The Post-HBM Fork — SK Hynix Opens HBF as an OCP Standard, Samsung Unveils 8x zHBM, and 2027 HBM Is Already Sold Out
The next-generation standard forked before the current one shipped — closed zHBM vs open HBF, on the same day.
Two Post-HBM Maps, Drawn in the Same Week
During the first week of August 2026, Korea's two memory champions used FMS 2026 and an OCP working group to publish diverging 'post-HBM' roadmaps on the same day. SK Hynix and SanDisk released the first standardized spec for HBF (High Bandwidth Flash) through OCP as an open standard — up to 512GB per stack, 3TB/s peak bandwidth. In the same week, Samsung unveiled its own zHBM, claiming 8x performance over the current generation, and paired it with z-NAND-O to sketch a closed, vertically-integrated AI memory stack.
The same day both firms proclaimed the next generation, another headline confirmed that 2027 DRAM and HBM capacity is already fully allocated (Google News, 2026-08-04). The successor standard forked before the current generation had even shipped. That is the real signal of the week.
Open vs. Closed: the Nature of the Standard War Changed
The axis of competition in the HBM era was 'who ramps 12-hi first.' JEDEC was common; differentiation lived in yield, bonding and TSV. SK Hynix's HBF move rewrites that axis. By pinning a NAND-based, 512GB-per-stack bandwidth tier as an OCP open standard, SK Hynix is trying to reshape the entire memory hierarchy, not just extend its HBM lead (THELEC, 2026-08-04). The open route has obvious advantages: hyperscalers (Google, Meta, MS) and TSMC — who owns CoWoS — can commit faster, and the standard rides directly on the ASIC custom-silicon wave already underway.
Samsung's zHBM is the opposite bet. Combined with z-NAND-O, it presents a stack that only closes inside Samsung's own fabs and packaging lines. The 8x performance headline is powerful, but it is a demo of what Samsung can do end-to-end, not an open spec (THELEC, 2026-08-05). With these two paths visibly forked, the market now has to re-compute which standard lands first on hyperscaler RFQs.
The Cycle Is Still Tight — and That Is What Accelerated the Fork
Korean customs data explains the timing. Semi exports hit a record $33.6B in June, up 173.9% YoY. April $25.2B → May $29.4B → June $33.6B: three consecutive record months. DDR5 16Gb spot was $51.33 as of Aug 5, with strength confirmed across HBM, GDDR7 and DDR4 (Money Today, 2026-08-04). Samsung posted a record KRW 89.5T Q2 operating profit; combined H1 for Samsung + SK Hynix approached KRW 250T.
Paradoxically, the sold-out book is exactly what allowed the standard fork to happen now. If 2027 volumes are already committed, hyperscalers have already moved past 2027 in their planning; the negotiation table is 2028+. Whoever lays a spec on that table first sets BOM share for the next five years. SK Hynix's open-standard play and Samsung's closed 8x demo are two different tactics aimed at the same negotiation window.
Why the Stocks Sold Off -8% — Standard Risk Just Entered Valuation
Despite the records, Samsung and SK Hynix printed -8% intraweek, and roughly KRW 870B exited 3x leveraged semi ETFs into Micron and SK Hynix ADRs (FN News, 2026-08-05). Goldman Sachs raised Samsung's target to KRW 490,000 and added it to APAC top picks (Edaily, 2026-08-03), but investors were not celebrating the print. They were re-pricing a new risk: if the post-HBM standard forks, CapEx doubles. An HBF line and a zHBM line require different equipment, different bonding, different test. The 2027-2028 CapEx bar consensus is now visibly stale.
The Fork Reaches Down Into Equipment and Materials
The HBF/zHBM fork ripples into Korean equipment and materials. Yujin Tech completed sub-10nm TiN ALD equipment for DRAM (THELEC, 2026-08-04) and HachTech began supplying temperature sensor ICs, ending Texas Instruments' monopoly. LB Semiconn entered volume for Qualcomm PMICs; DB HiTek printed +43% Q2 operating profit at 100% foundry utilization. A fork in the memory standard means the addressable market for Korean equipment doubles — but which side of the fork a supplier ends up on will determine the next 12 months' winners.
PM View: Two Payout Tables
1) Open-standard (HBF) wins. SK Hynix + SanDisk lock hyperscaler RFQs and integrate into TSMC's CoWoS roadmap. Equipment/materials tied to NAND stack, bonding and OCP-compliant tooling benefit. 2) Closed-stack (zHBM) wins. Samsung's real weapon — foundry + memory + z-NAND-O vertical integration — finally activates commercially, triggering an HBM market-share reversal.
The scenarios are not exclusive. If hyperscalers insist on dual sourcing, both sides win. What matters this week is that the fork became visible to public markets for the first time. That is a moment worth logging.
Key Sources: - SK Hynix, SanDisk publish first HBF standard spec (THELEC, 2026-08-04) - Samsung unveils zHBM with 8x performance claim (THELEC, 2026-08-05) - Samsung, SK Hynix 2027 DRAM/HBM already fully allocated (Google News, 2026-08-04) - Memory prices at record highs but Samsung/SK Hynix stocks plunge 8% (Google News, 2026-08-03) - Yujin Tech completes sub-10nm DRAM TiN ALD equipment (THELEC, 2026-08-04) - plus 15 more
If this analysis was helpful · ☕ Support Us · ✈️ Telegram