From Supplier to Stakeholder — Why K-Memory Just Wrote Equity Checks Into the AI Stack

The $1T Twins and an Anthropic Cap Table

May 29, 2026 may stand as the day K-memory's structural pivot crystallized. In the same week, Samsung Electronics and SK Hynix both crossed the $1 trillion market cap threshold, joining the global mega-cap club together. On the same day, both names appeared — alongside Micron — in Anthropic's $65 billion Series H round at a $965B post-money valuation. The Korean memory duo did not just sign a supply contract; they bought equity in the AI builder itself.

That one sentence compresses the inflection point. Through the prior HBM3/HBM3E cycle, SK Hynix held Nvidia's sole-source slot while Samsung played catch-up. This cycle is different: both Korean champions now sit on the cap table of a frontier AI lab, repositioning the equity story from a spot-DRAM beta into a call option on AI compute demand itself.

Samsung's Speed Run — HBM4E Samples Just 3 Months After HBM4

The most dramatic shift is Samsung's compressed roadmap. On May 28–29, Samsung announced it had shipped the industry-first HBM4E 12-high samples — 32Gb DRAM dies, up to 16 Gbps — to global AI accelerator customers, just 3 months after HBM4 mass production began. The same week, Samsung also delivered 7th-generation HBM (HBM5) samples roughly 6 months ahead of the industry roadmap. SK Hynix countered by pre-announcing an 8th-generation HBM leap, formally opening what Korean press is calling the "leapfrog war."

KB Securities raised SK Hynix's target price to KRW 3.8 million on expectations that HBM prices will more than double in 2027 amid sustained AI demand and tight supply. This is not a vanilla cycle call — whoever locks first qualification on HBM4E 12-high will set the memory ASP and BOM share on every AI accelerator shipping over the next 18 months.

The Foundry Leg — Taylor Fab and Tesla AI5/AI6

The capital-partner thesis is not confined to memory. At Samsung Foundry's SAFE Forum, US VP Margaret Han confirmed the $17B Taylor, Texas fab is "ready," with customer production starting in 2026 and Tesla AI5/AI6 chips slated for the SF2P+ 2nm line. ADTechnology used the same stage to announce that its ADP620 — a 2nm HPC CPU chiplet using Samsung's 2nm process and Arm V3 cores — had hit 3.695 GHz on silicon. SemiFive showcased 3D-IC and Big Die ASIC solutions in a Samsung co-development project.

Stitch the pieces together and the picture sharpens: Anthropic equity on memory, Tesla AI5/AI6 production on foundry, a 2nm HPC chiplet ecosystem on design services. The K-semiconductor complex is repositioning from "AI supplier" to "AI stack co-owner."

The Fundamentals Underwrite the Capital Game