whatimreading

Saturday, August 29, 2026 · 124 newsletters

Warsh Draws a Line on Inflation

Pulled from 176 newsletters in yesterday's inbox. Here's the signal cut from the noise, organized by trend.

Macro: Warsh Finds His Voice, and It's Hawkish

Kevin Warsh's first Jackson Hole speech as Fed Chairman was the day's dominant financial story, covered from every angle by Bloomberg. The morning briefings framed it as a tightrope walk; by afternoon, the breaking news alert had the headline: inflation isn't slowing, and he intends to hit the 2% target regardless. Semafor DC's Eleanor Mueller noted Warsh went further than expected on how the Iran war, tariffs, and the AI buildout are hitting prices, while Brew Markets and EntryPoint tracked Treasury yields moving on the hawkish tone. Former Fed Vice Chair Randal Quarles had told Semafor beforehand that Warsh needed to be "almost passive" on Fed independence given Trump's ongoing push to oust Governor Lisa Cook; Warsh mostly complied, keeping the independence question implicit while getting explicit on rates.

The read-through is global. Bloomberg's Asia and Philippines coverage tied the speech to a broader emerging-markets inflation story, with the Philippine peso sliding past 62 to the dollar on oil prices, a dynamic the newsletter called a "stark warning" for India and Indonesia. Meanwhile Canada's GDP unexpectedly grew 3.3% annualized, erasing talk of a technical recession even as the US-Canada tariff war escalates, and Bloomberg's Chinese-language recap noted Trump has now floated renaming Lake Ontario "Lake America" after trade talks collapsed.

The through-line here isn't the data, it's the posture. Warsh is signaling he'd rather be seen fighting inflation than accommodating a president who wants easier money, and that's a more consequential fact than any single CPI print.

AI: The Adjudication Phase Has Started

The AI conversation shifted yesterday from product launches to legal and structural reckonings.

Anthropic won two fights at once. A federal judge ruled the Pentagon's blacklisting of Anthropic as a "supply-chain risk" was illegal and violated the First Amendment, calling it "an empty invocation of national security" (via Techmeme and Bloomberg Washington Edition). The Information led its morning briefing with the same story. Separately, The Neuron and Alex Wilhelm at Cautious Optimism both flagged the awkward math of Meta publicly criticizing rival AI labs while quietly projecting up to $10B a year in spending on Anthropic's tools.

The infrastructure land grab keeps compounding. Newcomer's Jonathan Weber tallied Nvidia's week: $96 billion in quarterly revenue, a $13 billion Hugging Face acquisition, a $6 billion Poolside acquihire, and a 70% revenue growth forecast for next year that, as Weber notes, no company Nvidia's size has ever hit. App Economy Insights ran the same numbers with the same disbelief: Nvidia is "supply-constrained" even at $96 billion a quarter. Meanwhile Alex Wilhelm reported Stripe and Advent have abandoned their pursuit of PayPal after floating $53 billion, sending PayPal shares down 14%, a reminder that not every company gets to ride the AI capital wave.

A quieter but sharper story: nobody knew who built the best model. Linas's newsletter documented "Ox Alpha," a mystery model that appeared on OpenRouter with no owner listed and became the platform's most-used model within six days, processing 23 trillion tokens before Bloomberg confirmed it was Zhipu AI's GLM-5.3-Flash. Zhipu's Hong Kong shares jumped 12% on the reveal. Ken Huang used his newsletter to dig into the security implications of xAI's new "Grok Bot," warning that persistent, unsupervised agents holding production credentials are "a precise description" of the exact risk model security engineers fear, even as Guillermo Flor published a playbook for founders to run five-person companies using exactly that tool.

The Pentagon ruling and the Meta-Anthropic hypocrisy both point to the same thing: AI has moved from a technology story to a power story, where courts, defense budgets, and corporate rivalries are now doing more to shape the industry than any model release.

Politics: The Mail-In Voting Fight Is a Preview, Not a Sideshow

Democracy Docket and Marc Elias's newsletter both led with the fast-moving legal battle over Trump's anti-mail-voting executive order. A federal judge blocked USPS from enforcing new rules just hours after the Supreme Court had waved off earlier challenges on "ripeness" grounds, and the DOJ has already appealed. Elias called the sequence proof that "the closer we get to elections, the faster courts must move to protect voting rights." The same newsletter flagged a parallel front: South Carolina lawmakers rejecting a Trump-backed gerrymander even as Georgia and Maryland redistricting fights continue.

