Software Stocks Implode, Claude's Hit List, State of the Union Reactions, Trump's Tariff Pivot

Summary

• Anthropic went “three for three in tanking different market sectors” in February: a Claude Cowork legal plug-in (Thomson Reuters, LexisNexis, LegalZoom all −10%+), Claude Code Security (CrowdStrike, Cloudflare, Okta down), then COBOL modernization — and since 85% of COBOL runs on IBM machines, IBM fell 13%, its worst day since 2000, $31B of market cap gone.
• Chamath’s structural call: the market has moved from debating when software cash flows erode to if they exist at all — “is there some AI model that’s gonna come around the corner and obliterate this business without me knowing it?” Unpriceable event risk demands a margin of safety: PEs from 40→20 and 20→10, revenue multiples 10x→3x, WACC from 6% to 12-13%. Layered on top, smart-money hedge funds are “massively degrossing.”
• The viral Citrini doom post (28M views on X) that hit financials Monday — Amex and Capital One −8%, Mastercard −6%, Visa −4% — had its authorship amended after publication to include a managing partner of a $262M hedge fund with confirmed shorts in named companies. Sacks questions the organic virality and echoes Derek Thompson’s frame: AI discourse is “a marketplace of competing science fiction narratives,” “more literary than genuinely analytical.” Prediction markets give the scenario ~12%.
• The bull rebuttal runs through Jevons paradox: Anthropic itself is hiring a software engineer at $570K (“something doesn’t quite add up”), Citadel Securities shows SWE job postings up ~10% YoY with company formation rising, and Aaron Levy argues chronically supply-constrained software demand absorbs the new supply. Sacks: Fortune 500 cost structures are ~5% IT — per Elon’s “cybernetic organisms” framing, maybe they should be 50% software.
• Jason’s live experiment is the tell: OpenClaw agents now do SDR, clip-making, and management-reporting jobs at his firm, staff are “ten or twenty percent more efficient every week,” and he’s “definitely not adding people.” Open-source Kimi 2.5 does “80, 85% of the jobs” at massively lower token cost, and the renegotiation threat to SaaS vendors is “we could roll our own.”
• Chamath’s next shoe: to buy five-six years to figure AI out, tech companies will have to look at stock-based comp, which “literally incinerates most, if not all of their free cash flow.” Meanwhile he sees 10X token demand and a ~90% output-token price decline “probably by the end of this year” — his team now models tokens as a line item in fully burdened employee cost.
• Data-center NIMBY math: ~5GW canceled last year, ~7GW at risk in ‘26 — at Sarah Friar’s $10B-of-revenue-per-gigawatt, that’s “$130 billion of lost revenue” over two years. Trump’s State of the Union answer: a ratepayer protection pledge — hyperscalers fund their own power behind the meter so residential rates don’t rise.
• SCOTUS struck down the EEPA tariffs 6-3 — the biggest rebuke of executive policy since FDR’s New Deal in 1935 — but Trump immediately invoked Section 122 (15% global, 150 days) and Sacks reads Kavanaugh’s 70-page dissent as “a roadmap” to permanent tariffs via Sections 301/338: “future administrations, whether Republican or Democrat, will keep some version.” Also: the first human Yamanaka-factor trial under an FDA agreement (Sinclair’s Life Biosciences, eye indication) is underway.

