Sanders argues that OpenAI, Anthropic, Google DeepMind, xAI, and Meta's FAIR were all built on decades of publicly-funded research from DARPA, NSF, and the national labs, trained on a public-data commons, and continue to be subsidized through CHIPS Act allocations and federal compute contracts. His core claim is symmetry: if the public underwrote the upside risk, the public should hold the upside via a 50% non-voting equity stake whose dividends flow to a public trust.
The editorial points out that Sanders explicitly invokes Norway's oil fund, Singapore's Temasek, and Alaska's Permanent Fund — non-voting shares with no operational control. Read through that lens, the proposal is a passive financial instrument that captures public upside without state management, and the HN reaction split tracked exactly that framing question rather than left-vs-right ideology.
By submitting the op-ed to HN where it drew 130 points and 151 comments, droidjj surfaced a proposal whose comment thread split largely between people reading it as nationalization and people reading it as a Norway-style sovereign fund. The submission framing treats it as a serious policy idea worth technical-community debate, not a fringe socialist talking point.
The editorial argues the proposal is DOA in this Congress but is the third major 'AI as utility' trial balloon in 18 months, after the EU AI Act's foundation-model tier and the UK's shelved AISI compute-disclosure rules. The framing — treating frontier labs like New Deal electric utilities whose outputs are too foundational for normal capital-formation rules — is winning faster than the industry seems to notice, having moved further in two years than social-media regulation did in a decade.
On the Senate website, Bernie Sanders published an op-ed titled *The Public Should Own Half of the Big A.I. Companies* — a proposal that the federal government take a 50% non-voting equity stake in any AI company valued above $100 billion. The argument: OpenAI, Anthropic, Google DeepMind, xAI, and Meta's FAIR have all been built on decades of publicly-funded research (DARPA, NSF, the national labs), trained on data scraped from a public commons, and increasingly subsidized through CHIPS Act allocations, federal compute contracts, and tax credits on data-center construction. Sanders' framing: if the public underwrote the upside risk, the public should hold the upside.
The op-ed hit Hacker News at 130 points — modest engagement, but the comment thread is the real story. The split isn't left-vs-right. It's between people who read "public ownership" as nationalization (and recoil) and people who read it as a sovereign-wealth-fund model (Norway's oil fund, Singapore's Temasek, Alaska's Permanent Fund) and find it boringly reasonable. Sanders explicitly invokes the latter: non-voting shares, dividends flow to a public trust, no operational control.
The proposal is almost certainly DOA in the current Congress. But it's the third major "AI as utility" trial balloon in 18 months, after the EU AI Act's foundation-model tier and the UK's now-shelved AISI compute-disclosure rules. The Overton window on what counts as a normal way to regulate frontier labs has moved further in two years than it did for social media in a decade.
The interesting question isn't whether this passes. It's what assumption it encodes. Sanders is treating frontier AI labs the way the New Deal treated electric utilities: a category of private company whose outputs are so foundational to the rest of the economy that the normal rules of capital formation don't apply. Once that framing wins — and it's winning faster than the industry seems to notice — the specific mechanism (equity, windfall tax, golden share, mandatory licensing) becomes a detail.
Compare the approaches actually on the table:
- EU AI Act: tiered obligations based on training compute (10^25 FLOPs threshold). Regulatory, not financial. Already in force for general-purpose models. - UK (pre-election): voluntary safety commitments + AISI pre-deployment evals. Now largely abandoned under the new government's growth agenda. - China: de facto state direction through licensing, algorithm registration, and the "national team" model (Baidu, Alibaba, ByteDance, Tencent all coordinating with MIIT). - US (Sanders proposal): equity stake, public dividend, no operational control.
The Sanders model is the only one that doesn't try to regulate model behavior — it just claims a share of whatever the labs end up producing, good or bad. That's either elegant or evasive depending on your priors. Elegant because it sidesteps the impossible question of what "safe AI" means and just aligns financial incentives. Evasive because it does nothing about the actual concentration of compute, talent, and decision-making in five companies.
The HN commentariat's strongest objection isn't ideological — it's mechanical. A 50% non-voting stake at a $100B threshold creates a sharp cliff: labs would have every incentive to stay at $99B, spin out subsidiaries, or reincorporate offshore. The history of corporate-tax thresholds is the history of companies engineering around them. Anthropic's recent structure (PBC + Long-Term Benefit Trust) and OpenAI's capped-profit-then-uncapped restructuring already show how creative the cap table gets when the stakes are this large.
There's also the small matter that two of the five labs are not US companies in any clean sense. DeepMind is a UK subsidiary of a US company. Mistral is French. Anthropic took a $4B investment from Amazon and a $2B from Google. The notion of clean national ownership maps poorly onto entities whose compute lives in three countries and whose investors span four.
If you're a senior engineer reading this thinking "interesting policy debate, but not my problem" — the second-order effects are very much your problem.
First, the regulatory ratchet only goes one way. Even if Sanders' specific proposal dies, the framing that frontier labs are quasi-public infrastructure is now mainstream enough to appear in a US senator's op-ed without being dismissed as fringe. Expect the next administration — either party — to extract *something* from the labs: a windfall tax, mandatory model access for federal agencies, compute disclosure requirements, or a golden share. Price that into your build-vs-buy decisions for anything dependent on frontier model pricing past 2027.
Second, the labs will respond by entrenching. The fastest way to make a 50% public stake politically impossible is to become so embedded in critical infrastructure that the threat of disruption is unthinkable. Watch for accelerated push into federal contracts (Anthropic's recent Palantir partnership, OpenAI's DoD work), healthcare integrations, and education deals. The strategic logic is the same one electric utilities and telecoms followed in the 1930s: become the load-bearing wall before anyone gets serious about ownership.
Third, open-weights becomes a hedge, not just a philosophy. Meta's Llama strategy, Mistral's open releases, and the Chinese open-weights wave (Qwen, DeepSeek, Yi) all look different in a world where closed frontier labs are politically vulnerable. If you've been deferring serious investment in your local-inference stack because the closed APIs are good enough, the policy environment is now an additional reason to revisit that calculus.
The specific proposal won't pass. But the assumption underneath it — that the social contract between frontier AI labs and the public is unfinished business — is now load-bearing in the political conversation, and it won't unload itself. The labs that survive the next decade won't be the ones with the best benchmarks; they'll be the ones that figured out a politically durable answer to "who owns the upside" before someone else answered it for them. Sanders just gave the industry a free preview of what one possible answer looks like. The smart move is to read it as a forecast, not a manifesto.
Why half of AI and not half of Walmart & Exxon & Apple?Government spending is already ~40% of GDP.And what do we get with this half?A sovereign wealth fund? That seems like a great tool for a certain corrupt politician to use as a carrot to make CEO's bend to his/her whims.What ben
I love it, this is exactly the kind of thing government is meant to do, bring externalities to profit under control. There is something that's been stolen from all of us, collectively, and certain authors and artists, specifically, the creative soul we all pour into our expression, here, there,
Strictly speaking, the big A.I. companies _want_ the public to own half of them. Passively. In index ETFs in their 401(k)s and other retirement portfolios. That way the get all the money without any of the actual influence.
Instead of "The Public.." read it as "The Government Should Own Half of the Big A.I. Companies" because that's what it really means. It doesn't sound all that great now does it? Imagine what a red/blue administration would do with complete control of those AI capab
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.
> A.I. is built on our collective intelligence: our books, songs, artwork, journalism, computer code, scientific research, videos, conversations, images and ideas spanning generationsI know many here would scoff at nationalizing a private company, but AI is a usurpation of human knowledge and qui