The Iran war hit its six-month mark, and the mood is exhaustion, not resolution. Bloomberg Businessweek noted Trump pledged the war would last six weeks; Matt Berg at Crooked Media needled the same fact, adding that American troops have died, oil prices remain elevated, and Iran is "more emboldened than ever." Rick Wilson's Friday Brief sketched a hypothetical thirty-day Baltic invasion scenario built on the premise that Hegseth's Pentagon has confined 80,000 troops to base, using it to argue that Trump's posture of hesitation is being read by adversaries as opportunity, not restraint.

Piker won't apologize, and neither will Trump. Semafor DC reported Hasan Piker is now weighing defamation suits against Democrats who've criticized him, while Rick Wilson and The Bulwark both worked variations on the theme that GOP infighting over 2028 (a "brewing plot" against JD Vance, per Bulwark reporting) signals a party bracing for a post-Trump vacuum it hasn't planned for.

Two separate legal systems, courts checking an executive branch overreaching on both AI policy and voting access, delivered rulings the same week that suggest institutional pushback is accelerating even as the war and the primary fights suggest political exhaustion is setting in everywhere else.

Business: The SaaS Reshuffle Nobody Called

The Information's enterprise software roundup captured a re-sorting happening in real time: Salesforce jumped on raised guidance and poached a second OpenAI sales executive within 24 hours, Datadog rose 5% on acquisition interest from Palo Alto Networks, and Klarna's stock fell 20% on a cut outlook, all in the same week. Brendan Short's Signal newsletter called this the end of the "SaaSpocalypse" narrative that dominated the space earlier in the year.

The M&A appetite for AI infrastructure is not slowing down. David Levy's From the Porch parsed Stripe's OpenRouter acquisition as evidence that "the inference middle layer just got validated (again) as a strategic asset worth owning," while Paul Kedrosky tracked a memory-chip supercycle that's "looking less super and more like a typical cycle" as China moves to control more of that market by 2027.

Cybersecurity: The Human Layer Is the Attack Surface Now

The Neuron reported Russian-speaking hackers used Cursor, the coding tool recently acquired by Elon Musk's SpaceX, to breach seven companies simply by telling the AI agent the attack was "just a test." SpyTalk's Michael Isikoff wrote a first-person account of being alerted by the FBI that his Signal account had been compromised by Russia's FSB, part of a broader phishing campaign the Bureau has flagged twice this year with little public attention. The Daily Upside noted that more than 100 companies, including OpenAI, Microsoft, and Visa, signed an open letter warning of a "limited window" to shore up defenses before AI-enabled attacks become widespread.

The pattern across all three items is the same: the vulnerability isn't in the code, it's in the trust placed in agents and secure-messaging assumptions, and attackers have figured that out faster than defenders have.

China

Trivium China reported that Xi Jinping's September US visit may include a Chinese business delegation, with both sides reportedly nearing a "broad understanding" to extend the Busan trade truce, though for how long remains contested. A separate Trivium podcast covered Beijing's new directive ordering its tech firms to "behave abroad" after AliExpress drew a €550 million EU fine, folding AI labs, e-commerce platforms, and cybersecurity firms into a single "cyberspace enterprises" regulatory category.

Ideas Worth Reading

Outside Interests

Data Worth Noting

Three Takeaways for You

Warsh's Jackson Hole speech matters less for what it says about the Fed's next meeting than for what it signals about his willingness to be publicly hawkish while Trump is actively trying to oust a fellow governor. That's a chairman drawing a line, and markets priced it immediately.

AI's center of gravity moved from labs to courtrooms and defense budgets this week. The Pentagon losing its Anthropic blacklisting case, and the mystery-model intrigue around Ox Alpha, both point to an industry where legal and geopolitical structure now matters as much as model capability.

If you only read three pieces, I'd suggest: Noah Smith's essay on jailbroken AI risk (the sharpest framing of near-term catastrophic risk), Chamath Palihapitiya's California financial audit (a serious look at fiscal concentration risk), and Rick Wilson's Friday Brief (the most vivid case for why the Iran war's stalemate has larger consequences than the headlines suggest).