Record exports, record imports, and then a GDP miss 24 hours later. One of my assumptions about how these numbers fit together is wrong.

I have been running China as a single macro bet for two years, and the data sequence on July 14 and 15 broke something I thought I understood.

On July 14, customs reported June exports at $412.4 billion, up 27% year over year, an all time record. Imports hit $286.8 billion, also a record, up 36%. The arithmetic is clean: $412.4 billion minus $286.8 billion equals about a $125.6 billion surplus, the second largest on record, per Bloomberg and CNBC. My priors said record exports plus record imports in the same month equals an economy accelerating on both ends of the ledger. That was the bet.

Then July 15. Q2 GDP 4.3%, below 4.5% consensus, down from 5.0% in Q1. Same statistical system, 24 hours later, saying deceleration. NBS via Reuters.

I keep running the same reconciliation. First half chip exports nearly doubled to $177 billion, up 96%. That is a real industrial performance. But here is the honest caveat I cannot price: much of that jump is DRAM and NAND price inflation, not volume. I have not found a clean price versus volume split in the public releases, so I cannot tell how much of that $177 billion is more silicon and how much is the same units at higher prices. The record export number could be mostly a price effect.

The equity market read it as sector, not macro, on July 14. Shanghai closed up 1.36% at 3,967, Shenzhen up 2.77%. The leadership was optical module makers. Zhongji Innolight rose 6.86%, Eoptolink about 11%. Not broad beta. Narrow leadership in the hardware layer feeding global data center buildouts.

My working claim, and I want readers to shoot at it: the export record is the global AI capex cycle paying record prices to one slice of Chinese industry, and that slice can compound even if headline GDP keeps disappointing because its demand is overseas, not domestic. The test is falsifiable. If memory prices normalize and chip export values fall back while GDP stays soft, the record was a price mirage and my claim fails. If export values hold as prices cool, the volume was real.

The second unknown that breaks my prior: I do not know whether those record imports were components bound for re export or genuine domestic absorption. That is exactly the split I need to square the ledger with the GDP print, and I do not have it.

I am unwinding the single China exposure this week. The export facing AI hardware complex and the domestic demand economy just printed opposite directions in 24 hours. They may deserve opposite weights. That is not a conclusion. That is a confession that my old framework was too coarse, and the July 14 and 15 sequence caught it.

The sector versus macro split is why this stays its own small sleeve rather than a bigger China position, but sizing even that sleeve runs into the same mapping problem. The two optical names that led the July 14 session, Zhongji Innolight and Eoptolink, are A share listings, and KWEB holds zero A shares so it carries neither of them, while CQQQ applies a limited 25% A share inclusion factor. CNQQ's published May 2026 top holdings include Zhongji Innolight at roughly 6.4% and Eoptolink at about 2%, alongside its broader split of roughly 58% A shares and 42% Hong Kong listings as of end 2025. It is a small fund with short live history, launched September 2025, so liquidity and tracking are real questions I have not answered for myself yet.