Qualcomm HBC, AMD MEXT, and the Burry short — what the successor narrative gets wrong
The Narrative vs. What the Articles Actually Say
The most distinctive pattern across Korea semiconductor news over the last three days is the sudden acceleration of the 'post-HBM successor' narrative. Qualcomm launched HBC (High-Bandwidth Cache) as a challenge to NVIDIA. AMD acquired Japanese memory firm MEXT. Korean media began labeling HBC and HBF as 'the next generation after HBM.' Layer on Chinese memory makers entering the HBM market and Michael Burry expanding his short against Korean chipmakers, and the tape starts to price 'HBM peak' — which is exactly what shook the KOSPI semi complex on June 30 and July 1.
The problem: the narrative contradicts what the underlying articles actually describe.
Take the Three 'Successors' Individually
Qualcomm HBC. The article's own conclusion states "Samsung Electronics and SK Hynix memory demand is expected to rise." HBC is a cache layer that sits beside the GPU. The DRAM cells that fill that cache still come from Korean suppliers.
AMD's MEXT acquisition. The Korean coverage explicitly says the deal "signals memory hierarchy restructuring, not HBM replacement." What AMD is buying is bridge IP between CPU and HBM — not HBM itself.
HBC and HBF as 'successors.' Cache (HBC) and flash (HBF) are layers that sit above and below HBM in the memory hierarchy. They are not substitutes; they are additional layers in the stack. And Korean suppliers end up filling most of them.
The word 'successor' is marketing nomenclature. The substance is a memory hierarchy that keeps adding layers — each one addressable by Samsung and SK Hynix.
The Fundamentals Are Pointing the Other Way
Line up the numbers that dropped in the same three-day window:
- June total exports: crossed $100B for the first time in history, semiconductors as primary driver
- May semi exports: $29.4B, up 154.3% YoY. March $24.9B (+138%) → April $25.2B (+158%) → May $29.4B — accelerating
- SEMI 2026 memory equipment CAPEX: $52B, +29% YoY, first-ever 80T won milestone
- DDR5 16Gb spot (July 2): $46.67
- Jefferies Q3 memory price outlook: +50%, with stability not expected until 2028
- SK Hynix P&T7: ~200 HBM4 testers locked in (~$300M)
- Samsung Chungcheong investment: 140T won ($100B), designating Onyang/Cheonan as global HBM hub
Exports, equipment capex, spot prices, tester orders, cluster investment — five separate gauges, all pointing up. Those are the numbers Burry is shorting against.
What Burry's Thesis Misses on Timing
Burry's case is 'the AI narrative has peaked, and Korea's capex will produce a glut.' The numbers above show that capex hasn't even reached run-rate. Samsung's 140T won Chungcheong build-out and SK Hynix's Yongin Phase 2 are multi-year ramps. The proof: SK Hynix awarded a 500B won wastewater treatment contract to Techros this week — wastewater infrastructure typically precedes wafer-in by three to four years.
What Burry calls a 'peak' is closer to groundbreaking. The real short-side risk is being forced to cover while Jefferies' +50% Q3 price move is playing out.
Google's Taiwan Pivot Is a Different Axis
One apparent negative signal is Google shifting its AI-chip partnership toward Taiwan. Read carefully, though, that's foundry — deepening ties with TSMC — not memory sourcing. As long as HBM bottlenecks sit with SK Hynix, Google's move is really about locking in CoWoS packaging capacity. It's orthogonal to the memory hierarchy story.
The Micron-vs-Korea valuation debate has the same shape. Yes, Samsung and SK Hynix are trading below Micron on comparable metrics. But this week's news is a catalyst set for closing that gap: Jefferies' +50% Q3 call, record equipment investment, and the reframing of HBC/HBF/MEXT as Korean-supply expansion rather than replacement. Three re-rating triggers, not one.
Risks Named Explicitly
Fair-weather short-side risks worth flagging: (1) the US consumer class action alleging Samsung/SK Hynix/Micron colluded on DRAM prices is a near-term headline risk. (2) Chinese HBM entry could pressure premium margins over a multi-year horizon. (3) Foreign firms aggressively recruiting Korean engineers signals R&D drain. (4) Power and water constraints in the southwest cluster could delay capex ramp.
None of the four flows into 2026 H2 numbers. What does flow through: Q3 memory prices, HBM4 ramp at P&T7, the +154% May export trajectory, and the 200-tester order.
Bottom Line
This week presents a rare gap between nomenclature and substance. The tape reads 'HBC/HBF/MEXT successors' as a replacement threat; the article bodies describe them as Korean-supplied layers being added to the stack. Burry is betting on the confusion. Silicon Nexus view: this week's pullback in SK Hynix (000660) and Samsung Electronics (005930) is a narrative trade, not a fundamentals trade.
Key Sources: - Memory Equipment Investment Hits Record $52B in 2026 (TheElec, 2026-06-30) - SK Hynix Locks in ~200 HBM4 Testers for P&T7 Launch (TheElec, 2026-06-30) - SK Hynix taps Techros for Yongin Phase 2 wastewater system (TheElec, 2026-06-30) - Qualcomm HBC to challenge NVIDIA; Samsung, SK Hynix memory demand to rise (Google News, 2026-07-01) - Jefferies: Memory Prices to Rise Up to 50% in Q3 (Google News, 2026-06-29) - plus 8 more
If this analysis was helpful · ☕ Support Us · ✈️ Telegram