Burn It Down

Sam Altman just published a 13-page plan to save capitalism from AI. He's the one building the AI. His company won't be profitable until 2030. And this was somehow not the wildest thing that happened at OpenAI this week.

Part of the ongoing Big Tech's War on Users series. If you haven't read They're Racing to Stay Ahead of the Fuse — this one lands harder with that context. But it stands alone.

I was waiting on a database download when Sam Altman decided to save capitalism.

Feed refresh. There it was. A 13-page document titled "Industrial Policy for the Intelligence Age: Ideas to Keep People First." Robot taxes. A public wealth fund seeded by AI companies. A 32-hour workweek pilot. Automatic safety net triggers when AI-driven displacement metrics hit preset thresholds. Sam Altman, CEO of the most expensive private company in Silicon Valley history, publishing a blueprint for how the government should redistribute the wealth generated by the technology he is racing to ship before his own math catches up with him.

I've spent the last two weeks writing about burning fuses. I did not expect the next one to introduce itself on a Monday afternoon while a database downloaded in the background.

But before we get into what this document actually is — let me tell you about the week it landed in.

The Week In Full

March 31st. OpenAI closes a $122 billion funding round at an $852 billion valuation. Largest private funding round in Silicon Valley history. The press release is triumphant.

Buried: $35 billion of Amazon's $50 billion is contingent on OpenAI going public or achieving AGI by end of 2028. Not in the door yet. One analyst called it "a put option disguised as a capital commitment." Nvidia and SoftBank's combined $60 billion arrives in tranches. Also not all in the door yet. I covered the full breakdown four days ago.

Also buried: CFO Sarah Friar — the woman who just closed that round — has been privately telling colleagues she doesn't think OpenAI will be ready to IPO in 2026. She has concerns about whether the spending commitments are supportable given slowing revenue growth. Altman's response, per The Information: stop inviting her to financial planning meetings. Including a meeting with a major investor about server procurement. The CFO who closed the largest private funding round in history now reports to the head of applications instead of the CEO and is being frozen out of conversations about the company's financial plans.

Read that again.

April 2nd. OpenAI acquires TBPN — the Technology Business Programming Network, a daily Silicon Valley talk show with 58,000 YouTube subscribers — for reportedly low hundreds of millions of dollars. Altman calls it his favorite tech show. The deal is announced by Fidji Simo, head of AGI deployment. TBPN will report to Chris Lehane — OpenAI's chief political operative. Its advertising revenue winds down under the new structure.

April 3rd. Fidji Simo — the executive who announced the TBPN acquisition two days earlier — goes on medical leave for a worsening neuroimmune condition. The CMO, who has been managing a late-stage cancer diagnosis, steps down to focus on her recovery. The COO is shuffled to special projects reporting directly to Altman. Leadership is redistributed across four executives. The company is preparing for an IPO.

Also April 3rd: Iran's Islamic Revolutionary Guard Corps releases a video with satellite imagery of the $30 billion Stargate AI datacenter under construction in Abu Dhabi — the flagship of OpenAI's global compute infrastructure — and threatens its "complete and utter annihilation." This is not a hypothetical. AWS facilities in UAE and Bahrain have already been hit by strikes. Oracle's Dubai datacenter was reportedly targeted. The video message: "Nothing stays hidden from our sight, though hidden by Google."

April 4th-5th. The New Yorker publishes a lengthy investigation by Ronan Farrow and Andrew Marantz into Sam Altman. Eighteen months of reporting. Over a hundred interviews. Two previously undisclosed documents. We'll get to what's in them.

April 6th. Today. Sam Altman publishes a 13-page document about saving civilization. Axios exclusive. New Deal comparisons. Half-hour interview.

I cannot make this up. I was waiting on a database download.

Some Of This Sounds Good On Paper

I want to be straight about something before getting into the rest of it, because the voice of this series isn't reflexive cynicism and this document doesn't deserve to be dismissed wholesale.

A public wealth fund that gives every American a stake in AI-driven economic growth? Worth discussing seriously. Automatic unemployment triggers that kick in when displacement metrics cross preset thresholds and phase out when conditions stabilize? That's actually thoughtful policy design — responsive rather than permanent, tied to real conditions rather than political cycles. Portability of benefits not tied to a single employer? Long overdue regardless of AI. The 32-hour workweek framed as an efficiency dividend — if AI genuinely creates productivity gains, the argument that workers should capture some of that as time rather than watching it flow upward is legitimate.

