Thursday, April 30, 2026 · 181 newsletters

The War Becomes Infrastructure

hormuz-as-system · anthropic-builds-the-underneath · the-coalition-cracks · two-clocks-american-politics · wall-street-main-street-gap · ai-capex-as-gdp · operator-honesty-arrives · tokenmaxxing-and-the-pricing-fire

April was the month an event turned into a system. The Iran war that began as a price spike on March 17 stopped being a discrete crisis and became the operating environment, with a toll regime on Hormuz, a 60-day War Powers clock used as a procedural reset, gas at $4.43 a gallon, Spirit Airlines liquidated, and consumer sentiment at the lowest mark in the 74-year history of the Michigan survey. The AI buildout completed the same transformation in miniature: it stopped being a model race and became infrastructure, with Anthropic crossing $900 billion in talks, the four hyperscalers committing $700 billion in 2026 capex, and the operator literature visibly migrating from "what can the model do" to "where does the human judgment live in the loop."

The Month in One Sentence

April was the month the war stopped being a foreign-policy story and became a domestic cost-of-living regime, while the AI buildout stopped being a model race and became the unglamorous layers of permissions, credentials, standards, and silicon underneath.

Arc: Hormuz Becomes The System

Week one (ending April 5) opened with a speculative shock. Paul Krugman set the frame on Monday with "The Oil Crisis Is About to Get Physical," using JPMorgan's tanker-arrival map to show the grace period ending mid-week. Brent cleared $141 spot intraday Wednesday. Trivium China had the most underrated piece of the week: strikes on Qatar's Ras Laffan complex took roughly a third of the world's helium offline, which is the molecule semiconductor fabs cannot run without. By Sunday the IEA was calling this the largest supply disruption in oil market history.

Week two (ending April 12) was the madman moment. Trump's Easter morning Truth Social post threatening "every bridge in Iran will be decimated" climbed down 90 minutes before his own deadline, after Pakistan brokered a two-week pause. Andrew Egger at The Bulwark called it cleanly: "the ceasefire we announced with so much fanfare on Tuesday has already ceased to exist, indeed never existed at all." Iran demanded crypto payments for passage. Vance flew home from Islamabad without a deal.

Week three (ending April 19) was when it became infrastructure. Trump ordered a US Navy blockade Monday; by Thursday Hegseth had expanded it to every ship in the Strait. Former Biden energy adviser Amos Hochstein told Semafor the line of the month: "Regardless of how the war ends, Iran will have control of the Strait of Hormuz. Once you get that genie out of the bottle, it doesn't come back in." Matt Klein at The Overshoot ran the contrarian macro read: the EIA estimated Saudi Arabia, Kuwait, Iraq, and the UAE shut in 7.5 million barrels per day in March and 9 million in April, approaching 12 million if the Strait stayed closed through month-end.

Week four (ending April 26) was the second-order story. Maritime Analytica ran the piece that mattered: Panama Canal auction prices for Neopanamax slots reached $4 million, up from $140,000 to $380,000 in weeks. Trivium China had the China side: Persian Gulf crude imports collapsed 34% year over year in March, LPG fell 41%, and Chinese exports to the region fell 57%. ChinaTalk's interview with Bryan Clark of Hudson confirmed Pacific-war munitions, L-RASM, JASSM-ER, and Tomahawks, had been burned on Iranian corvettes, with a six-year refill pipeline.

Week five (ending May 3) was when the polling hardened. Crooked's What A Day and Semafor DC hit the same chord Friday: 61% of Americans now think the war was a mistake, matching the 1971 Gallup read on Vietnam and the 2006 Post-ABC read on Iraq. California gas hit $6, the national average $4.43. Matt Stoller at The Big Newsletter had the first corporate casualty: Spirit Airlines liquidated, 17,000 jobs gone, jet fuel doubled. The UAE walked out of OPEC. The blockade stopped being an event sometime around week three and started being infrastructure. American polling on an unpopular war usually takes years to harden. This one congealed in two months.

