In this week’s video, I walk through what ended up a quiet summer week with the S&P down 61 basis points, the Qs off 1.5%, and IWM posting a fourth consecutive week of sub-1% absolute moves where falling volatility, better small-cap and equal-weight performance, and improving breadth all point to earnings strength rather than deterioration. With almost a third of the index reported, 86% have beaten estimates against a 10-year average of 76%, an 8-to-1 beat-to-miss ratio, and margins are still grinding higher (roughly 14.4% ex-Google). Estimate revisions, the dominant factor since the iPhone came out, are telling you the market is fine.
The big signal of the week came from China. Beijing unleashed broad state support to halt the tech sell-off state funds buying nearly 9 billion dollars, state-backed insurers adding equity, listing standards changing all because capital markets have become critical to its AI ambitions. This is a two-horse race with two different strategies: China leaning open-source and abundance, the U.S. leaning capitalism. DeepSeek’s $7 billion raise gave the government the only outside voting rights. Moonshot’s Kimi K3 is seeking a $50 billion valuation while Anthropic is a two-to-three-trillion-dollar company. Capital is the constraint, and the issuance wave, Google’s $80 billion, the SpaceX unlock, widening hyperscaler spreads is exactly why the tape is chopping in thin summer liquidity.
On Google, the negative free cash flow print is the noise. The backlog is the whole ballgame: cloud up 82%, backlog from $106 billion to $514 billion in a single year, and they are still supply constrained. That’s good for memory, chips, optical and photonics. Short-term noise versus long-term flow and I’m still buying the consolidation.
Timestamps
- (00:00–02:20) Setup: a calm summer week, but earnings are the story for the rest of the year. China came to the market, Google’s nitty-gritty beyond the free-cash-flow headline, a new Mosaic prompt, Kimi K3, the Vera Rubin report, oil, and the Clarity Act.
- (02:21–06:33) The tape: S&P −61bp, Qs −1.5%, IWM’s fourth straight sub-1% week, tech momentum +8% after a gruesome four weeks (helped mostly by Google selling off), thematic portfolio +1.5% with its best week versus hyperscalers since late March. Breadth at its best since July 10th, still above a rising 200-day. Bought last week, will keep buying the consolidation because when people start saying “Anthropic is done,” that’s when I get more interested.
- (06:33–09:32) Earnings and margins: 86% beats, 3% in line, 11% misses, an 8-to-1 ratio versus a 76% ten-year average. S&P margins still progressing higher, ~14.4% ex-Google. Then China unleashes broad state support to halt the tech sell-off, because with house prices falling every month they need some asset class going higher.
- (09:32–12:44) The capital race: DeepSeek’s unusual $7 billion structure, fresh raises across Chinese AI developers, Moonshot’s $50 billion pre-IPO round. Google’s $80 billion, the SpaceX unlock, hyperscaler bonds declining and hitting global bond markets. All of this liquidity being soaked up by issuance in the thinnest month of the year. Eric Schmidt: cash, not energy, is the real limit.
- (12:44–15:31) Google, honestly: revenue +24% aggregate, search +17%, cloud +82%, YouTube ads +13%. Backlog up another $50 billion with just over half — roughly $260 billion expected to be recognized over the next 12 months. Operating cash flow $39 billion against $45 billion of capex. They are not managing toward near-term positive free cash flow, and they don’t need to.
- (15:31–21:53) The bear case and the odds: Bob Elliot’s $30 billion of excess cloud earnings for $140 billion of capex, and why you can’t run that math without the backlog going from $106 billion to $514 billion. Fable 5 puts 70–75% odds the capex program proves value accretive over three to five years, with three real risks, chip obsolescence, off-balance-sheet commitments, and backlog concentration. Then the drawdown math: 15% annualized since 2019 and six separate rolling three-month periods down double digits.
- (21:53–27:05) The Mosaic method: three podcasts into one context window, and the overlap is model abundance plus compute scarcity. Intelligence can be copied; compute cannot. Jevons paradox, enterprise adoption moving from token maxing to token efficiency, and routing as the inevitable winner. AI as the conductor of the symphony rather than a serial processor listening one at a time.
- (27:05–31:34) Kimi K3 and QE for the mind: K3 is not a small model I can’t run it, nobody at home can, and Moonshot had to pause subscriptions because they ran out of capacity. Potentially negative for OpenAI and Anthropic, net positive for essentially every other company in the world. Under ZIRP you borrowed at 2% to buy back stock; now a 5% expense reduction is a 20% EPS jump. Low-margin, high-friction businesses are the ones with the most to gain.
- (31:34–37:32) The evidence check: corporate profits to nominal GDP at all-time highs, and recessions arrive after margins and profits are already falling. Estimate revisions the dominant factor since 2010 say the market is fine. GPU availability tightening back up, Vera Rubin in full production, and why I expect a step-up function in optical and photonics. Marvell is still my biggest position; added AAOI this week.
- (37:32–40:36) Token index and headwinds: compute is becoming a financial market, demand fragments, enterprises settle around orchestration and routing. Cheaper models pressure the frontier but expand the total market the pie gets bigger faster. Then the honest headwind list: cyber, data-center backlash, power, cooling, memory bottlenecks, open-source competition. That friction is what keeps this from becoming a bubble.
- (40:36–44:15) Oil, CPI, and the Fed: December crude back near the highs at 78 versus a 68 midpoint, and everyone panicking. Inflation swaps and two-year breakevens haven’t budged smart money isn’t falling for it again. First non-zero CPI comparison doesn’t land until mid-September. Meanwhile wages and participation say the labor market isn’t strong, heading into the least predictable Fed meeting in years.
- (44:15–49:12) Clarity Act and crypto: one-yard line in the Senate per Bessent, but odds back in the 30s. Why agentic AI and blockchain are inseparable think 50 First Dates, where the context window remembers nothing and consumer agents need massive persistent memory to work. Once they do, crypto volumes explode. Bitcoin stable around 64,000, still a bear market until the 200-day breaks, and Dogecoin below its 20-day for a record 67 days retail energy is still nonexistent.
