timed for the same reason. The clock is the political forcing function.

**Fact 3: The 2025 Intel precedent is now a template.** In August 2025, the Trump administration took a 10% equity stake in Intel, structured as a conversion of CHIPS Act grants into equity plus additional cash, for a total government position worth roughly $11 billion at the time. The structure — convert existing public subsidies into preferred equity, take a board observer seat, retain voting restrictions — is now the playbook for how the federal government acquires equity in tech companies it has helped fund. The AI companies have all benefited from federal AI policy support (the EO, the export controls, the DOE compute programs, the implicit tariff protection on Nvidia chips). The 10% Intel template is the precedent. The argument for applying it to OpenAI, Anthropic, and xAI is mechanical: they have received comparable public support, the public is entitled to comparable equity.

Those three facts are not political. They are structural. Both parties notice them. Both parties reach the same conclusion. The mechanism differs. The destination is the same.

What A Public Wealth Fund For AI Actually Does

Let me get specific about what a Public Wealth Fund for AI looks like under each proposed mechanism, because the differences matter and the press has been sloppy.

**The Sanders mechanism (50% stock tax):** The federal government would receive 50% of the equity of every AI company above a threshold (presumably $1B in valuation or compute, with the exact number to be fought over). The equity would be placed in a sovereign wealth fund governed by a board with public-interest representation. The fund's dividends would be distributed to US citizens, either as a universal basic dividend, as supplemental Social Security, as funding for public services, or as a one-time wealth grant. Alaska's Permanent Fund is the model, scaled by 10,000x.

The political advantage: the public directly receives a check. The political disadvantage: every AI company becomes half-owned by the government, which has the effect of socializing the upside and privatizing the downside only if the fund structure is wrong. The financial-market response to a 50% equity seizure would be a 30-50% valuation reset, a freeze in venture funding for AI, and a wave of restructurings to move IP offshore. This is the mechanism that will most likely pass the House and die in the Senate, or pass the Senate and get litigated into a different shape by the courts.

**The Trump mechanism (negotiated equity stakes):** The federal government would negotiate bilateral equity positions in AI companies, in exchange for some combination of federal contracts, accelerated permitting, export-control support, and continued access to the executive-order framework for covered frontier models. The Intel template suggests a 10% position, with a board observer seat and restrictions on the use of proceeds. The government would not be a passive shareholder; it would be a strategic shareholder with policy preferences.

The political advantage: voluntary, negotiable, faster to execute, no constitutional takings issue. The political disadvantage: the public does not directly receive a check, the equity is held by the Treasury or by a controlled entity like the US Investment Accelerator, and the deal-by-deal structure creates enormous opportunities for regulatory capture. The 10% Intel stake is a 10% Intel stake held by the US government. It does not give every American a vote. It gives the executive branch leverage.

**The OpenAI mechanism (seeded Public Wealth Fund):** AI companies would voluntarily contribute a portion of their equity (or its cash equivalent) to a federally chartered wealth fund, governed by a board of public and private representatives, with the fund's returns distributed as a universal dividend. The framing in the OpenAI paper is closer to Norway's Government Pension Fund Global than to Alaska's Permanent Fund: a long-horizon diversified sovereign investor that uses AI returns to fund the broader federal budget or a universal dividend.

The political advantage: cooperative, market-friendly, and consistent with both parties' rhetorical commitments to shared prosperity. The political disadvantage: voluntary contributions do not scale. The fund would be 1% of the AI economy under voluntary contribution, which is roughly the size of the Alaska Permanent Fund and orders of magnitude smaller than what either Sanders or Trump is proposing. OpenAI's proposal is more of a talking-point framework than an actual mechanism.

**The hybrid that is most likely to pass:** A bipartisan compromise that combines a small (5-15%) mandatory equity position for AI companies receiving federal support, negotiated case-by-case under the Trump mechanism, with the equity placed in a federally chartered Public Wealth Fund under the OpenAI model, with distribution rules closer to Alaska's Permanent Fund than to Sanders' universal dividend. The mechanism would land in late 2027, after the IPOs price, structured in a way that does not trigger a market collapse, and governed by a board that gives both parties a seat. The political class wants the win. The financial markets want the certainty. The hybrid is the only outcome that gives both.