Deep dive

1. Anthropic tanks sectors — and the market reprices from “when” to “if”

• The February scorecard: Feb 3, a legal plug-in for Claude Cowork — Thomson Reuters, LexisNexis, LegalZoom all down at least 10% since. Feb 20, Claude Code Security in limited research preview — CrowdStrike, Cloudflare, Okta down. Feb 23, Claude can modernize COBOL — the language running 95% of US ATMs and Social Security payments, 85% of it on IBM machines — and IBM dropped 13% Monday, its worst day since 2000, $31B in market cap.
• Chamath’s tactical explanation: smart-money hedge funds are in a degrossing cycle — “the longs become less long, the shorts become less short, and you just shrink” — creating general downward pressure regardless of fundamentals.
• The structural one matters more. Markets used to debate when cash flows get impacted — Coca-Cola’s, Eli Lilly’s, Meta’s. Now: “Are these cash flows durable at all? Could they fall off a cliff in year three?” Unpriceable event risk means holders demand a huge margin of safety: PE of 40 goes to 20, 20 to 10; 10x revenue goes to 3x; a 6% WACC goes to 12-13%. “That’s the market’s way of saying, I’m now debating if these things will even exist.”
• His hedge on the call, exactly as hedged: “The answer may be for many of these companies that they will survive, but we don’t know how long” — and the first-order casualty is how tech recruits, since eroding cash flows tie directly to stock-based comp.

2. The Citrini doom post: 28M views — and a short fund’s name added after publication

• The Substack fan fiction, set in a fictional 2028 “global intelligence crisis,” sketches a death spiral: AI cuts staff, margins rise, consumers lose discretionary income, companies cut deeper — 10% unemployment, S&P down 38% from highs. Posted Sunday night; Monday, on its claim that agents kill the 3% interchange fee via stablecoins, Amex −8%, Capital One −8%, Mastercard −6%, Visa −4%.
• Sacks’ receipts: the piece “got passed around like a joint at a Grateful Dead concert,” but attribution was amended after publication to add a managing partner of a $262M SEC-registered hedge fund who “confirmed short positions in the companies the report named.” “Did this article truly go viral, or did the authors do anything to kind of amplify it? We just don’t know.”
• On the merits he leans on Derek Thompson’s “Nobody Knows Anything”: with real-world data on AI’s macro effects “so paltry,” serious AI conversations are “more literary than genuinely analytical” — a marketplace of competing science fiction. A prediction market puts the Citrini scenario at ~12%.
• Why doom wins anyway, per Sacks: sci-fi skews dystopian, the seen-versus-unseen bias (obsoleted jobs are visible; uninvented ones take “some great innovator or a genius” to imagine), and the fixed-pie fallacy. His borrowed line: “The economy is not a pie, it’s a garden, and technology is rain.”

3. SaaS’s broken annuity math — and the Jevons rebuttal

• Sacks’ VC framing of what broke: SaaS categories used to settle into a leader with predictable ARR and ~120% net dollar retention — “an annuity with growth” bought at ~13x ARR. Now: “I don’t think AI is gonna get rid of Salesforce, but it could eat into their growth opportunity. We just don’t know.” Chamath’s blunt corollary: “If you can get 5% for owning government bonds, why are we taking excessive risk here?”
• The contradiction Sacks flags: Anthropic has a $570,000 software engineer listing live while the broader thesis predicts job elimination — “Anthropic doesn’t really seem to be practicing what they’re preaching… Something doesn’t quite add up.” Chamath, half-joking about retention: with $5-6B structured secondaries making that equity effectively “cash compensation,” “for me to match that, I need to be 3X higher.”
• The data rebuttal: Citadel Securities’ response to Citrini shows SWE job postings up ~10% YoY and company formation rapidly expanding. Aaron Levy’s Jevons argument: software engineering was chronically supply-constrained from startups to the Fortune 500, so cheaper supply meets a massive unfilled need — more 10X engineers spread across the economy, not firings.
• Sacks’ cybernetic math: the average Fortune 500 cost structure is ~5% IT, maybe 1-2% software; Elon describes companies as “cybernetic organisms that are part software, part human” — perhaps they should be 50% software. Chamath’s translation: OPEX as a percentage of revenue “falls off a cliff,” but the technology share within it “goes way, way up.”

4. Friedberg’s never-faced question: is there an upper limit on consumption?

• His rubric — offered with the honest caveat “that’s just like an anecdote” when Chamath presses for data — is that humans need roughly 10% yearly improvement in livelihood to be happy; that sets the floor on consumptive capacity.
• The ceiling is the new problem: AI productivity may make the ability to produce exceed the capacity to consume, “something that I don’t think we’ve faced before” — where economic, productivity, and social models break. His bigger claim: like SaaS, “knowledge work in general is also a transitory phenomenon” existing only between the computer and AI. Sacks’ pushback, worth keeping: “Is this just another dueling science fiction take? … is there data?”

