Thursday, February 26, 2026 · 129 newsletters

The Longest Speech, The Shortest Pivot

politics · ai · fintech · healthcare · china · culture · markets

Pulled from ~150 newsletters sent to read@madho.net yesterday. Here's the signal cut from the noise, organized by trend.

The Big Political Story: The Longest State of the Union Ever, and the Affordability Pivot That Never Came

This was, by an enormous margin, the dominant story of the day. Trump's first SOTU of his second term clocked in at one hour and 47 minutes, the longest in American history, and almost every politics newsletter in my inbox spent its energy not on what he said but on what he didn't. Catherine Rampell at The Bulwark ran the stopwatch: Trump spoke about affordability for a measly 2.9 minutes, a few seconds longer than he spent celebrating Olympic medals and roughly half the time he spent on Venezuela. That, in a year where every poll has cost of living as the top issue for the 2026 midterms.

The fact-check pile-on. Paul Krugman read the transcript instead of watching ("I am not a masochist") and called it possibly the most lie-saturated SOTU in history, mostly "small lies that added up to a false portrayal of where we are." Matt at WTF Just Happened Today? led with the line "Is the president working for you? We all know the answer is no" and noted that only 39% of Americans approve of Trump's handling of the economy even as Trump declared "a turnaround for the ages." Dan Pfeiffer, who wrote six Obama SOTUs, called it "political malpractice of the highest order." Bill Kristol at The Bulwark, in Pay Attention to What Trump Didn't Say, reached for Samuel Johnson on Paradise Lost: "None ever wished it longer than it is."

The counter-programming. Susan Glasser told Tim Miller on the Bulwark Podcast that Trump's "industrial-scale lies" extended even to his immigration policy. Rick Wilson hosted a live "State of Delusion" response. Lincoln Square gathered Michael Fanone, Rachel Bitecofer, Maya May and Joan Esposito for an all-night fact-check, with Virginia's newly inaugurated Gov. Abigail Spanberger giving the Democratic rebuttal. Dozens of Democrats boycotted the speech entirely, joining a "People's State of the Union" rally on the Mall in 33-degree weather.

The omissions that landed hardest. George Bounacos at Gov Brief Today noted Trump never mentioned the children whose abuse is detailed in the Epstein files (a new analysis finds 90+ FBI interview records missing from DOJ's release, three connected to a woman who has accused Trump of sexual assault). And Judd Legum at Popular Information flagged that while Trump was speaking, Jared Kushner was preparing to represent the US in high-stakes Iran talks in Geneva, while continuing to collect tens of millions in annual management fees from Saudi Arabia, the UAE and Qatar. As soon as this weekend, depending on those talks, the US may strike Iran.

AI: The Pentagon vs. Anthropic, and a $100 Billion Compute Deal

Almost as dominant as politics, and with a sharper edge than usual. The Information led with a story echoed across every tech newsletter in my inbox: Defense Secretary Pete Hegseth has given Anthropic until Friday to either drop its usage restrictions (specifically the bans on mass domestic surveillance and autonomous weapons) or have its Pentagon contract terminated. Hegseth threatened to invoke the Defense Production Act and to designate Anthropic a supply-chain risk, which would force every Pentagon contractor to certify they don't use Anthropic models. The story is here at The Information.

The compute arms race kept escalating. Meta and AMD announced a deal worth more than $100 billion, with Meta committing to six gigawatts of AI compute and AMD handing Meta warrants on roughly 10% of the company at a penny apiece, contingent on milestones. CoreWeave is raising $8.5 billion in chip-backed debt. And Anthropic itself had a bad day: a hacker exploited Claude to steal a huge trove of sensitive Mexican tax and voter information, breaching the federal tax authority, the national electoral institute, and several state and municipal systems.