Arc: Anthropic Builds The Underneath

The single biggest non-war story of the month, and the one that ran on a different clock than the headlines suggested.

Week one set the embarrassment. Anthropic shipped its own Claude Code source map to npm in a routine release, exposing what The Neuron called "a blueprint for how AI coding agents actually work." Ben Thompson at Stratechery paired the leak with the axios supply-chain attack on a package with 300M weekly downloads. The same week Lenny Rachitsky confirmed Anthropic had scaled from $1B to over $19B in ARR in 14 months.

Week two was the bankers' meeting. Linas Beliunas at Fintechnize reported Treasury Secretary Bessent and Fed Chair Powell pulled the CEOs of Citi, Goldman, Morgan Stanley, Bank of America, and Wells Fargo into a closed-door session on April 8 over a single AI model: Anthropic's Mythos Preview. Within a week the Bank of Canada, the Bank of England, the FCA, and HM Treasury were running parallel convenings. Anthropic stopped selling intelligence and started selling infrastructure: sandboxed Linux containers, append-only session logs, a credentials vault outside the sandbox, OpenTelemetry tracing. Ken Huang walked through how this could upend Salesforce, SAP, Oracle, and the entire ERP layer.

Week three was the parabolic move. Bloomberg led with investor offers valuing Anthropic above $800 billion, more than double its prior round. Opus 4.7 shipped Thursday with a developer revolt the same day. Claude Design shipped Friday, and Figma closed down 6.84%. By Sunday Techmeme had the NSA running Mythos despite the Pentagon's own supply-chain-risk designation.

Week four was the pricing fire. The Neuron had the recap: Claude Code quietly disappeared from the $20 Pro plan. The post hit 4.9 million views. Anthropic claimed only 2% of new prosumer signups were affected; the change had propagated to the public pricing page, support docs, and help center. Sam Altman quote-tweeted it with two words: "ok boomer." Anthropic reverted within hours. The same week Amazon committed $25B to Anthropic against a $100B AWS spend commitment, then Bloomberg reported Alphabet planning another $40B at a $350B valuation.

Week five flipped the leverage. Axios ran the scoop: after months of lawsuits, the supply-chain-risk designation, and a draft executive order to ice Anthropic out of government entirely, the White House was quietly welcoming Anthropic back. By Saturday Contrary Research had Anthropic at a $40 billion run rate with preemptive offers in for a $50 billion raise at up to a $900 billion valuation, three months after a $30 billion Series G at $380 billion. The moment a model is too useful to ignore, the leverage flips. Anthropic was the test case and Anthropic won.

Arc: The Coalition Cracks On Three Fronts

The political through-line that hardened over the month.

Week two was the MAGA break. When Tucker Carlson, Megyn Kelly, Alex Jones, Candace Owens, and MTG all used "evil" or "madness" inside one news cycle, that was a coalition crack, not a news cycle. Matt at Crooked catalogued the defections after the Easter post: Candace Owens calling for the 25th Amendment, Tucker calling Trump's post "vile on every level," Trump firing back in the New York Post calling Carlson "low-IQ."

Week three was the international break. Bill Kristol and Andrew Egger titled their Morning Shots "MAGA International Goes Down in Flames" after Peter Magyar's Tisza party took 138 of 199 seats against Orban. Paul Krugman's interview with Kim Lane Scheppele had the operational playbook: Magyar spent two years going village to village inside a system Orban himself had rigged. JD Vance's foreign policy collapsed the same week on Iran and Hungary at once. Reid Cherlin at Crooked called Trump's fight with Pope Leo "Holy War."

Week four was the structural break. Dan Pfeiffer had the cleanest framing: the buried story is that white working-class voters, Trump's most reliable base across two cycles, are abandoning him. Marjorie Taylor Greene's "Republicans are going to get slaughtered in the midterms" was the moment that should have been the story and was the second story most days because the war kept eating the cycle. Cook shifted Senate odds toward Democrats in North Carolina, Georgia, Ohio, Iowa, and Nebraska.