These are real ideas with real intellectual history. Some serious people have been making versions of these arguments for years. The fact that OpenAI is now publishing them doesn't make them wrong.

It does raise a question.

With what money?

(Worth noting: the administration that would presumably implement any of this recently cited military spending priorities as the reason it couldn't afford a fairly long list of domestic programs. That's not a political observation. That's just the room the 13-page document walked into.)

The Math Has Not Changed

Four days ago I ran these numbers. They have not improved since Monday.

OpenAI is burning $25 billion this year. $57 billion in 2027. $665 billion through 2030. They are generating $2 billion in revenue per month — which sounds like a lot until you note that adjusted gross margins fell to 33% last year because inference costs quadrupled. The revenue curve is going up. The cost curve is going up faster. That gap is the structural problem no funding round fixes.

And now we know the CFO — the person legally responsible for the accuracy of the company's financial statements — has been frozen out of financial planning meetings for raising concerns about whether the spending commitments are supportable. The woman who closed the largest private funding round in history isn't being included in conversations about how to spend it. Because she asked if the math works.

This is the company proposing to seed a public wealth fund.

They announced they'd donate a kidney. They're on dialysis. And they just iced the doctor who asked about the bloodwork.

The more interesting question isn't whether the proposals are good. It's whether OpenAI ever expects to be in a position to fund them — or whether the document is written for a future that requires them to still be standing when it arrives. If the competition goes broke first, if the IPO hits before the quarterly numbers become undeniable to everyone holding the bag, if the government becomes a stakeholder in your survival before the regulations get written — then maybe you don't actually need to afford it. You just need to still be in the room.

Jensen Already Ran This Play

Three weeks ago I wrote about Jensen Huang's token compensation proposal — paying engineers half their salary in compute credits they can only spend on Nvidia infrastructure. Company scrip. Spends at the company store. Workers spend it right back into the demand metrics. Circular all the way down.

You load sixteen tokens, what do you get? Another day older and deeper in debt.

OpenAI's wealth fund is that logic at national scale.

Seed the fund with AI company contributions → government becomes a stakeholder in AI success → government can't regulate too aggressively without hurting its own investment → companies keep building → companies generate the growth the fund depends on → circular all the way down.

Jensen's company town had workers. Altman's company town has a country.

And the 32-hour workweek framed as an efficiency dividend? Jensen already proposed the efficiency dividend. He called it tokens. The underlying logic is identical — AI creates productivity gains, here's your share. Jensen's version keeps it in-house as compute credits. Altman's version nominally gives it back as time. Both are mechanisms for deciding where the efficiency gain lands before it distributes on its own. The difference is Altman's version requires an act of Congress.

Which means it probably won't happen. Which might be exactly the point.

Phase 3, Congressional Edition

The series opener laid out the playbook: they built things we genuinely needed, got us hooked, then figured out how to monetize the removal. Phase 1: collect the trust. Phase 2: consolidate the position. Phase 3: profit.

The GitHub post documented Microsoft running it — Nadella's 2018 promise that GitHub would "retain its developer-first ethos, operate independently, and remain an open platform." Seven years of trust accumulation. Then the independence ends quietly in a reorganization, and individual developer code routes into the Microsoft AI training pipeline by default. Nadella said judge them by their actions. He was right to invite that.

OpenAI's document is the most sophisticated Phase 3 this series has covered. They're not just extracting value — they're proposing to institutionalize the extraction as government policy. Make the safety net structurally dependent on the companies causing the disruption. Embed the government as a stakeholder before the regulations get written. That's not Phase 3. That's Phase 3 with a bibliography, a New Deal comparison, and a Ronan Farrow investigation dropping the same morning.

Altman has his own version of Nadella's formulation. Asked why people should trust him, he told Axios: "I think almost everybody involved in our industry feels the gravity of what we're doing. We all take that responsibility very seriously. We feel that way every day."

Okay. Judge them by their actions. Still good advice. Especially today.

The Character Arc

Here's where the New Yorker piece matters — not because it's an opinion column, but because of who wrote it, what they found, and how long it took.

