Samsung's Memory Boom Reshapes Investment Strategy as Foundry Capacity Crunch Looms in 2027
nashnova research
Samsung posted the tech industry's first ₩100-trillion single-quarter operating profit, but the memory super-cycle that funds its new mega-deal playbook is also devouring foundry capacity from the inside — a 2027 shortage is already locked in.
A ₩100-trillion quarter — where is the money coming from?
Q3 2026 operating profit hit roughly ₩107.4 trillion at an operating margin of about 55% — making Samsung the first tech company to clear ₩100 trillion in a single quarter.
The engine is memory. This means → the higher the peak, the steeper the potential drop if the memory cycle turns.
In plain terms = Samsung is flying on one engine, and that engine just set a thrust record — but the backup engine is not yet installed.
Why has the investment style shifted from "spray and pray" to "fewer, bigger, lead"?
Over the past year deal count fell while ticket size rose to $50 m–$250 m per round. Samsung moved from co-investor to lead, and from early-stage to late-stage growth companies.
Key deals: co-led Mistral AI's ~€3 billion Series D; Catalyst Fund led Normal Computing's $50 m strategic round and co-led OXMIQ Labs' $35 m Series A; also participated in Anthropic's Series H.
Exits were equally active — Marvell acquired Celestial AI for up to $5.5 billion (including earnouts); Analog Devices closed its $1.5 billion acquisition of Empower Semiconductor. Both were Catalyst Fund portfolio companies.
This reflects a deliberate pivot: memory-cycle cash is being redeployed from exploratory bets to strategic positions.
A brand-new M&A team — where is the money headed?
Samsung stood up its first dedicated M&A unit in November 2025, then published a corporate-value plan in March 2026 committing over ₩110 trillion to facilities and R&D for the AI-semiconductor era.
Four stated M&A priorities: advanced robotics, medtech, automotive electronics, and HVAC. Two deals already closed — ~€1.5 billion for Germany's FläktGroup and ~€1.5 billion via Harman for ZF Group's ADAS business.
Two new funds launched: SVIC 82 (₩500 billion, semiconductor start-ups) and SVIC 83 (₩300 billion, future consumer-electronics tech).
This means → Samsung is using the memory peak to fill a "second growth curve" beyond semiconductors.
A 15% price hike — where does the bargaining power come from?
Samsung raised prices on select wafer-foundry services in July, with some orders up as much as 15%. Strong demand and existing commitments to U.S. clients already limit room for new orders.
Marquee contracts keep landing: Tesla signed a deal extending to 2033, worth $16.5 billion; a Broadcom MoU covers HBM, 2 nm foundry, and advanced packaging with an estimated scope exceeding $200 billion through 2030; Nvidia's Groq 3 LPX inference accelerator — whose LPU (Language Processing Unit, a custom chip built to run large-model inference) is fabbed by Samsung — entered mass production.
Samsung also signed an agreement with AMD to deepen advanced-memory cooperation and explore foundry opportunities. This means → Samsung may leverage its HBM supply as a bargaining chip to pull some of AMD's advanced AI chip production into its foundry, but no confirmed capacity transfer has been publicly announced.
Why will capacity "blow past 100%" by 2027?
On its Q3 earnings call Samsung Foundry disclosed that by 2027, demand across all nodes below 8 nm is expected to exceed capacity, pushing utilization above 100%.
The biggest capacity consumer is not an external client — it is Samsung itself. The foundry division produces the logic base die for Samsung's own HBM4 memory — the bottom logic layer of an HBM stack that handles signal routing — and HBM4 is Samsung's fastest-growing product.
In plain terms = Samsung's foundry lines must "pour the foundation" for its own memory division and fill external orders at the same time, both competing for the same fab lines. The window for outside customers will keep narrowing around 2027.
2 nm and the Taylor fab — what is the real timeline?
The 2 nm project pipeline more than doubled year-on-year in Q3, with new customers from hyperscalers and AI / HPC. High-performance computing is expected to account for over 50% of foundry revenue.
Management was candid: 2 nm yield stabilization is still in progress, and volume production for new customers will not show up in revenue until H2 2027.
The Taylor, Texas fab's first building has begun cleanroom operations; equipment installation and test wafers are under way, with operations targeted for 2027. Samsung is evaluating further investment in the U.S. and Korea, including additional fabs and advanced-packaging capacity.
This reflects a structural tension: the memory super-cycle has given the foundry unprecedented pricing power, but internal chip demand plus unfinished 2 nm yields mean external customers won't fully benefit until after H2 2027.
市场有风险，内容仅供研究参考，不构成投资建议。