The AI Brief

CNBC reported Thursday, confirming Financial Times sourcing, that OpenAI has proposed handing the US government a 5% stake in the company. At OpenAI's March 2026 funding-round valuation of $852 billion, a 5% holding would be worth roughly $42.6 billion. CEO Sam Altman argued that giving the public a financial interest in the company is the best way to share the upside of AI, per two people familiar with the talks.

Altman framed the donated (not sold) stake as the foundation for a "Public Wealth Fund" modeled partly on the Alaska Permanent Fund, with talks running for more than a year with White House officials including Treasury Secretary Scott Bessent and Commerce Secretary Howard Lutnick. The proposed arrangement envisions other US AI companies, such as Anthropic, Google, and Meta, ceding similar stakes through a sovereign wealth fund vehicle. It is not clear whether any of those groups would agree.

Senator Bernie Sanders has proposed a competing approach, the American AI Sovereign Wealth Fund Act, which would impose a one-time 50% stock tax on OpenAI, Anthropic, and xAI , framing the equity question as redistribution rather than voluntary donation. It is not yet clear whether the Trump administration intends to pursue OpenAI's specific offer, whether the stake size will end up smaller than 5%, or how governance and voting rights on the equity would be structured. No final agreement has been reached, and terms remain subject to change. OpenAI and the White House did not immediately confirm the FT's account.

Global AI sales, excluding China, reached $25 billion in Q1 2026, exceeding the industry's estimated $21 billion in depreciation costs tied to investments in data centers and chips for the second consecutive quarter. The research comes from Exponential View and was cited by Bloomberg on June 25. While the milestone suggests AI companies are beginning to cover the cost of their capital spending, the margins are thin. Depreciation charges still consume more than two-thirds of revenue, leaving a small buffer to cover other costs such as power, labor, and financing.

The report states "the economics are holding," but the margin for error is narrow, with more financing risk shifting into capital markets through leases, debt, and equity, especially among so-called neoclouds. The report estimates that by end of 2026, cumulative AI-related capital expenditure by global hyperscale cloud vendors and emerging AI cloud platforms will reach approximately $2 trillion. Passing the depreciation threshold once could be noise; passing it twice is a structural signal, but cumulative historic depreciation from prior capex waves still has not been recovered.

Data from OpenRouter shows the share of tokens requested from Google, OpenAI, and Anthropic models fell to 33% in June 2026 from 72% a year earlier. Azeem Azhar of Exponential View said that reflects power users moving toward cheaper and faster models for simpler tasks. The shift tracks with the proliferation of capable open-weight models: Chinese lab Z.ai's GLM-5.2, a 744B-parameter MIT-licensed mixture-of-experts model, was previously covered in this brief as the strongest open-weight coding model at roughly one-sixth the per-token cost of GPT-5.5.

The OpenRouter data is an operator-level signal rather than a consumer one; it captures developers actively choosing between providers for production workloads, not casual ChatGPT users. The implication for enterprise procurement is that the "default to frontier" heuristic is eroding among technically sophisticated users, replacing it with task-routing logic. Azhar described the shift plainly: "You don't always need a Nobel laureate to extract a number from your receipt to put into an expense spreadsheet." The counter-signal: the same period saw frontier labs capture a larger share of high-complexity agentic workloads, suggesting bifurcation rather than uniform substitution.

OpenAI began a limited preview of GPT-5.6, introducing Sol, Terra, and Luna with stronger reasoning, coding, biology, and cybersecurity performance, a new ultra mode, tougher safety protections, updated pricing, and more predictable prompt caching. According to OpenAI, the company previewed the models to the US government before launch. At the government's request, the first wave of access went to a small group of trusted partners whose participation was shared with the government.

GPT-5.6 is priced per million tokens across three model sizes: Sol at $5 input/$30 output; Terra at $2.50 input/$15 output; and Luna at $1 input/$6 output. Sol introduces a "max" reasoning effort, giving it more time to think through difficult tasks, plus an "ultra" mode that uses subagents to accelerate complex work. OpenAI is also launching GPT-5.6 Sol on Cerebras at up to 750 tokens per second in July, targeting unprecedented inference speed for select customers.

The government-first release structure tracks with the broader pattern from the Fable 5 / Mythos 5 export-control saga: the Trump administration has signaled that frontier cybersecurity-capable models warrant pre-release government review. The equity-stake proposal reported today fits the same logic: Washington is positioning itself as a governance layer over frontier AI releases, using access approvals and ownership stakes as levers rather than formal regulation.

Over the ten weeks of the export-control episode, Anthropic worked closely with the US government as it developed the approach reflected in the June 2 Executive Order on Promoting Advanced AI Innovation and Security. Its engagement spanned the Office of the National Cyber Director, the Office of Science and Technology Policy, the Department of the Treasury, the Department of Commerce including CAISI, and relevant national security agencies.

Anthropic committed to continuing that work, building on nearly two years of pre-existing collaborations with US government partners on pre-deployment testing and evaluation. The formal commitments include providing designated government partners with expanded early access to models that materially advance the capability frontier in areas relevant to national security. In a letter to Anthropic, Commerce Secretary Howard Lutnick confirmed the company would no longer require an export licence, as it had agreed to proactively detect and address security risks, to work with the government on standards for upcoming models, and to inform the government of malicious activity.

The restoration commitments establish a structural precedent that goes beyond Fable 5. They make Anthropic the first commercial frontier lab to operate under a publicly documented, government-mandated pre-release review framework for future capability-advancing models. The GPT-5.6 government-preview launch and the equity-stake proposal both show OpenAI moving voluntarily toward a similar posture. Whether this converges into a formal inter-agency evaluation regime or remains ad hoc is the open question for any operator dependent on predictable model release timelines.

Disclosure: Claude, which generates this brief, is built by Anthropic.