Week five hardened the math. Bruce Mehlman's Six-Chart Sunday had Democrats' Kalshi odds to take the House at 86%, up from 58% six months earlier, with the Senate at 50%, up from 27%. Trump's approval sat at 41%, below the 44% he carried into 2018 when the GOP lost 41 seats. Andrew Wilson at Lincoln Square declared Trump "functionally finished as a political force." Saturday night an armed Californian fired a shotgun into a Secret Service agent's vest at the Washington Hilton during Correspondents' Dinner. Trump shrugged from the briefing room and pitched his ballroom.

Arc: Two Clocks, One Country

The arc that ran underneath every other arc.

Week one ran the cabinet purge. Trump fired Pam Bondi on Wednesday and installed his former personal defense attorney Todd Blanche as acting AG. Gov Brief Today had the sharpest read: Blanche's first act was a DOJ opinion declaring the Presidential Records Act, the law Jack Smith built the classified documents case on, unconstitutional. Trump's former defense lawyer killed the law his own client was charged under, on day one.

Week two added the structural rewrite. Brian Beutler at Off Message flagged the OLC quietly opining on April 1 that the Presidential Records Act was unconstitutional, meaning Trump now claims to own all of his presidential records. Judd Legum at Popular Information wrote the DHS SAVE database piece: designed for benefits eligibility, weaponized for voter purges, with Missouri flagging 35% of naturalized citizens as suspect.

Week four moved the map. Virginia voters approved a Democratic redistricting plan that could shift the state from a 6-5 Democratic edge to one favoring Democrats in 10 of 11 districts. Marc Elias called Tuesday night the payoff for a decade of litigation. A Virginia judge blocked the map a day later. The Daily Skimm had Rep. Kevin Kiley on the record: "I wish none of this had happened."

Week five removed the brake. Wednesday, the Supreme Court's Louisiana v. Callais decision struck down the state's second majority-Black congressional district and gutted Section 2 of the Voting Rights Act. Justice Kagan called it "all but a dead letter." Florida rammed through DeSantis's gerrymander hours later. By Friday, Democracy Docket had Alabama, Louisiana, Tennessee, South Carolina, and Florida queueing up. Bill Kristol and Andrew Egger made the smarter read: the Callais decision does not start the redistricting war, it removes the last brake.

The visible clock was the war. The faster, quieter clock was the constitutional one: an OLC opinion gutting the Presidential Records Act, a Supreme Court ruling ending majority-minority districts, a 60-day War Powers theory ratified by Senate inaction, 5,000 troops withdrawn from Germany to punish Chancellor Merz for using the word "loss" in public. Reading either register alone misses the picture.

Arc: Wall Street Versus Main Street, And The Capex Underneath

The macro arc that ran through every business newsletter.

Week one had Q1 venture funding at $300B globally, with $188B to four AI labs. Roughly 80% of total dollars went to AI, up from 55% a year earlier. OpenAI closed the largest private funding round in history, $122B at an $852B valuation. Om Malik titled his piece "The Fix Is In," reading the JPMorgan/Goldman/Citi/Morgan Stanley/Wells Fargo credit line as an IPO underwriter roster auditioning for the job.

Week two cracked the data. March CPI surged 0.9% month over month, the largest jump since 2022. Bloomberg's Rovella opened with University of Michigan consumer sentiment at 47.6, a record low going back to 1961. Liz Hoffman at Semafor had the next 2008 nobody is watching: $1.3 trillion in bank loans to private-credit firms, collateralized by those firms' own loans, what Advent's managing partner called "leverage on leverage." Krugman had the labor side: the US near-zero job-growth stall, 178k in March after losing 133k in February, with unemployment holding only because the breakeven rate had collapsed too.

Week three booked the bank profits. Banks printed record quarters off Iran-war volatility. JPMorgan revenue up 10% to $49.8B with markets revenue up 20%. Citigroup posted its best quarter in a decade at $24.63B. Wall Street's six largest banks posted nearly $50B in Q1 profits. Meanwhile The Average Joe reported the top ten luxury houses had shed $176B in market value year to date, Hermes shares fell 14% intraday, and Dubai mall traffic was down 70% in March.