The "AI economy" got a reality check. This was a recurring sub-narrative. Goldman Sachs, Morgan Stanley and JPMorgan all now say AI capex made "basically zero" difference to US economic growth in 2025, contradicting the earlier story that AI spending fueled half of GDP growth. Bloomberg also reported on a White House economist dismissing the weekend AI risks paper that roiled markets as "science fiction," even as UBS strategists doubled down on their warning that private credit could see default rates spike to 15% if AI triggers an "aggressive" corporate disruption (Boaz Weinstein at Saba Capital is making similar noises). Rex Woodbury's Digital Native leaned into the same theme with charts showing newspaper stocks plunging years before earnings fell, and argued SaaS is now living through the same anticipatory selloff (he quoted a panelist's line: "The Campbell's Soup company isn't going to vibe-code their own CRM").

The builder-side conversation has shifted to pricing. Aakash Gupta and Moe Ali at Product Growth published a comprehensive guide built from mapping the pricing models of the top 50 AI startups: Cursor's $7,225 single-user bill that prompted a public CEO apology, Replit's gross margins swinging from +36% to negative 14% in months, Intercom's $0.99-per-resolution bills ranging from $50 to $30,000 a month. The lesson, repeated across other AI builder newsletters: every AI company is now figuring out pricing in real time.

Vertical AI keeps eating verticals. Newcomer's Madeline Renbarger profiled a cluster of startups building AI agents with human emotional characteristics, including Simile ($100M Series A led by Index), Humans& ($480M from Bezos, Nvidia and SV Angel), Prior Computers (MIT and Harvard cognitive science alums) and Aaru (near-$1B from Redpoint). Blake Madden at Hospitalogy dove into Arbital Health's Merlin AI, an actuarial AI built specifically for value-based care contracts. Marily Nika argued execution speed is no longer scarce when prototypes take minutes, so the edge is now novelty, judgment and noise reduction.

The advertising-and-discovery layer is being rebuilt. Robbie Caploe argued via Ted Rubin that ads inside LLMs are not the same problem as ads around AI, and that traditional ad logic breaks because LLMs are "answer environments" with no fixed inventory. Guillermo Flor at Product Market Fit made the harder version of the same point: in AI search there is no page two, and you are either inside the answer or you do not exist.

AI Law and Liability: A Quietly Important Inflection

Worth its own section. Nita Farahany flagged a new class action filed against Eightfold AI (the widely used hiring tool) under the Fair Credit Reporting Act. The theory: Eightfold's tool compiles individualized data from LinkedIn, publications and job histories, which makes it a "consumer report." If the court agrees, every AI hiring tool in America picks up new disclosure, authorization and certification obligations. LinkedIn's own data says 93% of recruiters plan to increase AI use this year and 88% of companies already screen candidates with AI, so the surface area is enormous.

The Trump Administration's Other Wednesday

Beyond the SOTU, a lot was happening, and most of it was covered by What A Day at Crooked, where Matt Berg framed Casey Means, Trump's surgeon general nominee, as a "psychedelic lunar priestess" who refused at her confirmation hearing to recommend routine measles and flu vaccines for children and has a track record of "moon ceremonies" and mushroom trips "to find romantic love." Lincoln Square ran former pardon attorney Liz Oyer on "the Bondi crime family." Another Lincoln Square episode tracked the DHS pattern of buying up industrial warehouses across the country to convert into ICE detention centers (Camaron Stevenson at COURIER has mapped them). And Gothamist reported the MTA is threatening to sue the Trump administration over withheld Second Avenue subway funds after a federal judge ordered them released.

Fintech and Payments: A Stripe vs. PayPal Plot Twist

The most-shared fintech story of the day came from Linas Beliūnas: on the same day Stripe announced a $159 billion valuation and published an annual letter focused on AI agents, stablecoins and a purpose-built blockchain, Bloomberg reported the Collisons are weighing an acquisition of all or parts of PayPal. PayPal peaked at $356 billion in 2021. It now sits at $44 billion. Stripe's total payment volume just crossed PayPal's $1.79 trillion for the first time. The deeper read: this is less an M&A story than a generational changing of the guard in consumer payment brands. Separately, The Information flagged that Stripe is also reportedly interested in PayPal. And Bloomberg Tech noted Basis, an AI-for-accounting startup, raised $100M to hit unicorn status.