I am calling this the **AI Wealth Fund Compromise**, and I think it has a 60% chance of being signed into law by Q4 2027.

Why This Changes Every AI Architecture Decision Between Now And The IPO Window

The policy debate matters, but for engineering teams and AI architects the more immediate consequence is structural. A federal equity position in OpenAI, Anthropic, and xAI changes the calculus on every architecture decision that touches the frontier-model layer. Here is what I mean.

**First, vendor consolidation accelerates.** If the federal government is a 10% equity holder in OpenAI, it has a strong preference for OpenAI to succeed. That preference is going to be reflected in federal AI procurement, in the covered-frontier-model framework, in the export-control regime, and in the CHIPS Act successor programs. OpenAI is going to win a disproportionate share of federal AI business. The same is true for Anthropic (which has a less contentious political profile and a stronger safety positioning that plays well with the Sanders wing) and for xAI (which has the explicit political relationship with the current administration). Google DeepMind and Microsoft are inside larger parents that are also implicated, but the equity structure is more diffuse. The practical effect: if you are building an enterprise AI system today, the political logic of federal co-ownership is going to make OpenAI, Anthropic, and xAI the de facto preferred vendors for any system that touches federal money. Microsoft, Google, AWS, and the smaller labs become the "if you are not federally regulated" tier. That is a real bifurcation in the enterprise AI market, and it is happening in the next eighteen months.

**Second, the open-weights labs become structurally more important.** The labs that the federal government does *not* co-own — Meta (with Llama), Mistral, DeepSeek, Qwen, the smaller open-weights players — become the only frontier-tier models that foreign governments, regulated industries with conflict-of-interest concerns, and enterprises that do not want a federal government watching their AI traffic can use. Expect Meta, Mistral, and the Alibaba Qwen team to become more, not less, important over the next eighteen months. The federal-equity move is a de facto subsidy for the non-co-owned frontier labs. The architecture decision: if your enterprise cannot afford to have federal visibility into your AI traffic (and many cannot, for legal, competitive, or customer reasons), the non-co-owned frontier is where you should be investing your architecture effort.

**Third, the agent stack gets a public-trust layer.** The agent enterprise stack I wrote about last week — authoring, orchestration, governance, observability, deployment — gets a sixth layer almost overnight: **public-trust governance.** The Microsoft Agent Governance Toolkit, Okta's Auth0 for Agents, and the AWS Bedrock AgentCore governance primitives are going to be supplemented by a federally chartered layer that certifies agents for use in federally regulated workflows. The Public Wealth Fund's board (in any of the proposed mechanisms) becomes the natural home for an agent certification regime. If you are building agents that touch federal money, your certification surface is going to be federally curated. That is a real layer of the stack that did not exist last week.

**Fourth, the IPO window becomes the policy window.** The market is going to price the AI IPOs based on the policy risk of the federal equity regime. If a Sanders 50% tax looks likely, the IPOs price at a 30-50% discount to the private valuations. If the Trump negotiated-stake template is the most likely outcome, the discount is 10-20%. If the hybrid compromise is the most likely outcome, the discount is 5-15%. Every architecture decision that assumes a particular vendor's IPO price is implicitly a bet on the policy outcome. The teams that price the policy risk into their vendor selection are going to be the teams that survive the IPO window without a forced restructuring.

**Fifth, the talent market bifurcates.** Engineers who want to work at federally co-owned labs will trade equity upside for stability and federal-grade benefits. Engineers who want to capture the equity upside that the federal stake does not capture will go to the open-weights labs (Meta, Mistral, the Qwen team) or to startups that are not in the federally co-owned tier. The talent flow is going to track the equity structure, and the teams that recognize this will hire accordingly. The architecture decision: your team composition is implicitly a bet on which tier of the equity structure you expect to dominate.

