AI Bubble Pops, Zuck Freezes Hiring, Newsom's 2028 Surge, Russia Summit
2025-08-22 spoken.md · speaker-labeled ▶ watch ← E239 all episodes E241 →
Every number here comes from the conviction ledger — the same one the pool ranks on. Decay is what the 116 ideas nobody mentioned gave up this week; it applies only when an episode is processed.
Tier crossings
conviction thresholds crossed by this episode — 65 / 45 / 15 · ideas born here show where they landed. why 65 isn't always green
Kill dates that landed since E239
1 hit · 0 partial · 5 miss — windows that closed after 2025-08-15 and up to 2025-08-22, auto-scored against price data and never hand-set. result · return · vs SPY
| idea | result | return | vs SPY | scored |
|---|---|---|---|---|
| 📈 Multifamily breaks on the capital stack, not vacancy | WRONG | -23.5% | -74.8 | 2025-08-18 |
| 🏛️ A Google breakup unlocks value - sum of the parts beats the whole | RIGHT | +25.7% | +8.2 | 2025-08-16 |
| 🤖 Apple launches its own search engine and goes head-to-head with Google | WRONG | -25.7% | -43.2 | 2025-08-16 |
| 🏛️ Harris's federal food price-gouging ban is a discredited price control | WRONG | -33.7% | -51.2 | 2025-08-16 |
| 📈 Return-to-office wins - remote work retreats outside an elite minority | WRONG | -1.3% | -18.8 | 2025-08-16 |
| 📈 Starbucks is a sugar company - GLP-1s and sugar awareness are the real problem | WRONG | +2.0% | -15.5 | 2025-08-16 |
Who moved the board
each voice's push on conviction this episode, in conviction points — supports and opposes · share = % of this episode's movement
What got argued (2 ideas)
ordered by how hard each idea moved · quotes are verbatim from the transcript, timestamps deep-link into the episode
Demand for AI compute outstrips supply for at least the next 12-24 months; the data-center buildout is profitable across the whole chain (labs -> renters -> Nvidia -> fabs) and keeps running, with GPU-backed financing removing the capital constraint.
We've been selling into the enterprise at 8090. This is our first full year of business — we'll do about $40 million of bookings... all back-office type stuff... you can get extremely high rates of accuracy if you implement AI correctly in back-office tasks. [MIT study: 95% of gen-AI pilots failing] paints a very accurate picture — we now need to go through a sorting and a cleansing.
Episode digest
written during extraction and stored in data/extractions/ep240.json — the auditable source of truth, including everything market-adjacent that did not earn a capture
Recorded during the Aug-2025 AI sentiment wobble (MIT '95% of pilots fail' study, GPT-5 disappointment, Altman bubble comments, Meta AI hiring freeze, ~10% AI-stock correction). Sacks calls it a healthy correction — 'early to the middle' of the investment supercycle, explicitly not a bust (attached to the live capex idea, re-birthing it to Aug 2025) — while dismantling the rapid-takeoff/AGI-in-2-years narrative: models are clustering and specializing, 'a more normal technology race.' Chamath discloses 8090's ~$40M first-year bookings as receipts that back-office AI works while sales/marketing pilots fail (attached inside the window of his closed 2025 enterprise-software idea), predicts a social-media-style sorting where thousands of AI app companies collapse to a handful, and — notably, a flip from his 2024 'OpenAI EV goes down' call — makes the bull case for OpenAI at $500B reaching '$1.5 trillion' on DAU compounding (untradeable, private). Rest of the episode is non-market: Trump-Putin Alaska summit and the three-pillar peace framework (Sacks), Newsom 2028 surge analysis, DC politics. Untradeable skips also include Friedberg's continued socialism-momentum call. Labels: correct this episode (Friedberg appears as 'Erik Friedberg' — canonicalized on ingest).