AI Doom vs Boom, EA Cult Returns, BBB Upside, US Steel and Golden Votes

Summary

• Sacks’s central claim: AI doomerism is a funded political project amplifying legitimate concerns — a ~$1.6B network seeded by Dustin Moskovitz’s Open Philanthropy, tied to Anthropic by marriage (money-man Holden Karnofsky is married to Dario’s sister, an Anthropic co-founder) and personnel (top Biden AI staffer Tarun Chhabra and AI Safety Institute founder Elizabeth Kelly now work there). The agenda is “global compute governance,” and its adherents “need to be loomered, quite frankly.”
• Sacks’s risk ranking is the tradeable frame: China winning the AI race “might be, like, thirty percent” versus a “much lower” x-risk — and the only dystopia with tangible evidence is government using AI to control citizens, which he argues the Biden EO and Diffusion Rule were already building toward before Trump’s win moved the trajectory.
• Chamath’s tell: Anthropic’s safety warnings “tend to be pretty coincidental with key fundraising moments” — not nefarious, “it’s smart,” since anyone who isn’t sitting on Meta/Google “money gushers” needs an angle. His threshold question: “should you fear government overregulation or should you fear autocomplete?” Right now it’s “an exceptionally good autocomplete” — though “Claude kicks ass.”
• Friedberg’s counter to job doom is a capital-returns argument: one engineer outputting “20, 50 times as much software” means higher ROI on deployed capital → more capital deployed → more jobs, plus a deflation dividend — the $8 latte dropping to $2, the workweek compressing below 30 hours. Sacks adds Satya’s test: 10–20% white-collar job loss implies 10%/yr GDP growth — “do you actually believe that’s gonna happen?”
• Displacement is real at the entry level now: “new grads were our autocomplete,” per Chamath, and at his 30-person 8090 AI-native young workers thrive while L7s from Google/Amazon/Meta “just don’t thrive.” The unresolved fight: Jason ties Microsoft’s 6,000 layoffs to AI eating management; Sacks calls it “total confirmation bias… we’re still at the chatbot stage.”
• On the Big Beautiful Bill, Sacks issues corrections: reconciliation legally can’t contain DOGE’s discretionary cuts (Byrd rules), and the CBO scores the sunset removal as a spending increase when against a current-year baseline the bill cuts mandatory spending. Friedberg’s numbers: Medicaid’s $70B/yr cut still leaves it ~40% above 2019; the bill passed by one vote — Chamath says he doesn’t put the blame on the White House.
• The path out is GDP, and the bond market’s long end is the judge — Chamath calls it the second-biggest actor after Trump, and “they’ve steepened the curve.” He bets the Q2 GDP print comes in “above three” versus the CBO’s 1.7% assumption; the binding constraint is energy, where the US “literally consume[s] every single bit of energy that we make,” SMRs are 2035+, and renewables-plus-storage are the fastest supply.
• Trump cleared Nippon’s $14.9B US Steel deal as a “partnership” with partial ownership controlled by the USA; Chamath wants the model templated across five critical industries (pharma precursors, AI, lithography, EV deposition and chip-making, batteries, rare earths) à la Hu Jintao’s 2003 national champions. Friedberg dissents — “keep the government out of the markets” — and would rather invest Social Security’s $4.5T like Australia’s supers before its 2032 functional bankruptcy.

Deep dive

1. Sacks maps the “AI existential risk industrial complex”

• Sacks grants the premise before flipping it: AI risks are legitimate, but a claim like 50% of white-collar jobs lost within two years is “a level of specificity that I think is just unknowable” — headline-grabbing with a pattern behind it. Three years ago Anthropic’s bioweapon demo drove Rishi Sunak’s interest and the first AI Safety Summit at Bletchley Park — “it turns out that that particular output was discredited.”
• The money map: roughly $1.6B of EA funding across an “inflated ecosystem” — same names, acronyms, logos and talking points, one source: Dustin Moskovitz’s Open Philanthropy. Holden Karnofsky, who doled out the money, is married to Dario’s sister, herself an Anthropic co-founder. “These are not coincidences.”
• The personnel pipeline seals it for Sacks: Biden’s most powerful AI staffer Tarun Chhabra now works at Anthropic, as does AI Safety Institute founding director Elizabeth Kelly; Reed Hastings — who tried to drive Thiel off Facebook’s board in 2016 — just joined Anthropic’s board. Three nodes: the EA movement (Sam Bankman-Fried its most notable member), Biden staffers, and Anthropic — “a very tightly wound network.”

