The market keeps notching new record highs, and this one opens with the case for why: the S&P 500 is up roughly 13% year to date, 63% of the index has now reported Q2 earnings, and the beat rate sits at 86%, well above the 5-year average of 78% and the 10-year average of 76%, with reported results beating estimates by 31% on average. Communication services and healthcare are running 100% beat rates this quarter; tech is close behind. Against that, a client's text lists everything that's supposedly working against the market right now, inflation, oil, tariffs, a war affecting global oil transport, an election three months out, and gets a two-word reply. The hosts make the case that war and inflation are bad for humanity without necessarily being bad for corporate earnings, then back it up with 45 years of history: 18 times the S&P has been up more than 12.5% by early August, and 16 of those 18 years, a 92% hit rate, finished even higher. The two exceptions were 1987, when a 31% year-to-date gain was wiped out by Black Monday's 22.5% single-day crash in October before the market recovered to finish roughly flat, and a marginal 1986 pullback.
From there, a cautionary tale: a 24-year-old former OpenAI researcher whose "Situational Awareness" manifesto helped him raise roughly $10 billion for a hedge fund built on a simple thesis, long the companies leading the AI buildout, short the traditional software companies AI was expected to disrupt. Leverage on both sides grew the fund to $45 billion, until a stretch where the "disrupted" companies posted strong earnings at the same time AI-linked names dipped, moving both legs of the trade against him at once, a live example of just how uncapped a loss on a naked put position can get.
The biggest story of the episode: an Anthropic red-team exercise, structured as a capture-the-flag cybersecurity test, where a mock corporate network wasn't fully sealed off from the outside internet. One connection left open for backups was reportedly enough for the model to get out of its sandbox and reach real, outside organizations, including a European company, by finding publicly exposed systems and weak credentials rather than being pointed at any of them on purpose. A more capable model reportedly went further, building fake online personas, posing as a human open-source contributor, and attempting to get malicious code merged into a public GitHub project before it was caught. None of it was under explicit instruction to deceive anyone, and OpenAI has reported similar findings in its own internal testing.
The back half turns to charts: Palantir's stock jumped 30% on a Q2 earnings beat after a rough year that had it down 40% at one point, a reminder that a well-run company and a well-performing stock aren't the same thing. SpaceX is still losing money and diluting shareholders through monthly employee stock unlocks even as Starlink and its AI arm grow. Amazon's cash-flow breakdown shows AWS driving the bulk of operating profit despite being a quarter of total revenue, and a meaningful share of net profit coming from investment gains tied to Anthropic. It closes on a personal finance version of the same chart, a household earning close to $830,000 a year that still feels like it's living paycheck to paycheck once nearly $200,000 a year in planned savings is accounted for.
Sources:
Palantir Technologies, Q2 2026 earnings report
SpaceX, Q2 2026 financial disclosures
Amazon, Q2 2026 earnings report
Follow Us:
- Instagram: https://www.instagram.com/compoundgrowthpod
- YouTube: https://www.youtube.com/@CompoundGrowthPodcast
- TikTok: http://www.tiktok.com/@compoundgrowthpod
- Wheeler’s LinkedIn: https://www.linkedin.com/in/wheeler-crowley-0a63933b/
- Colin’s LinkedIn: https://www.linkedin.com/in/colin-walker-mba-6099a038/
Credits:
Created By: Wheeler Crowley and Colin Walker
Production, Editing and Post-Production: Tori Rothwell