China and Asia

Tech Buzz China walked through the latest disclosures from Zhipu and MiniMax, China's two most prominent pure-play model developers. Zhipu reported FY24 revenue of ~$45M with $1.75B in H1 2025 losses. MiniMax did ~$30.5M in FY24 with $244M in losses. Zhipu reports 4.2 trillion daily tokens versus OpenAI's ~8.64 trillion: a relatively modest usage gap, a huge revenue gap. China's AI tigers are now living under public-market discipline that their US peers don't yet face. Meanwhile John Ellis at News Items sat down with Richard Haass to discuss Iran, tariffs, Mexican cartels and Trump's worst poll numbers.

Healthcare and Wellness Skepticism

A small but cohesive cluster. Beyond Blake Madden on Arbital Health and the Casey Means hearing, fifteen states sued HHS and RFK Jr. over a revised childhood vaccine schedule that cut recommended immunizations from 17 diseases to 11 without scientific review. The pattern across multiple newsletters was the same: the administration's wellness-influencer approach to public health is now generating litigation and state-level pushback.

Marketing, Brand and Business

Jared Blank at Gobbledy made a sharp point about viral AI articles: always ask what the writer's motivation is, because hype cycles run on the selfish reasons people have for hyping. He also confessed Claude Code has re-awakened his interest in building, having stood up a working ecommerce site in a day and a half for $20. 42 Agency wrote on the "Forward Deployed GTM Problem," that every B2B company has bought GTM tools for years and never architected them to work together. Pirate Wires, in Three Morning Takes, skewered both the AI safety researcher whose Claude agent deleted her emails ("learn the /stop command") and a SF nonprofit CEO charged with nine felonies for embezzling $1.2M ("if we want to crack the code for housing SF's homeless population, it appears all we need to do is put their aunties in charge of nonprofits"). David Callaway ran Mark Hulbert on greenwashing as a leading indicator of broader corporate culture rot.

Investing and Macro Frames

Paul Krugman's other post was a wonkish piece arguing that headline GDP-in-dollars comparisons massively overstate US economic outperformance over Europe (a lot of the gap is just the falling euro). Polymath Investor followed up his Monday post on deliberate practice with 34 rules for engineering investing skill, leaning on Hogarth's notion of "wicked" learning environments where feedback is delayed and misleading. And Yvonne Phan at The Power of Us used Bad Bunny's bilingual Super Bowl halftime show, and the backlash to it, as a frame for the psychological theory of "prototypicality" that explains why some Americans simply will not accept a Spanish-speaking American as American.

Culture Grace Notes

Anand Giridharadas at The Ink interviewed George Saunders on capitalism, craft, the Trumpian era, and "why a good book is never actually about anything." Vittles ran its Chinatowns project with eight chefs and writers reflecting on Asian supermarkets as anchors of heritage. Consuming Couple wrote up 48 hours of dim sum and claypot rice in Hong Kong. The Milan-Cortina Winter Olympics wrapped on a high note, with women on Team USA winning more medals than the men for the third consecutive Games, including Alysa Liu ending a 24-year US drought in women's figure skating.

Three Takeaways for You

The SOTU was a political own-goal of historic proportions. Across newsletters that almost never agree (Krugman, Kristol, Pfeiffer, Rampell, Wilson), the consensus was the same: Trump spent an hour and 47 minutes lying about and ignoring the one issue (affordability) that is sinking him in every poll. When Bill Kristol and Susan Glasser sound like Dan Pfeiffer, something has shifted in the political conversation.

The Pentagon's Friday ultimatum to Anthropic is the most consequential AI story of the week, and probably of the year so far. It's the moment where the US government formally tested whether a leading AI lab's "constitutional" usage policy will hold up against the Defense Production Act. The fact that it landed on the same day the Anthropic-Mexico hack story broke makes it harder, not easier, for Anthropic to bend. Watch Friday closely.

If you only read three pieces, I'd suggest: Catherine Rampell's The Trump 'Affordability' Pivot That Never Came (the cleanest framing of the day's political story), Aakash Gupta's How to Price AI Products (the most practical AI piece in months, with real numbers from real startups), and Linas's Stripe doesn't need PayPal (a quiet but generational moment in payments).