2. The agenda is global compute governance — and the doomers are rebranding as China hawks

• What the network wants, per Sacks: regulation of computational resources and GPU access, supranational agreements, ethics built into policy — which maps directly onto the Biden record: the 100-page AI executive order with DEI requirements (“it led to woke AI… black George Washington”), the 200-page Diffusion Rule’s global GPU licensing, the AI Safety Institute.
• His risk arithmetic: x-risk is real but unpriceable, while China winning “might be, like, thirty percent.” The only dystopian risk with tangible evidence is Orwellian — government-controlled AI: “prior to Donald Trump winning the election, we were on a path of global compute governance, where two or three big AI companies were gonna be anointed as the winners,” with woke values as the quid pro quo. “We could very easily be moved back onto that trajectory.”
• The kicker he wants Republicans to hear: none of these regulations solves the actual problem — “they don’t solve for alignment. They don’t solve for the kill switch.” And the same ex-Biden agitators whose Gulf chip restrictions were “pushing them into the arms of China” now market themselves as China hawks because doomerism “is just not gonna get much purchase” under Trump. “They need to be loomered, quite frankly.”

3. Chamath’s tell: safety warnings track fundraises

• His opener: “these safety warnings tend to be pretty coincidental with key fundraising moments in Anthropic’s journey… put that into an LLM and try to figure out if what I just said was true… you find it’s relatively accurate.” Not nefarious — “It’s smart”: if you aren’t Meta or Google sitting on “money gushers,” you have to find an angle.
• The ideology is self-reinforcing through liquidity: funding rounds where “a large portion goes right back out the door via secondaries” mean lots of people getting rich holding this worldview — “it’s gonna cement that worldview, and then they are going to propagate it even more aggressively.”
• His threshold question for policymakers: “should you fear government overregulation or should you fear autocomplete?… It may get so good that it’s an AGI, but right now it’s an exceptionally good autocomplete.” Even so, no product bashing: “Claude kicks ass.”
• Jason’s synthesis — all three views can be true at once: EA true believers with “God complexes,” economic interests using those “useful idiots” to become one of the anointed winners, and a political ideology riding along to “secure the bag.”

4. Friedberg’s counter: AI raises the return on capital, and returns attract capital

• His reframe of the firing question: AI-written software doesn’t mean cutting 80% of engineers — one engineer outputs “20, 50 times as much software,” so the ROI on that salary explodes. “If you have a chance to make 20 times on your money, you’re gonna deploy a lot more capital. And this is the story of technology going back to… the caveman.” Hence more venture capital into startups, not less: “the premise that AI destroys jobs is wrong.”
• Jason’s ground-level confirmation: startups now regularly hit $1M revenue per employee, versus the $250K–$500K benchmarks of prior cycles.
• The second engine is deflation: the Industrial Revolution created a consumer market that didn’t exist; AI’s equivalent is costs halving — “suddenly, you only need to work 30 hours a week.” Best specimen: robotic food prep. “It’s eight dollars for a latte. This is crazy… What if that dropped down to two bucks?” Vision-action models can be trained and installed “in a matter of weeks” where it used to take a whole startup years.
• Friedberg’s power-vacuum warning doubles as his explanation of doomerism: every major transition invites people claiming “I am smarter, more educated… more moral… therefore, I should be in a position to define how the new system should operate.” “Fear is a great way of getting into power.”