**Sixth, export controls get teeth.** The June 2 EO established the covered-frontier-model framework, with a voluntary 30-day pre-release window for the federal government. The Public Wealth Fund turns that framework from a soft regulatory regime into a hard one. A federal equity holder in a frontier lab has a direct fiduciary interest in not having that lab's models used by US adversaries. Expect the export-control regime around frontier model weights, training infrastructure, and inference deployments to tighten in 2026-2027, with the Public Wealth Fund as the policy lever. The architecture decision: any AI system that depends on cross-border model weights, on training data sourced from US-adversary jurisdictions, or on inference in non-US data centers is going to face new compliance overhead. The teams that have already de-globalized their model supply chain are going to be the teams that win the next round of federal contracts.

What This Means For The Field

The Public Wealth Fund for AI is going to be one of the defining policy structures of the next decade, and it is going to be bipartisan for reasons that are not the usual bipartisan reasons. It is bipartisan because the structural facts — concentrated AI equity, an open IPO window, the Intel precedent — point the same direction from both parties' starting premises. The mechanism will be argued. The destination will not.

The implications for the AI field are not comfortable. A federal government that co-owns the frontier AI labs is a federal government that has a direct interest in those labs' commercial success. The cozy relationship between regulators and the regulated is going to be the central political risk of the late 2020s. The Public Wealth Fund's governance board is going to be the locus of that risk. If the board is captured by the labs, the fund becomes a wealth-transfer mechanism to insiders. If the board is captured by the political class, the fund becomes a slush fund. If the board is structured well — with term limits, conflict-of-interest rules, public disclosure, and a real seat for the workers whose equity is being partially socialized — the fund could be the most consequential economic institution the United States has built since the Federal Reserve.

The architecture decision for the next eighteen months is: build for a world in which the federal government is a co-owner of the frontier AI stack. Vendor selection, agent certification, talent flow, export-control posture, and the IPO-window pricing of your model dependency are all bets on how that co-ownership lands. The teams that price the policy risk correctly will be the teams defining the next decade. The teams that ignore the policy risk will be writing post-mortems about why their carefully built AI system got deprioritized in federal procurement, lost access to the most capable models under export-control tightening, or got out-executed by a federally co-owned competitor with structural advantages they did not anticipate.

The Take

Three announcements in ten days. Sanders' 50% stock tax. Trump's negotiated equity stakes. OpenAI's Public Wealth Fund. Two parties, one destination. The US government is about to become a co-owner of the frontier AI stack, and the AI press is still treating this as three separate stories. It is one story. The mechanism will be argued. The destination will not.

If you are building anything in AI in 2026, you need a clear answer to: how does your architecture account for the federal equity regime, and which tier of the co-owned-vs-non-co-owned frontier are you betting on. The "the policy will sort itself out" answer is a death sentence. The "we have a position in the new architecture" answer is the only one that survives the next eighteen months.

The Public Wealth Fund for AI is going to be one of the most consequential institutions the United States has built in a generation. The fact that it just became bipartisan should be the headline of the week. It is not. That is the kind of gap between what matters and what gets covered that I built this publication to close.

*Sources: Senator Bernie Sanders, "American AI Sovereign Wealth Fund Act," announced June 1, 2026 (NYT op-ed June 1; Fortune coverage June 3). President Trump remarks on Air Force One and White House equity-stake discussions, June 5-6, 2026 (CNBC, Bloomberg, Washington Post, Reuters). David Sacks comments on the corporate-government fusion risk, June 2026. OpenAI, "Industrial Policy for the Intelligence Age: Ideas to Keep People First," April 6, 2026. Intel 10% federal equity stake, August 2025 (Trump administration, CHIPS Act conversion). The 2025 GSA-xAI $0.42-per-agency precedent and the 2025-2026 export-control regime as the policy framework. Next: Apple WWDC 2026 keynote, June 8, 10:00 a.m. PT; the Anthropic Mythos public-release window expected to open in June 2026; the first AI IPO pricing window expected to open Q4 2026.*