Ronan Farrow and Andrew Marantz spent 18 months on this. They interviewed more than a hundred people. They obtained two previously undisclosed documents. It published this morning — the same morning Altman asked Congress to trust him to redesign the social contract for the intelligence age.

OpenAI was founded as a nonprofit. The premise: AGI was too consequential to be owned by anyone — it needed to exist outside the profit motive, governed for the benefit of humanity rather than shareholders. That's Altman's babyface run. The thing that made serious people believe the intent was genuine. Some very smart, very careful people signed on to that premise.

The New Yorker piece is what some of those people think now.

The first document Farrow obtained: a roughly 70-page confidential memo compiled by Ilya Sutskever — the former Chief Scientist, the man who led the board coup that fired Altman in 2023. The memo contains Slack messages, HR documents, and screenshots Sutskever allegedly took on a personal phone specifically to avoid company device monitoring. The first item on the list of concerns reads: "Sam exhibits a consistent pattern..."

The first word of that pattern: "Deception."

The second document: over 200 pages of internal notes by Dario Amodei — the former OpenAI executive who left to found Anthropic — titled "My Experience at OpenAI." Subtitle: "Private, Do Not Share." The central finding, stated plainly: "The problem with OpenAI is Sam himself." The notes include an allegation that Altman personally denied contractual terms to Microsoft executives face-to-face while Amodei read the terms out verbatim.

Microsoft's own executives, speaking to Farrow: "He distorts, twists, renegotiates, and violates agreements." One put the outer bound of the risk this way — there is "a small but real possibility he will ultimately be remembered like Ponzi scheme perpetrators."

A former board member's full assessment, on record: "He has two traits that are almost never seen in the same person. The first is a strong desire to please people, to be liked in any given interaction. The second is almost a sociopathic lack of concern for the consequences that may come from deceiving someone."

That board fired him in 2023 for failing to be "consistently candid in his communications." He engineered his way back. The board was replaced. The detractors left — several to found or join competing companies, apparently carrying detailed notes. The independent investigation promised on his return produced no written report. WilmerHale — the firm that investigated Enron and Tyco — gave only an oral briefing to two new directors. Insiders described it as seeming "aimed at limiting transparency."

Last year the nonprofit conversion completed. The board no longer has meaningful control. Critics got "we hear you" language and a restructuring that addressed the optics more thoroughly than the underlying tension.

This is the turn. Not sudden — turns rarely are when you're paying attention — but completed and now documented at length by a Pulitzer Prize winner who spent a year and a half on it.

"Private, Do Not Share" — the subtitle on Amodei's notes — lands differently after Do Not Blow Your Cover. The people who knew him best built their own architecture for what to hide and from whom. They took screenshots on personal phones. They wrote 200 pages they never intended to publish. They left and started over rather than stay in the room.

The New Yorker's title: "Sam Altman May Control Our Future — Can He Be Trusted?"

He published his answer this morning. Eighteen months of reporting published its answer at the same time.

Gary Marcus, watching both land simultaneously: "when he is in trouble, he often tries to change the narrative with hype."

In professional wrestling there's a character who engineers the chaos, positions himself as the only one with the vision to lead you through it, and needs you to believe — genuinely believe — that he's the savior, not the cause. The gimmick only works if the sincerity feels real enough that you're never quite sure. It only works if there was a babyface run to turn from.

The nonprofit founding was the babyface run. Sutskever's phone screenshots are the receipts.

What's Actually In It For Him

Follow the incentives, because there are more layers here than just the IPO.

The IPO is the obvious one. "Responsible statesman who warned Congress and proposed frameworks" is worth billions in retail investor enthusiasm from people who won't read the burn rate footnotes. They specifically arranged ARK ETF access in the last round. Retail investors buy stories more than institutions do. This is a very good story.

The regulatory moat may be more valuable long-term. If OpenAI proposes the compliance framework, they set it at the height they can clear. A well-funded startup reading this document now has to model a world where robot taxes are real, where compliance overhead scales with AI deployment, where the wealth fund contribution comes out of operating budget. Even a 20% chance of that world materially changes fundraising conversations and runway math for every competitor. OpenAI doesn't need Congress to act. They just need the uncertainty to hang in the air long enough for the field to thin on its own.

The document is called "initial ideas" and "starting point for debate" for a reason. Not a bill. A conversation starter that reshapes as the conversation develops. Costless to propose. Expensive for smaller players to defend against.

