@sama Government guarantees should be in the conversation. AI is comparable to electricity / the internet and will be essential for the modern industry to have a 99.99…% uptime. Governments should therefore have some skin in the game to ensure stability.

@sama I ain’t reading all this I’m happy for you Or sorry that happened

@sama Sam just wrote a 1,500-word essay explaining why he absolutely doesn’t want government guarantees… while outlining a $1.4 trillion moonshot that somehow needs them in spirit, if not on paper. It’s performance art at this point. 😂 He’s basically saying: “We don’t want

@sama At this point you can just copy and paste it into gpt5 and ask how he's being disingenuous: This statement is disingenuous because it presents itself as humble, responsible, and market-disciplined — while subtly masking the contradictions between what OpenAI says and what its

@sama Translation: "We don't want a bailout for us. We want a bailout for the hard stuff (infrastructure) so we can keep the high-margin easy stuff (software) for ourselves. Totally different. We love hyping up AGI so please support our research, it’s coming next week.

@sama I know we already have a 6 trillion hole, but listen. If you give me 6 trillion more, we will agi in 2025, I mean 2026, I only need I mean 8 trillion more usd. Then, by 2027 we'll have 10 trillion usd in income all we'll be green by then. If not, the taxpayers will save us.

@sama What are your thoughts on an American "Sovereign AI Wealth Fund"? Like how Norway has one for the oil fields they discovered.

@sama So you don’t want government guarantees… you just want the government to build the infrastructure for you and sell you cheap compute. Got it. That’s not capitalism, that’s state-sponsored monopoly building.

@sama Altman said OpenAI doesn’t want bailouts, but he never mentioned users. He talked about infrastructure, investment, governments, and markets, but not about the people who were forced to watch their AI relationships be deleted mid-sentence. This isn’t a vision, it’s just a

@sama "But we of course could be wrong, and the market—not the government—will deal with it if we are." The market is composed of many U.S. citizens. Also, the U.S. government is now a hedge fund. So yes, if you are wrong, both U.S. citizens and the government will have to deal with

@sama "I would like to clarify a few things." <proceeds to scrawl out 17,000 words clarifying nothing>

@sama When a trillion-dollar AI firm starts talking like it’s just a humble shop in the arena, you’re not hearing transparency. You’re hearing preloaded liability wrapped in mythic-scale PR. This isn’t a clarification. It’s a contingency spell. A public-facing firewall built

@sama Government-owned data centers would probably be as efficient/cost-effective as a government-run space program like SLS and STS. Everything would be contracted out and "legacy" companies, like Boeing, Lockheed Martin, Rocketdyne (not that they are AI developers but legacy

@sama the real tell isn't the $1.4 trillion number, it's that you're building infrastructure before knowing what it's for. "ai that can do scientific discovery" isn't a product category, it's a placeholder for "we think we're close to something fundamental but can't prove it yet."

@sama Your revenue needs to 15x in 8 years to justify $1.4T in spend. That’s the entire thesis. “Feeling good” about growth isn’t a strategy. It’s a prayer. The market sees a circular ecosystem of AI and chip firms recycling capital, not a sustainable business model. Deutsche Bank

@sama tl;dr OpenAI doesn’t want or seek government guarantees for its datacenters. Sam Altman says the market, not taxpayers, should decide winners. He supports governments building their own AI infrastructure, like a national compute reserve, but for public benefit. OpenAI only

@sama AI isn’t a public utility. You chose the burn rate. You chose the capex race. Nah. If you really believe in “open” and “competition,” then compete. If you can’t, someone else will.

@sama $1.4 trillion in 8 years is a massive bet. If compute becomes cheaper faster than expected, OpenAI could overbuild. The timing here is everything.

@sama Every empire justifies its subsidies by calling them ‘strategic infrastructure.’ Today it’s compute instead of corn—but it’s still the same story: privatize the upside, socialize the risk.

@sama All of this is objectively really absurd if you step back and wonder “what if LLMs truly do hit a wall well before general agency?” though

@sama

@sama A forth, crucially important question is: "How do we get enough, reliable energy production fast enough and what can the government do to aid this?" Are you banking on fusion or do you have any other plan? Afaiu energy is one of the largest bottlenecks and needs to be

@sama Inb4 openAI gets a government contract valued at 1.4T to build a data center... And the government rents it out to openAI. Guys, it's not a bailout. That are doing it for the government. Lmao.

@sama Many problems cannot be concealed by lengthy articles.🫠 If AI companies face insufficient computing power, they should carefully manage resources rather than expand indefinitely. OpenAI commits $1.4 trillion over 8 years, yet 2024 revenue is only $20 billion annualized. How will

@sama 1.4 T$ in 8 years implies 175 B$/yr in CAPEX, even assuming 200 B$+ in revenue by 2030 + equity/debt capital, the math feels stretched. IMO, the government will step in eventually. With strategic importance, subsidies are inevitable. Strategic assets rarely stay purely private.

@sama The risk isn’t overbuilding AI - it’s not building enough. Governments shouldn’t pick winners, but a national AI infrastructure? Absolutely.

@sama Tldr Create bubble, print money

@sama I asked GPT5 if you’re just coping. This was the reply:

@sama but when the government builds "public ai", will it actually be public, like open models and infra, or locked behind govt walls?

@sama What benefit would nationalized compute infrastructure provide over privatized ones?

@sama Sam: “We don’t want bailouts” Also Sam: “Please build my trillion-dollar AGI playground, government”

@sama the sovereign compute angle is the most interesting part here. separating 'government owns infrastructure and gets upside' from 'government guarantees private buildout' is a distinction that changes the incentive structure entirely if governments buy computing power at scale and

@sama Bro, you’re creating training data for ChatGPT by writing posts this long.

@sama Sam, can you clarify the distinction between not wanting government guarantees in the first paragraph, while suggesting the government might want a form of equity in their own AI infrastructure….who is the IP progenitor of government AI infrastructure? As a sovereign, the

@sama "Building a strategic national reserve of computing power makes a lot of sense. But this should be for the government’s benefit, not the benefit of private companies." Sounds like The ATOM Project (for open models) https://atomproject.ai/

@sama Have you guys given thought to compute refresh cycle? Meaning will you spend same $$, everytime (every 2 years) Nvidia releases new chip?