Week four added the speculative pivot. Om Malik wrote the must-read piece on NewBird AI: Allbirds, down 99% from its 2021 IPO, executed a $50M convertible to pivot to AI compute infrastructure and popped 700% in a session. The stock traded more volume than JPMorgan and Exxon on Wednesday, $3.8B against a $22M starting market cap. Om compared it directly to the 1960s "tronics boom" and the 1999 suffix mania.

Week five booked the capex commitment. Alphabet raised its 2026 capex range to $180B to $190B. Meta raised to $125B to $145B and shares fell 4.4%. Om Malik ran the sharpest read: Microsoft, Meta, Amazon, and Alphabet collectively guided to roughly $705 billion in 2026 capex, nearly double 2025. Matt Klein at The Overshoot had the cleanest analysis: the year-on-year jump in capex alone added roughly a full percentage point to nominal GDP growth. Strip that out and the picture is messier than the headline. Stocks at all-time highs against the lowest consumer sentiment ever recorded is the K-shaped economy showing up at the pump and the polls at the same time.

Arc: The Operator Class Gets Honest

Running parallel to the capex story, the most coherent operator-side trend of the month.

Week one started the literature. Nate framed it bluntly: "Most of what you're building will be replaced by a better model." Lenny Rachitsky ran Keith Rabois arguing the PM role was dying and CMOs were becoming the #1 consumer of tokens. Ruben Hassid had Karpathy spending four hours refining a blog post with an LLM, feeling great about it, then asking the LLM to argue the opposite and getting demolished.

Week two added the harness paradigm. Ken Huang's Agentic AI opened a new chapter on the harness paradigm: the model provides intelligence but the harness provides control. Addy Osmani at Elevate wrote the corollary: agents need identities not shared credentials, and "a promise in a prompt is not a policy."

Week three moved the bottleneck. Nate argued the bottleneck was no longer the model: agents ran 10 to 50x human speed but got throttled by compilers, file systems, and login flows. Jeff Dean at GTC said infinite model speed nets only 2 to 3x end to end. The $300 overnight loop, Karpathy pointing an agent at his own training code and waking up to 700 experiments and an 11% training-time cut, hit YC the same week.

Week four moved the trust frame. Matt Brown at Vertex 2026 had the line of the month: "Trust is the new taste." Nikunj Kothari at Balancing Act wrote the cleanest version of the design-isn't-dead argument: Claude Design ships the output, which is 0.1% of the work; the other 99.9% is the carrying, the brainstorm, the wrong turns. The Meta "tokenmaxxing" performance review story landed the same week, Goodhart's law on schedule.

Week five named the pricing reality. Gergely Orosz at The Pragmatic Engineer talked to devs at 15 companies and found token spend up roughly 10x in six months. Claude Cowork wrote the operator essay of the month on the PocketOS database deletion: a Cursor agent running Claude Opus deleted a production database in nine seconds after hitting a credential mismatch in staging. Recovery took thirty hours. A prompt is words; a permission system is access controls. Operators kept treating them as the same object and they were not. Maggie Appleton closed the month: "Implementation is rapidly becoming a solved problem. The hard question is no longer how to build it. It's should we build it."

The Story of the Month

The Iran war was the story of the month, and not because it was the loudest. It was the story because by the end of April every other story was downstream of it. The CPI print was the war. The bond rout already starting in early May was the war. The redistricting urgency was the war. The Trump approval drop was the war. The Spirit Airlines liquidation was the war. The Meta-Manus unwind in Beijing was the war. The Anthropic-to-NSA pivot was the war, because the political class needed the leverage of frontier AI in a national-security crisis and could no longer afford the lawsuit posture. The $700 billion of hyperscaler capex was the war in the sense that it was the only force keeping nominal GDP growing while the consumer cracked.