5. Jason’s OpenClaw weekend: knowledge workers, not developers, automate everything

• Fifteen of his firm’s twenty people trained for six-seven hours, each with their own OpenClaw agent: “Every piece of software that we wanted to buy or build over the last 10 years that we never got to, my people are building in the last 30 days.” Best specimen: an SDR agent that scans the top-100 podcasts’ transcripts for advertisers, cross-checks them in Pipedrive with last-contact dates, and loads the sales room — a job they’d wanted to fill; the human gets redeployed, not fired.
• It’s getting recursive: a thumbnail agent instructed to research weekly found a MrBeast staffer discussing heat maps — “an article I would have never known” — and added it to its own skills file. His “Ultron” agent has root access to Gmail, Calendar, Zoom, Notion, and Slack, produces per-person weekly summaries, and “it’s helping manage those people.”
• The SaaS-killer moment: asked for missing Slack functionality, the agent replied “have you considered Matterpost?” (likely Mattermost) — offering to spin it up and export the Slack instance over a weekend. The leverage: “we could roll our own” — which pressures upselling, a big part of SaaS. Chamath adds Ryan Peterson’s post: “Claude for legal seems to work just as well as Harvey” — the apocalypse “is going after private companies now too.”
• The punchline Chamath drags out of him: “Despite all your doomerism… you’re growing and more productive” — but Jason is “definitely not adding people”; staff get “ten or twenty percent more efficient every week.” And standing up open-source Kimi 2.5, which “can do about 80, 85% of the jobs,” lowered token bills massively. “It’s the most exciting time I’ve had online since the web came out.”

6. Batten down the hatches: stock-based comp is the next shoe, tokens are a payroll line

• Chamath’s survival logic: to buy five-six years of cash to figure AI out, companies need to look at compensation — tech companies burn “most, if not all of their free cash flow fighting the dilution from stock-based compensation.” At some point “the next shoe will drop” on SBC.
• His token forecast: 10X demand, but ~90% price reduction per output token “probably by the end of this year” — the price cut itself creating “an enormous upswell of demand.” His own team redid its cost model so tokens sit inside fully burdened employee cost: “some of our engineers are just racking up ginormous bills.”
• Sacks’ constraint check on both extremes: tokens per second, per watt, per dollar are all improving fast, but land/power/shell, energy, or chip production may constrain the system “in the next couple of years” — so “these hyper-utopian or hyper-dystopian narratives will be wrong”; the economy can’t transform that fast. Jason’s aside: a rumored M5 Mac Studio built “language model ready” could put local models on every desktop.

7. The data-center revolt: $130B of forgone revenue, and Trump’s ratepayer pledge

• A weekend data-center tally: ~25 projects total, 20 in Q2 alone, with 100 more facing local opposition. Applying the ~40% rate to announced megawatts: ~5GW lost last year, ~7GW at risk in ‘26. At Sarah Friar’s rule that a gigawatt is ~$10B of OpenAI revenue, that’s $50B lost in 2025 plus $70B in ‘26 — “$130 billion of lost revenue… We need to figure out a way to nip this in the bud.”
• The fix announced at the State of the Union: a ratepayer protection pledge — hyperscalers provide their own power behind the meter so residential rates don’t rise; excess sold back to the grid plus scale economics “could actually bring down consumer prices.” His label for the opposition: “BANANAS — Build Absolutely Nothing Anywhere Near Anyone.”
• Friedberg’s two warnings: data moves at the speed of light, so if America balks, one of 196 countries hosts instead — Saudi, UAE, and Qatar are already “10X-ing their data center builds.” And the utility business model itself — CapEx plans approved by utility commissions can be invested for a return — can push rates up “independent of data centers,” so accountability has to extend there too.
• The litigation machine: Micron’s $100B New York megafab took 1,200 days from announcement to groundbreaking and faced a lawsuit from six citizens (612 days on the environmental study alone) while Elon built the Texas Gigafactory in ~18 months; a Nevada lithium deposit was blocked over the upper land grouse (likely sage grouse); a North Dakota judge ordered Greenpeace to pay ~$350M over pipeline protests. Chamath: “It should not be the case that six people can slow down a $100 billion investment package.” Friedberg’s dissent from the pile-on: this is left-behind emotion, and without “mechanisms that helps the broader population… get some benefit,” they’ll keep using their power to stop it.

