Visit ai.22vresearch.com and visser-labs.com for more information
Email Mark Whaling mwhaling@22vresearch.com
In this week's video, I begin with Stan Druckenmiller's op-ed which was framed as an attack on Bessent, but is really an op-ed on the situation we have created over decades through the spending in the old aging system. I agree with him on what should happen. We don't make money on what should happen. At 240 to 250% market cap to GDP, heading toward 300%, with nominal GDP at 6.5% and climbing, ten-year yields arguably belong above 8%. They aren't there and letting them go there is a risk, and the August pattern of behavior tells you why Bessen has no stomach for the risk: yen intervention, refunding language changes, a buyback framework, larger buybacks, and now open discussion of tapping the TGA.
The deeper point is that this pressure is being generated by AI. Hyperscaler issuance is already 9% of all IG supply and has doubled in a year, crowding out Treasury demand. Warsh talked hawkish at Jackson Hole and the long end backed up anyway, because the bid for capital is coming from the AI build-out, not from inflation expectations.
That leads to the two constraints I keep returning to. Compute is scarce, and scarce compute does not commoditize, which is why OpenAI and Anthropic keep winning: highest margins buy the most compute, which trains the best models, which earns the most revenue. And AI compresses time, which forces finance onto programmable rails. Japan and Korea are already moving on tokenization. Stripe is buying toward it. Assets are becoming software.
In a week when Solana ran 46% month-to-date and the tokenized index broke out, the only thing macro people asked me about was gold. Gold is fine. It will not be the fastest horse in this race.
Timestamps
• (00:00–05:48) Week setup and the Druckenmiller op-ed: the reaction was off base. The real issue is Congress, the K-shaped economy, and a 240–250% market cap to GDP economy heading toward 300%.
• (05:48–08:26) The debt endgame runs on human time. Growing out of it is unlikely on a linear path, but it is the same bet the hyperscalers are making with debt: revenues arriving faster than the debt does damage.
• (08:26–13:18) The yield suppression pattern: yen intervention, refunding language, buyback framework, bigger buybacks, possible TGA use. Steno Larsen on repo and pushing debt to the front end. The basis trade and stablecoins as engineered demand for long-term Treasuries. Warsh's Jackson Hole comments moved hike odds from 35% to 60% without saying anything new.
• (13:18–17:35) Crowding out and concentration: hyperscalers now 9% of all IG supply, doubled in a year. Broadcom's $70 billion deal and widening single-name CDS. Index CDX isn't confirming, exactly like equity vol versus index vol. Credit isn't telling you a story here, concentration is. Dollar/yen weakens to 160.
• (17:35–21:07) Dylan Patel and Dwarkesh Patel on compute centralization. OpenAI and Anthropic are treated as having already won because their margins let them outbid everyone for compute. Open source is not discussed as competition. The U.S. has pulled well ahead on AI compute deployment.
• (21:07–28:01) Where I disagree with their $11 trillion capex and sovereign shock framing: AI also destroys debt through defaults and closures, and lets startups scale without debt or headcount. AI creates a higher return clock for capital.
• (28:01–31:22) Agentic tooling in practice, plus why the deficit story keeps circling back to rates. My read is Warsh doesn't raise in September despite wanting the inflation-fighter legacy, because the timing is wrong with the long end pressing and the Treasury working the other side. Nobody risks the long end going into the midterms.
• (31:22–38:29) Roman Yampolskiy reaches my conclusions from a completely different direction. AI compresses scientific and technological time until progress runs on a different clock from human society.
• (38:29–47:19) Bitcoin's 200-day slope turned up after more than 100 days pointing down, only the fourth such instance, and prior lows held in every case. Meanwhile the only question I got all week was about gold.
• (47:19–56:21) Tom Lee on why agents cannot function without smart contracts, and why AI forces finance from human precision to programmable constraint. Ethereum as the institutional settlement layer, Solana as 24/7 market speed, Bitcoin as protection against time debasement. Tokenization turns assets into software: a stock, a bond, a royalty stream, a membership all become programmable objects.
• (56:21–1:03:22) Nvidia's blowout, with next-year growth guided to 70% against 44% expectations, and the multiple compressing anyway. Breadth is still weak with 29% above the 50-day.