GS Basics Gross Concerns. Operation Buyback. Token Cost Collapse

Asia: Asia traded well overnight, led by a pretty extraordinary ~6% rip in KOSPI after yesterday’s selloff. Some of that is just the mirror image of the previous move… lower US long end yields removed an obvious pressure point and SK Hynix’s KRW40tn buyback gave Korea a pretty enormous company specific catalyst. Given the weight of Hynix/Samsung in the index, that quickly becomes an index event. China added more incremental property support overnight, while EM broadly traded well with the dollar weaker.

Gross / AI: Very violent market for gross the last few days. I’m surprised by the magnitude… GS prime data looks more subdued (neutralish momentum...people added more nets than gross). Last 48 hours there has clearly been some position pain with GSPUMOMO down nearly -7% over the period. I'm surprised...we just had a big de-risking event in July...so while people clearly still have exposure, we are a long way from peak positioning (at least on our data). The harder question is fundamentals. The frontier lab debate has shifted towards whether growth can continue to match already extremely elevated expectations. At the same time, model economics keep improving stupidly fast. Qwen3.8 is another example… increasingly useful intelligence from increasingly small/cheap models. Useful intelligence per dollar is exploding and competition is multiplying. That is very bullish for AI adoption and the broad market, but less obviously bullish for everyone selling tokens/model access. Moats get harder to defend. Same argument for parts of the picks and shovels chain… think whole space keeps de-rating as EPS moves up...defensible toward low end of the valuation range but difficult to chase into 2027/28 uncertainty.

NVDA: Nvidia next Wednesday probably prints numbers that look incredible on the headline. That said I see this as more of a mixed catalyst. The questions increasingly move from demand to financing, margins and how the economics get distributed across the chain. Memory inflation/supply is part of that margin discussion, so you could easily get an emphatic “AI spend is amazing” message while some parts of the memory chain digest the details less well. Think the more supportive catalyst come with the heavy September conference calendar.

Treasury / USD: Treasury doubled the maximum size of long end liquidity buybacks from $2bn to at least $4bn per operation in the 10-20yr and 20-30yr sectors through early November. The words “at least” matter… this doesn’t read like a hard ceiling and clearly opens the door to doing more if required. Mechanically this isn’t QE and they still have to fund whatever they buy

back. There also isn’t some requirement that every operation is perfectly duration neutral… Treasury manages duration across the whole funding program. But the signal seems pretty clear: they are willing to use buybacks and issuance composition more actively to stabilize the long end. Call it quasi-Twist, Twist-lite, whatever… the label matters less than the reaction function. I actually think the more lasting implication may be for the dollar/gold than rates. Rates can eventually re-clear higher because the underlying supply hasn’t disappeared. But if the market starts believing that sufficiently aggressive long end tightening will be met with larger buybacks, shorter issuance or other liability-management tools, some of that adjustment has to migrate elsewhere. A flatter curve and weaker dollar feels like a pretty logical expression of that. For equities/gold I think that is pretty bullish. The important information yesterday wasn’t whether $4bn itself changes the supply/demand balance… somewhat marginal. It was learning that the response to pressure in the long end can include changing the way Treasury manages duration. That strikes me as more important than the initial move in bonds.

Risk: We’re trading as if the fast money crowd has aggressively re-risked back into momentum and semis. Think illiquid August tape, although leverage in semis space has come down there is still a lot in the system and large moves down drive forced supply. Feels like violent moves both ways that look much more fundamental than they probably are. I…