The legal shield is pre-filed. If something catastrophic happens — the cyberattack Altman says is "totally possible this year," the novel pathogen he says is "no longer theoretical" — and you're on record having warned Congress and proposed frameworks, your personal culpability looks radically different than if you were silent. "We told them" has real value when the thing you predicted arrives. Especially when the Farrow investigation is already in the archive.

The nonprofit conversion is still under AG scrutiny. Hard to prosecute the guy Congress is currently consulting as a responsible statesman.

And if it all still implodes — if the burn rate wins, the math becomes undeniable, the CFO's concerns prove correct — Altman the Visionary Architect of the Intelligence Age has a much softer landing than Altman the guy who ran the biggest cash burn in tech history while freezing out the person whose job it was to say stop. He's building the exit parachute into the narrative right now. The document exists in a future where OpenAI doesn't have to.

The Real Architecture

On April 2nd, three days before publishing his civilizational manifesto, Altman's company bought TBPN — the daily Silicon Valley talk show where tech founders and CEOs come to shape the industry's self-image — for reportedly low hundreds of millions of dollars.

The show has 58,000 YouTube subscribers. It generated $5 million in ad revenue last year. Its advertising revenue winds down under the new structure.

It reports to Chris Lehane. OpenAI's chief political operative.

The Information's framing: Elon has X. Now Altman has TBPN.

The "editorial independence" guarantee is precisely as meaningful as Nadella's 2018 GitHub promise. The hosts say they can say whatever they want because they're live. They just can't pay their own bills without OpenAI anymore. The financial independence went with the ad revenue. What remains is the credibility of the brand, now owned by the subject it covers, parked under the political operation.

Do Not Blow Your Cover was about a different kind of architecture — the code kind. The one-way doors. Instructions hardcoded below the context window so nothing could drop them. The system designed to manage what you see, staying undercover regardless of what else the context does.

TBPN is that at narrative scale. You don't suppress the conversation. You own the room where it happens and call it editorial independence.

The policy paper tells Congress what Altman thinks. TBPN shapes what Silicon Valley thinks Congress should think. And the undercover architecture manages what the developers building on all of it are allowed to see. Three layers. One direction of travel.

The Fuse Is Still Burning

Four fuses, I wrote last week. The burn rate, the safety gap, the resource spiral, the geopolitical wildcard. I described the fourth fuse as the one they can't control — the one where no funding round, no policy paper, no narrative architecture reaches.

It's no longer a warning.

The $30 billion Stargate datacenter in Abu Dhabi — the flagship of the compute infrastructure this entire plan depends on — has been named by Iran's Revolutionary Guard Corps for "complete and utter annihilation." They released a video with satellite imagery to prove they know exactly where it is. AWS and Oracle facilities in the region have already been hit. The message: "Nothing stays hidden from our sight, though hidden by Google."

The compute infrastructure underpinning the $665 billion plan is in a geopolitical crossfire that no 13-page document addresses. You can propose a wealth fund. You cannot negotiate with a missile.

This document is not a new fuse. It's the existing fuses — all four of them — getting a press cycle and a policy framework while they keep burning. The burn rate doesn't change because a manifesto exists. The inference costs don't drop. The agents still have context compaction problems. The DRAM shortage doesn't normalize. The CFO being frozen out of meetings doesn't stop being right just because she's no longer in the room. The Stargate datacenter doesn't move.

What the document does is buy time — in the court of public opinion, in the halls of Congress, in the narrative running alongside the quarterly numbers — by positioning the company as the architect of the solution rather than the accelerant on the problem.

Three days before the manifesto: buy the room where the narrative gets made. Same morning as the manifesto: Ronan Farrow publishes 18 months of receipts. Same morning as the manifesto: the CFO's concerns become public record. Same week as the manifesto: the datacenter is in a missile threat video.

And if you know your Seth Rollins — the screaming, the pyro, the whole entrance built around a single phrase — you already know that Burn It Down was never a warning.

It's what he does when he walks in.

Find me on Mastodon at @ppb1701@ppb.social. The thread, as always, keeps not running out.

Part of the ongoing TheranasAI series, a sub-series of Big Tech's War on Users.

Read the terms. They're more honest than the marketing.