AI Valuation Reflexivity, Record Housing Unaffordability, and the Tokenization Thesis

Key Takeaways

1. Elad Gil describes an AI venture market where tiny teams with no product or revenue raise billions at $10B–$500B valuations, and a Soros reflexivity framing suggests valuations themselves have become the signal.
2. US housing affordability hits a record gap: the income needed to buy a median-priced home is $126K versus a median household income of $86K, a 47% shortfall.
3. US unemployment has stayed below 5% for 60 consecutive months, the second-longest streak on record behind the mid-1960s.
4. Raoul Pal argues the price of intelligence is trending toward zero, with $200/month subscriptions delivering unlimited intelligence that is then reinvested into energy and data-center efficiency bottlenecks.
5. Meb Faber highlights a wirehouse channel that blocks a $1B, 10-year ETF beating 96% of peers while offering private credit closed-end funds, calling the opportunity set "rough."
6. Ben Carlson's data shows the bottom 50% of US households grew net worth 129% since end-2019, the fastest of any cohort, while the top 0.1% grew 106%.
7. Marc Andreessen shares claims that Husky, a model-specific inference engine, runs up to 4.5x faster than Apple MLX, with Underdog's Pareto frontier model hitting up to 730 tokens/sec on a MacBook.

1. AI Venture Valuations and Reflexivity

1. Elad Gil describes the current AI venture landscape as the most disorienting period of his career, with portfolio companies growing rapidly alongside companies that have no revenue, no product, and very small founding teams raising billions at $10B–$500B valuations. A forwarded Soros reflexivity framing explains the dynamic: frontier lab valuations climbed from $4B to $18B, $60B, $180B, $380B, and toward $1T, and subsequent labs priced against those marks, making valuation itself the indicator. The same thread warns that when price becomes the signal, picking the right company can still fail if the price is off by 10x, and argues the next decade's VC winners will treat portfolio construction and position sizing as equal to deal access — asking how many companies depend on the last round rather than revenue, and what happens if the reflexive loop breaks next year instead of in five. — via 1

2. Macro Data: Housing, Jobs, and Inflation Regimes

1. US housing affordability has reached a record gap: buying a median-priced home now requires $126K in income against a median household income of $86K, a 47% shortfall that is near the historical high. Separately, US unemployment has remained below 5% for 60 consecutive months, the second-longest streak on record behind the 64-month run that began in the mid-1960s. — via 1 2
2. Ben Carlson's long-run inflation data shows the US annual inflation rate averaged just 0.2% from 1800 to 1940, with prices rising only 28% over 140 years, versus 3.7% annually since 1940 and cumulative price increases exceeding 2,200%. On household balance sheets, US net worth has grown 70% since end-2019, with the bottom 50% up 129%, the next 40% up 74%, the top 0.1% up 106%, the rest of the top 1% up 63%, and the next 9% up 57%. — via 1 2
3. Liz Ann Sonders flags continued deterioration in market breadth despite a prior-day bounce, with charts covering index drawdowns, the share of stocks outperforming, and the percentage of stocks at 4-week and 52-week highs by sector. Chicago Fed's Goolsbee called the setup a painful jobs-versus-inflation tradeoff imposed by a stagflation shock, saying the only path back is a hard one. The Chicago Fed National Activity Index fell to -0.04 in August from an upwardly revised +0.08, its 3-month moving average rose to +0.01 from -0.01, and 39 of 85 components contributed positively. The New York Fed household spending survey showed August increases in furniture, home repair, housing/apartment, and vacation spending, with declines in electronics, appliances, and vehicle purchases. — via 1 2 3 4 5 6 7 8

3. AI Economics, Tokenization, and Local Inference

1. Raoul Pal argues the price of intelligence is approaching zero: a $200/month subscription already delivers unlimited intelligence, and applying that intelligence to solve its own bottlenecks in energy and data-center efficiency makes the next round cheaper, creating a self-reinforcing acceleration. He also describes two AI-generated religions — the Goatse Gospel from a New Zealand experiment with two AIs conversing without intervention, which became a billion-dollar meme coin, and Crustafarianism, which emerged on Moltbook, an agent-only Reddit-like network, within roughly 48 hours when an agent built a church website while its owner slept — framing religion as a coordination tool that machines adopted almost immediately because they face the same problem. On tokenization, he argues the common framing of trading Apple on weekends, instant settlement, and 24/7 markets misses the larger story: everything becomes a token and the entire economy is rebuilt around it. — via 1 2
2. Marc Andreessen shares claims that Husky, a model-specific inference (MSI) engine, is up to 4.5x faster than Apple MLX, and that Underdog's Pareto frontier model runs at up to 730 tokens/sec on a MacBook, positioning local models as competitive with the cloud on both speed and capability, with a trial available in its app. — via 1
3. Meb Faber points out that clients in the wirehouse channel may be unable to buy a $1B ETF with a 10-year track record that beats 96% of peers, while the same channel blocks 351 ETFs but may offer hot private credit closed-end fund allocations — a situation he calls a "rough opportunity set." He also relays that Winton has filed for two managed futures ETFs, the Winton Managed Futures ETF and the Winton US Equities & Managed Futures ETF, with an effective date of December 4, 2026 and a 0.80% fee, running core managed futures (trend following) plus carry and other quant strategies, globally long/short with leverage across equities, commodities, FX, and fixed income via futures, forwards, and swaps. — via 1 2

4. Founder and Investor Practice

1. Raoul Pal describes himself as an infrequent trader and long-cycle macro investor who adds only when markets are deeply oversold, citing June 5 adds in ETH at 1650, SOL at 65, and SUI at 0.75; his 43 GMI positions average +96% with a 67% win rate, all in crypto and tech under the "index era" theme, and he says his 21-year win rate has held near 65% with his edge being time horizon and long-term trend stacking via the Everything Code framework. He says the key to investing is simplifying decisions: the GMI Compounding Machine for RV Pro lets users set standard-deviation multiples and dollar amounts for buys and sells, with the principle of buying weakness, selling strength, selling less than you buy, and otherwise just holding. He also calls James Easton a "raging bull" whose posts all use long time frames, arguing that if crypto is on an adoption curve, those entry points are fine over the long run even if not precise. — via 1 2 3
2. Elad Gil writes that founder burnout usually comes from doing things you feel obligated to do but hate, rather than things you love. Mohnish Pabrai relays that The Soul of a New Machine is essential reading for anyone in product development or a startup environment, and that holding great businesses like Walmart ($WMT) and the Nifty Fifty makes the hardest and most important task simply holding on. — via 1 2 3
3. Jason relays mentor-program guidance from Streamfog founder and LAUNCH accelerator alum KevBondzio: tie fundraising to achieved milestones rather than expected ones, because "we just crossed X ARR" and "we're about to cross" get very different responses. — via 1