Treasury Doubles Long-End Buybacks as Marvell Takes a Slice From Broadcom
US Stocks · Options · News · Views
Breadth Held; SOX Did Not
The tape on Wednesday, August 19, was still heavy, and Nasdaq was the unique decline. In the SOX basket, 26 names fell and only four rose. Under the S&P surface the tape was fine: 288 up, 214 down, seven of eleven sectors higher, and the equal-weight index up more than 1%. Health care jumped 3.51% and printed a new high. The index looked like tech was the drag. Internally, most groups had a bid. The weak line was semiconductors. Mega-cap software barely moved; leave it.
One name explained the health-care high, not the SOX tape. MRNA ripped as much as about 176% after the personalized mRNA melanoma vaccine with Merck hit its main Phase 3 endpoints. That stock has not been a coverage name since the pandemic. No fundamental or technical map here. It tells you why health care could make a high. It does not tell you why SOX went 26-to-4.
The July FOMC minutes were a sideshow. Most officials still preferred to hold; a minority leaned toward a hike. The vote was 9-3 to keep the funds rate at 3.50% to 3.75%. The new item was meeting frequency. Chair Warsh floated six scheduled meetings a year, roughly every two months, so more data can accumulate between decisions. No vote. The rest of 2026 stays on the current calendar. Five committees are working on framework questions into year-end. That is not a trade yet. The minutes were a touch hawkish. They were not the driver.
What changed the tape was Treasury stepping on the long end. The 30-year had already tagged about 5.3% to 5.4% in the prior sessions, near the highest since 2007. Equities got the valuation hit from yields first, then the semiconductor location second.
Buybacks at Least Double; Squeeze the Shorts First
Treasury will lift liquidity-support buybacks in 10- to 20-year and 20- to 30-year nominal coupons from a $2 billion cap to at least $4 billion per operation, effective September 9 through November 4, the rest of this refunding quarter. The next size decision waits for the November 4 refunding. Simple addition across the remaining operations this quarter gets the market to an extra tens-of-billions number. That annual total is not an official promise. Yields fell on the news. Equity futures jumped.
Where does the cash come from? TGA general-account cash that has not been spent yet, or proceeds from new issuance. Left pocket, right pocket. Existing cash or new borrowing buys older long bonds. The deficit is still there, so net debt does not shrink. The law allows it. Treasury cannot create reserves the way the Fed can. Each dollar bought back has to be matched, in principle, by a dollar of new issuance or a dollar of tax receipts. The budget is still in deficit, so cash pulled from taxes in the short run likely gets replaced by more issuance later. This is not QE. It can drain float and put a bid in for a while. It does not create base money.
Speculative shorts plus higher inflation expectations make a one-way yield spike easy. Buybacks are a regular liquidity tool and a signal: endless shorting of long bonds will get a response. Covering showed up. TLT gained 1.67% on the session. That is not a bull market. Debt is heading toward $40 trillion, inflation has not fully cooled, and hike odds have not been pulled. Until rate expectations move down and the debt path slows, the long bond lacks a structural bid. Bounces are allowed. A long-term allocation still looks like a leftover.
Three paths. If yields trend lower, growth multiples and AI funding costs ease together, and stocks benefit. If this is only a short squeeze and the long end turns back up, equities have to reprice again, and Bessent can still size up — $4 billion is not a ceiling. That is expectation management, not a repaired fundamental. The middle case is a high-range chop in long yields: stocks get a short-term relief bid, and SOX still has to be handled on its own levels. More issuance, unrestrained spending, or a slower tax base would lift the long end again.
AMD Still Rich; SOXX Stuck in 489 to 532
Semiconductors were the actual damage. SOX had already dumped the rebound and lost 532 the prior session. Nothing was repaired today. AMD dropped 3.71% to around 466. The 2027 forward median is about 430, so the stock is still above that median. 2028 looks less stretched, but that is later. SOX has been under pressure for several sessions. The next support has not broken. The tape already looks ugly.
461 to 476 is a tiny shelf. Ignore it. The zone that matters is 427 to 453, just under the 100-day, a medium support. Lose that and 400 does not do much. The next real area is 334 to 366, near the 200-day and a more reasonable multiple. 427 has not broken. 334 to 366 is a map, not a forecast. Getting the stock back into a strong regime still needs 506 to 528, which is distant. The fundamental kicker is a late-2027 earnings ramp, and that is still far away. SOX is weak and the swings are small. Anyone still constructive has to wait for a location. There is not much of a swing to trade.
SOXX is still 489 to 532. Volume was a bit lighter than Tuesday. Treat it as a range. 489 is the lower edge. 532 is the level just lost. Until a high-volume stop-decline prints, this is not a new long. Buy it and then lose 489, and the swing plan is done. Quiet mega-caps are not cover for SOX.
MRVL Ties Up Google; AVGO Loses a Slice
MRVL jumped on an expanded custom-chip partnership with Google. Marvell issued Google a warrant at $206.58, covering up to about 58.97 million shares, or roughly $12.2 billion if fully exercised. About 1.36 million shares vest in equal quarterly slices in the first year. The rest vest from fiscal Q3 2027 through fiscal 2033: one tranche for every $500 million of custom-product revenue, 240 tranches in total. The work sits around the TPU stack: AI inference accelerators, storage and network controllers, memory interfaces, and near-memory compute. The warrant dilutes EPS, but vesting is tied to purchases. Google has to buy more silicon to earn more shares. For Marvell, that is more visible revenue from 2027 through 2033. If other demand slips, this slice can offset part of the dilution worry.
Google’s main TPU partner has been Broadcom. This is a second leg, not an eviction. Broadcom is still in. A single supplier can raise price with little pushback. Two suppliers make the long-term negotiation cleaner. Do not read the filing as Broadcom being kicked out.
Price still has to clear the forward. The January 2027 fiscal-year upper bound is about 165, and that year is almost done. January 2028 tops out around 247. After-hours already tagged about 240. There is not much room left, and technical resistance is right there. 253 to 272 can still chop. 278 to 318 looks hard. Watch for stalling. The stock is no longer cheap. The deal bid and the multiple are sitting on top of each other.
The slice that accrued to Marvell came off AVGO, down 4.61%. Broadcom’s October 2026 year ends sooner. On October 2027, the same valuation band is about 572 to 763 with a 668 median. Five-year growth rates are similar. The gap is size: Broadcom around $1.72 trillion, Marvell around $200 billion, an order of magnitude, the NVDA versus AMD problem. In a messy tape it is easier to chase the smaller cap than to lift the giant. Broadcom at 324 to 365 is still a reasonable seat if growth is not cut. Lose 324 and the supports underneath are weak — a breakdown, not a dip. If the Street already has slower growth in the model, price will lag the research. Trading Broadcom is closer to trading Nvidia: core exposure, large cap, wait for the location. Marvell is the beta side, and today’s print already prepaid some of the revenue visibility.