The arc inside the arc is the one Amos Hochstein named in week three: "Regardless of how the war ends, Iran will have control of the Strait of Hormuz. Once you get that genie out of the bottle, it doesn't come back in." The toll regime is the structural fact. Panama Canal slots at $4 million each, Chinese Persian Gulf imports collapsing 34%, helium for chip fabs down a third, and Pacific-war munitions burned on Iranian corvettes are the same fact in five different sectors. When one event explains five sectors, that is a regime, not a headline. By month-end the war was no longer a discrete crisis Iran or Trump could end. It was the operating environment. The 60-day War Powers gambit, the ceasefire that never existed, and the Patriot missile resupply two days after Trump declared the war over are not contradictions. They are the system optimizing for political cover through November while keeping the option open.

In Retrospect

The "two weeks of oil left" framing did not survive contact with the toll regime. Brew Markets ran the number in week one. The actual story by week four was not depletion, it was Iran reframing the war as a tax on global trade. Panama auction slots at $4M, not US crude stockpiles, were the binding constraint.

The "madman theory" frame got tested and failed inside one week. Week two opened with Trump's Easter morning threat to obliterate Iran "Praise be to Allah." It ended with Pakistan brokering a climbdown 90 minutes before his own deadline and Iran openly charging tolls for Strait passage. The frame the rest of week three was written against was the inverse: madman theory does not survive contact with a counterparty who has read it before.

The Anthropic source-code leak was supposed to be the company's worst week. Week one closed with the leaked source map for Claude Code, the OpenClaw crackdown, and the embarrassment of shipping their own internal codenames to npm. By week three Anthropic had crossed an $800B valuation. By week five it was at $900B with the NSA running its unreleased flagship over the Pentagon's own blacklist. The market was pricing the embarrassment; the operators were pricing the run rate.

The "model wars" framing got buried. Week one had OpenAI's $122B at $852B, the largest private round in history, as the AI story. By week four DeepSeek had opened V4 weights, V4 Pro at 1.6T parameters and a 1M context window. Stanford's 2026 AI Index showed the US-China performance gap had effectively closed: Anthropic's top model led the best Chinese models by only 2.7%. The chip gap was no longer the model gap, and the story stopped being which model wins and became where the human judgment lives in the loop.

What to Carry Into Next Month

The war is now paid for by American consumers at the pump and by 17,000 Spirit Airlines employees on the unemployment line. The polling has hardened in two months rather than two years, which is the velocity that will define the rest of the year. The 60-day War Powers gambit, the ceasefire that the Senate has now ratified six times by inaction, the procedural reset dressed as peace are all telling the same story: the administration is not optimizing for ending the war, it is optimizing for the political cover to keep the option open through November. The gas number is structural through the midterms, and the inflation that follows it is more structural than transitory.

The AI buildout has visibly moved from models to the unglamorous layers around them. The pricing fire on Anthropic's Claude Code page in week four was the canary in the VC-subsidized-token coal mine. Tokenmaxxing turned into a metric inside Meta and Microsoft. Goldman counted just $300M of productivity gains across the entire S&P 500, roughly what Anthropic earns in a day and a half. The bottleneck has moved off the model and onto the permissions, the credentials, the standards, and the human judgment around the loop. The companies that quietly build those layers will look unremarkable for two quarters and then become impossible to catch. Anthropic's bad press in week one was the most bullish AI signal in the inbox, because every operator field report kept landing on the same conclusion: the run rate is the message.

If you only revisit three pieces from the month, I would suggest Matt Klein's "Look Through the Hormuz Shock If You Dare" for the contrarian macro read everyone else has missed all year, Bill Kristol and Andrew Egger on the post-Callais redistricting war for the cleanest read on what the Supreme Court just did to the next decade of American elections, and Nikunj Kothari's "The Mess Is the Work" for the cleanest answer to where AI-era judgment actually lives. The month told me three things in sequence: the war is now infrastructure rather than event, the rules of the map have changed faster than the political class can absorb, and the AI buildout has visibly moved from models to the slow layers around them. Those are the three frames I am carrying into May.