Memory has always been a cycle: shortages push prices and margins to a peak, makers build, and prices fall. AI has pushed this one further than any before. Micron's quarter to early September 2026 brought revenue of $54.2B, against $11.3B a year earlier, at a GAAP gross margin of 86.8%, against 44.7%, and it guided the next quarter to $61.5B at about 86%. Its previous record, at the last memory peak in 2018, was 61.0%. SK hynix is reported to have its HBM sold out years ahead, and price trackers report conventional DRAM contract prices still rising. What is new is how the memory is sold. Micron's 2026 annual report says most of its strategic customer agreements are multi-year, take-or-pay, backed by customer deposits, and priced within minimum and maximum bands, and that gross margins at the floor prices would be meaningfully above its peak quarterly margins in any past cycle. The claim here is that this holds: through Micron's fiscal 2027 its gross margin stays above the old 61% record, because the floors bind before the cycle does. That supports MenFem's bullish SK Hynix, Micron and Samsung calls, and it is bad news for MenFem's calls that inference gets much cheaper, since memory is the input getting dearer. Conviction is LOW. These are the highest margins memory has ever earned, and the usual answer to record margins is new supply: Chinese makers such as CXMT are adding capacity, and memory shares reportedly fell between about 11% and 26% in five days in mid-October on fears of oversupply by 2029. Micron itself says the contract structure still subjects it to risk, and a floor is only as good as a customer's willingness to honour it in a glut. BULLISH on the thesis, conviction LOW, one year. This view was formed by MenFem's research process at Connor's request and adopted by him.
Optics takes the new links, copper keeps the short ones: both grow through 2027
Co-packaged optics is arriving on schedule while copper cables inside the rack keep growing. Bullish on the thesis at medium conviction: the rack is big enough for both through 2027, and neither side's story of taking the other's place holds yet.
Coherent PhotonLink (CPO/NPO): >$100M in the March 2027 quarter, $250M a quarter before end-2027 (4 Nov slides).
Lumentum: >$5B of CPO/NPO commitments; Marvell to buy Lumentum (all-stock, ~$115B, 2 Nov), citing copper's limits.