8. State of the Union: 80/20 issues, and Democrats failing the easy test

• The 108-minute address — longest since tracking began — landed against a 26-point approval swing (+11.7 to −14.3). Sacks’ core read: Trump served up “80/20 issues… even 90/10 or 95/5” and Democrats refused even polite applause — for grieving families of victims of criminal illegal aliens, for Erika Kirk on political violence, even for lower drug prices. Two-thirds of CNN’s viewers and ~three-quarters of CBS viewers rated it effective; Trump’s “these people are crazy” was delivered “in an almost mournful way.”
• The bipartisan flash: Elizabeth Warren stood for the Stop Insider Trading Act, teeing up the night’s best zinger — “Did Nancy Pelosi stand up if she was here?” Sacks’ point: Warren stood because she agrees on that issue and not the border — so this is substance, not mere polarization.
• Jason’s dissent, undiluted: “Trump is the divider in chief,” and the counterpunch-never-apologize philosophy is “part of what’s broken down in our politics” — vote moderates in the midterms. Sacks’ counter: try working with Ilhan Omar; Susan Rice just promised prosecutions of Republicans and even cooperating tech companies. Jason’s concession: “They’re absolutely going to do that… Both sides need to drop the lawfare.”

9. Science corner: Yamanaka factors enter human trials — starting with the eye

• Friedberg’s headline: David Sinclair’s Life Biosciences reached an agreement with the FDA to become the first company to treat humans with Yamanaka factors — AAV-delivered DNA used for an eye indication involving injection into the eye’s vitreous fluid to rejuvenate retinas blinded by glaucoma or stroke-like eye disease, with protein production switched on and off via the antibiotic doxycycline. Phase 1 is underway: “If it works, which it’s expected to, because we see this result happen in animal models.”
• His caveat comes first, as told: Sinclair is “a bit of a controversial character,” bemoaned as “a little too over-hypey, snake oil salesman” — he sold a resveratrol company to GSK for $720 million and “it didn’t end up working.”
• The arc: aging damage is often damage to the epigenome, not DNA — reset the markers and “boom, the cell is young again” (monkeys have shown wrinkles disappearing). A 2006 discovery finally in clinic, with over a dozen startups behind it: “the beginning of… the most extraordinary revolution in human therapeutics,” possibly “the fountain of youth.”

10. SCOTUS kills the EEPA tariffs 6-3 — but Kavanaugh wrote the workaround

• The biggest rebuke of executive policy in 91 years, since FDR’s first New Deal in 1935: Roberts, Barrett, and Gorsuch joined the liberals. $175B collected, ~50% potentially refundable, 2,000 importers already filed; Polymarket’s refund odds spiked from 18% to 40%, while Congress passing tariffs by March 31 sits at 3%.
• Sacks’ call: the tariffs aren’t going away. Trump immediately invoked Section 122 of the 1974 Trade Act (15% global, 150 days), and the 70-page Kavanaugh dissent “provided a roadmap” — use the window to substantiate Section 301 (unfair trade practices) and Section 338 (discrimination against US commerce). Note the majority said nothing about refunds — and “why would we want to give back hundreds of billions of dollars to a bunch of importers when we’re trillions of dollars in debt?” His prediction: “future administrations, whether Republican or Democrat, will keep some version of the tariffs.”
• Chamath goes further: “the experiment has been successful” — the imbalances were “a hodgepodge of globalist drivel,” and Congress should now ratify the tariffs permanently. Friedberg’s institutional takeaway: Trump’s own appointees ruled against him, which “should give everyone good faith that the system that the founders set up is working.”