5. Entry-level hiring is shrinking — the fight is over what falls next

• Chamath’s tweet unpacked: “If a GPT is a glorified autocomplete, how did we used to do glorified autocomplete in the past? It was with new grads. New grads were our autocomplete.” Incumbent employees rise without “new grad grist for the mill”; OPEX shrinks as a percentage of revenue and revenue-per-employee climbs. His prescription for grads: steep yourself in the tools, “go to younger companies or start a company. I think that’s the only solution.”
• His hiring evidence from 8090 (~30 people): senior mentors plus “an overwhelming corpus of young, very talented people who are AI native.” The L7s from Google, Amazon and Meta with high salary demands “just don’t thrive… they push back on the tools.” His own precedent: he was the Luddite annoyed that Facebook used PHP and Python — abstraction “grew the top of the funnel of the number of developers by 10X,” and this is the same leap again.
• The unresolved fight: Jason ties Microsoft’s 6,000 layoffs (3% of workforce, record profits) to AI eliminating management, citing Sergey Brin’s demo of feeding internal Slack to an LLM and asking “who are the underrated people… who deserve a raise” — it got the right answer. Sacks: “total confirmation bias… What is the AI agent that’s doing management right now? This theory doesn’t even make sense… We’re still at the chatbot stage.” Chamath sides with Sacks on management.
• Sacks’s structural doubt: fully eliminating a multifaceted job (a salesperson negotiates, implements, gives product feedback) is very hard, and coding assistants may be a leveler that makes junior programmers 5–10X better — “I don’t think we know how this cuts yet.” Regulating pre-emptively is “a total power grab” before the risk has even manifested.

6. Is there an AI race? Kumbaya versus the iron cage

• Friedberg’s provocation: “who won the Industrial Revolution?… who won the internet race?” — value accrued everywhere. AI is a substrate under “every interaction humans have,” a continuous process with no finish line, and abundance “means more peace” in a less resource-constrained world.
• Sacks partially agrees — it’s an infinite game — but insists a race exists anyway: per Mearsheimer’s All-In Summit phrase, the US and China are in an “iron cage.” Nuclear stabilized because the technology “hit an asymptote,” enabling arms control; AI is “potentially on some sort of exponential curve” where “even six months ahead could result in a major advantage” — so he doubts an agreement to slow down is possible. Armies of the future are AI-powered drones and robots.
• His definition of winning, pressed by Friedberg on what losing even means: “the whole world consolidates around the American tech stack… If we have, like, 80 to 90% market share, that’s winning. If they have 80% market share, then we’re in big trouble.” The template is Huawei leapfrogging on 5G — “they weren’t concerned about diffusion.” And when prosperity and balance of power conflict, “power is ultimately privileged.”

7. BBB cleanup: what reconciliation legally can and cannot do

• Sacks corrects his own prior concession: the DOGE cuts cannot be in the Big Beautiful Bill — reconciliation (the Byrd rules, 50-vote threshold) only touches mandatory spending; discretionary cuts need 60-vote appropriations or a separate rescission bill. “If the DOGE cuts don’t happen through rescission, I’m gonna be very disappointed… but it’s just wrong to blame the big, beautiful bill.”
• Second correction: the bill “does actually cut spending” — the CBO scores removing the 2017 tax-cut sunset as a spending increase, but against a current-year baseline it’s a mandatory spending cut.
• Friedberg’s numbers via the besties’ call with Senator Ron Johnson: mandatory is 70% of the federal budget, with interest “well over a trillion dollars a year on its way to a trillion five.” Medicaid: $627B in 2019 → $914B in 2024; the $70B/yr cut lands at $840B — “still roughly, call it, 40% above where you were in 2019.” SNAP: $15B of the $120B buys soda; the cut to $90B still sits 50% above 2019’s $60B. The bill passed the House by one vote — Chamath says he doesn’t put the blame on the White House.
• Pressed on whether Trump balances the budget in four years, Sacks punts: “you’re gonna have to get Scott Bessent on… this is just not my area” — the administration’s bet is growth. Jason’s closing shot: it’s “a lack of leadership on Trump’s part” — if he can criticize Taylor Swift and Zelensky on Truth Social, he can criticize Congress on spending; “tone down the tariff chaos” and lean into intelligent immigration.

