In this week’s video, I separate signal from noise as the “AI bubble” chorus and rate panic connect, starting with Ray Dalio’s warning that a burst may be near. Speculative, capacity and credit bubbles all require supply to get ahead of demand. That isn’t happening. In past cycles humans were both the supply and the demand. Now the demand side is agents: thousands today, billions next year, eventually trillions, all consuming tokens. Demand is still greater than supply, Nvidia trades nowhere near dot-com multiples, and much of the new borrowing is asset-backed against compute.
The true bubble that does exist is in the old economy: companies that only survived because rates stayed near zero. The economy has changed and is why only 18% of utilities and 12% of autos sit above their 200-day moving averages while tech and financials are above 70%. Since the iPhone, alpha has meant long the digital economy, short everything kept alive by debt. Pandemic, the fastest hiking cycle since the 1980s, a historic bank run and tariffs all failed to end it.
Alpha is now concentrated. Energy and tech are the only sectors beating the S&P this year, and 71% of my agentic AI names in the S&P 1500 have outperformed since ChatGPT launched, versus 24% of the universe. The next legs are compute and crypto. Stripe, growing revenue 41% with a fraction of Citigroup’s headcount, is the template for why banks face their own “SaaS apocalypse” moment, and why tokenization and privacy become investable themes.
Timestamps
- (00:00–04:50) Intro: We are in an exponential era, and my own shift from macro to the digital economy began in 2012–2013, when I stopped traveling to China and went to Silicon Valley. I lay out the signal, alpha and agency framework that runs through every video.
- (04:50–09:30) Bubble debate: Ray Dalio says an AI bubble may burst, but a speculative, capacity or credit bubble requires supply to outrun demand, and agents are now the demand side. An AI-built risk scorecard using my weekly data shows demand strong, compute scarce and capital on the watch list.
- (09:30–15:20) The real bubble: The bubble sits in rate-dependent old-economy companies, which is why only 18% of utilities and 12% of autos are above their 200-day moving averages while tech and financials are above 70%. The Morgan Stanley leverage index, CCCs vs. high yield, Europe vs. the U.S. since the iPhone, and the longest gaps between all-time highs all point the same way.
- (15:20–24:10) Exponential math: Kurzweil’s curve shows we are at the inflection point, and Aaron Levie explains why personal agents and agent swarms will need far more compute. Tokens are the new oil, but agents don’t grow with nominal GDP, and Henry McVey now sees a broader, more durable and more inflationary investment cycle.
- (24:10–31:10) Where alpha lives: The trade moved from Nvidia to memory, with Micron going from 100 to 1,400, and the next legs are compute and crypto. The equal-weight S&P just triggered a MACD buy signal that has been positive six months later 22 out of 22 times, and 71% of my agentic names in the S&P 1500 have beaten the index versus 24% of the universe.
- (31:10–40:00) Bank apocalypse: Stripe is building the rails for agentic commerce through Privy, Bridge and the OpenRouter acquisition, and grew revenue 41% in the first half. Its valuation per employee is 15 times Citigroup’s, which mirrors the gap between Anthropic and Google.
- (40:00–41:50) Technicals: Lumentum says it is sold out through 2029, and my agentic portfolio is forming a cup-and-handle ahead of a likely breakout. The 10-year yield shows a triple RSI divergence while the MOVE index and bonds put in an outside reversal week.
- (41:50–44:30) Humanoids and labor: Consumer agents bring humanoids and FSD closer, and 2027 could be the year agent output exceeds the work of four to five billion people. At the same time, healthcare hiring is slowing and jobs are harder to get, which pressures wages and widens the labor vs. capital gap.
- (44:30–49:10) Market structure: Every company now falls into one of three buckets: adapters, winners or the disrupted, which forces portfolio managers to run more concentrated books. Tokenization could move trading from 32.5 hours to 168 hours a week, raising LTCM-style risk for levered multi-strats and options traders.
- (49:10–52:30) Privacy and crypto: Vitalik Buterin argues that agentic AI turns privacy into a control problem, which makes privacy an investable theme through names like Zcash and Near. Asia is moving fast, with Samsung adding USDC to its wallet and Standard Chartered offering crypto custody in Singapore.