8. The only way out is GDP — and the long end of the bond market is the judge

• Chamath actually downloaded the CBO’s model: “at best, it’s spartan” — undisclosed discount rate, brittle assumptions — so “I don’t think a financial analyst or somebody that controls a lot of money will actually put a lot of stock in their model.” Navarro’s op-ed “nails it right in the bullseye”: the CBO’s fatal error is assuming 1.7% GDP growth — build a sensitivity at 2.2%, 2.7%. Chamath’s bet: “I think the GDP print’s gonna come in above three. Not quite four, but above three” in Q2.
• The market structure as he sees it: the biggest actor is Trump; the second is “the long end of the bond market” — central bankers, long-bond holders, macro hedge funds who set America’s cost of capital. “What’s happened over the last little while is that they’ve steepened the curve, and they’ve made it more expensive for us to borrow money.”
• The precedent for optimism, from Friedberg: in 2017 the CBO projected 1.8–2% growth and it came in at 2.9% after the tax cuts. His honest caveat stays honest: “I don’t think anyone fucking knows how much the GDP is gonna grow” — and deregulation isn’t captured anywhere: drugs to market in 5 years instead of 10, reactors in 7 instead of 30, and investment follows.
• Sacks’s FRED anchor: federal receipts have averaged ~17.5% of GDP ±2% whether top marginal rates were 90%, 70% (Carter) or in the 20s — revenue tracks the economy, not the rate; the ~20% top tick was the 2000 boom. Outlays ran ~20% of GDP for decades, spiked to 30% in COVID, and sit in the low 20s — closing that gap to a 3% deficit is Bessent’s 3-3-3 target.

9. Energy is the binding constraint on the whole thesis

• Chamath’s chart, “without judgment… just the facts”: the US is at a supply-demand standstill — “we literally consume every single bit of energy that we make,” demand grows ~3% a year, and on the margin “sometimes there are brownouts” because capacity wasn’t added.
• The timelines if you start a project today: SMRs are 2035-plus even fully permitted; an unplanned nat-gas plant is four years out; only three mothballed nuclear reactors are restartable (2027–2030); 24 gigawatts of planned nat-gas sit in a queue that can’t turn on. Strip away the partisanship: “we have ready supply of renewable and storage options today — it’s the fastest thing that you can turn on.”
• His skin in the game and his bottom line: he announced a one-gigawatt data center in Arizona on Tuesday, “little old me” next to people “ripping in huge, huge, huge checks.” “We cannot lose the energy market because that is the critical driver of all the GDP” — narrowly fix the bill’s energy provisions, let Trump get what he wants, “and I think we live to fight another day.”

10. US Steel and golden votes: national champions versus keeping government out

• Trump cleared Nippon Steel’s $14.9B acquisition of US Steel on Friday — the deal Biden blocked — reframed as a “partnership” creating 70,000 jobs and involving partial ownership “controlled by the USA.” Chamath’s context: America is “always on the wrong side of these deals,” showing up for stranded assets (TARP, the auto bailouts) and getting “not much in return” — while Brazil (Embraer, Vale), the UK (Rolls-Royce) and China (ByteDance, CATL) hold golden votes in thriving companies.
• The model he wants copied: Hu Jintao’s 2003 plan for ten national champions “in all the critical industries… and they did it.” His five: pharma precursors, AI, upstream lithography, EV deposition and chip-making, batteries, and rare earths plus the specialty chemical supply chain — “if you have those five, you are in control of your own destiny.”
• Sacks’s conditional endorsement: the past 25 years weren’t a free market — China subsidized champions under WTO developing-country cover and dumped products to become the low-cost producer. So steel, aluminum and rare earths are strategic (“clothes and toys we don’t necessarily have to reshore”); empower the free market where America wins, protect where national security demands.
• Friedberg’s dissent — worth keeping in full: “I don’t like it. Keep the government out of the markets.” It’s inefficient and a slippery slope where “every industry suddenly gets government intervention”; use tariffs and trade incentives to pull manufacturing onshore instead. If government must invest, do it through Social Security’s $4.5 trillion, currently “completely ripped off” earning ~3.8–4.5% lending to a downgraded Treasury — invest it in equities like Australia’s supers before the program’s 2032 functional bankruptcy. Congress ducking that reform is “a degree of disgrazia in this bill.”