Samsung Electronics vs SK hynix: Who Leads the HBM Race in 2026?
Data as of August 16, 2026
Quick Take
The HBM race between Samsung Electronics and SK hynix is no longer a simple question of which company can build the fastest memory chip.
SK hynix currently has the stronger strategic position around NVIDIA, supported by a multiyear technology partnership covering next-generation AI memory and HBM. Samsung, however, has significantly narrowed the gap in 2026, becoming the first to commercially ship HBM4, shipping HBM4E samples ahead of rivals, and expanding relationships with AMD and Broadcom.
For global investors, the key question is therefore changing from “Who is the HBM leader?” to “Which company can convert AI memory demand into sustainable earnings, customer diversification and higher returns on semiconductor investment?”
The most realistic 2026 outcome may not be a winner-takes-all market. SK hynix appears better positioned in the NVIDIA ecosystem, while Samsung is building a broader HBM strategy across GPUs, custom AI accelerators, foundry and advanced packaging.
What Changed in the HBM Race in 2026?
The biggest change is that Samsung is no longer merely trying to catch up.
Samsung announced commercial HBM4 mass production and shipments in February 2026. Its HBM4 uses sixth-generation 10nm-class DRAM, known as 1c DRAM, together with a 4nm logic base die. Samsung says the product operates consistently at 11.7Gbps per pin and can scale to 13Gbps, delivering bandwidth of up to 3.3TB/s per stack.
Samsung then began shipping 12-layer HBM4E samples to major global customers in May. The HBM4E product operates at 14Gbps with scalability to 16Gbps and provides up to 3.6TB/s of bandwidth per stack.
SK hynix, meanwhile, began mass shipments of HBM4 during the second quarter of 2026 and plans to ramp production through the second half. It also completed HBM4E sample shipments during the first half. The company says its HBM4 has met customer-required speeds while maintaining power efficiency, yield and cost competitiveness.
That distinction is important.
Samsung has regained technology and time-to-market momentum. SK hynix retains a powerful advantage in customer integration and supply relationships.
The race is therefore becoming much more balanced.
Why HBM Matters More Than Ordinary DRAM
High Bandwidth Memory, or HBM, stacks multiple DRAM dies vertically and places the memory extremely close to an AI accelerator.
The goal is not simply to store more data. It is to move enormous quantities of data between memory and processors fast enough to keep expensive GPUs and AI accelerators fully utilized.
HBM4 doubles the interface width from 1,024 I/O connections used in HBM3E to 2,048. Samsung says its HBM4 can deliver up to roughly 2.7 times the bandwidth of its previous-generation HBM3E product.
This matters because AI computing increasingly faces a memory bandwidth problem.
An accelerator can contain enormous computing power, but that power provides limited value if data cannot reach the processor quickly enough.
That changes the economics of the memory industry.
Traditional DRAM suppliers historically competed heavily on production scale and cost. HBM adds several other competitive variables:
- DRAM process technology
- stacking technology
- thermal management
- advanced packaging
- base-die design
- power efficiency
- yield
- customer qualification
- accelerator co-development
- ability to deliver large volumes on schedule
HBM is therefore moving memory suppliers closer to the center of AI system design.
Samsung's HBM Strategy: Technology Catch-Up Is Turning Into Customer Expansion
Samsung's strongest HBM advantage is its semiconductor manufacturing structure.
The company operates memory, foundry and advanced packaging businesses internally. Samsung argues that the combination allows it to optimize DRAM dies, logic base dies and packaging as one system. Its HBM4 uses Samsung's own 1c DRAM and a Samsung Foundry 4nm logic base die.
This could become increasingly important as HBM becomes more customized.
Future HBM products are likely to require tighter integration between memory and the customer's AI accelerator architecture. Samsung's ability to offer memory, foundry and packaging creates a potential one-stop AI semiconductor model that a pure memory supplier cannot replicate internally.
Samsung is also diversifying customers.
In March 2026, Samsung and AMD expanded their collaboration around HBM4 for AMD's next-generation Instinct MI455X accelerator. AMD said the companies would align on Samsung as a primary HBM4 supplier for the MI455X. Samsung was already a major HBM3E partner for AMD's MI350X and MI355X accelerators.
In July, Samsung and Broadcom announced a strategic semiconductor partnership covering memory and foundry cooperation through 2030. Samsung specifically identified HBM as a memory solution that will support Broadcom's next-generation AI accelerators.
Investors should be careful with the headline figures surrounding these agreements. An MOU or broad strategic cooperation agreement does not mean that the entire announced cooperation value represents committed HBM revenue.
The more important implication is customer diversification.
Samsung does not necessarily need to displace SK hynix everywhere. Winning significant positions in AMD accelerators and custom ASIC platforms could create a second route into the AI memory market.
Samsung expects its HBM sales in 2026 to more than triple from 2025 levels.
SK hynix's Moat: NVIDIA, Customer Co-Development and Execution
SK hynix's competitive advantage is different.
Its moat increasingly comes from deep customer integration rather than product specifications alone.
NVIDIA and SK hynix announced a multiyear technology partnership in June 2026 covering next-generation memory aligned with NVIDIA's AI infrastructure roadmap. The partnership includes co-development across NVIDIA's Vera Rubin systems and other AI computing platforms.
The relationship was expanded again in July. NVIDIA said the companies would establish a long-term AI memory partnership to secure and jointly develop next-generation memory, including HBM. NVIDIA also confirmed that SK hynix HBM4 would power Vera Rubin infrastructure planned for the SK Telecom-NVIDIA AI factory.
That provides SK hynix with something particularly valuable in semiconductors: visibility into the customer's future architecture.
HBM is increasingly designed years before the AI accelerator reaches mass production. Early collaboration can influence specifications, qualification schedules and production planning.
SK hynix said in its second-quarter results that it had finalized long-term agreements with around 10 customers and was discussing additional agreements with major industry clients.
This reduces some of the uncertainty that historically made the memory industry highly cyclical.
The company is also investing aggressively in capacity.
In August 2026, SK hynix approved approximately KRW 54 trillion of investment for the Yongin Y2 and Cheongju M17 fabs. Y2 is intended to support HBM and next-generation DRAM production, with its first cleanroom scheduled for June 2029.
Its existing expansion plans are closer to production: SK hynix is accelerating M15X while preparing the first Yongin cleanroom for early 2027.
Samsung vs SK hynix: The HBM Race at a Glance
| Category | Samsung Electronics | SK hynix |
|---|---|---|
| HBM4 status | Commercial mass production and shipments began in Feb. 2026 | Mass shipments began in Q2 2026 |
| HBM4E | Samples shipped in May 2026 | Samples shipped in H1 2026 |
| Major strategic AI relationship | AMD, Broadcom; also works with NVIDIA | Deep multiyear NVIDIA partnership |
| HBM4 architecture | 1c DRAM + 4nm logic base die | Focus on customer requirements, yield and supply stability |
| Structural advantage | Memory + foundry + advanced packaging | HBM specialization + customer co-development |
| Main investment thesis | HBM share recovery and diversified AI customers | Defending AI-memory leadership and NVIDIA exposure |
| Main risk | Qualification, yield and execution across advanced processes | Customer concentration and expectations already embedded in earnings |
| U.S. investor access | KRX shares; LSE GDRs | KRX shares plus Nasdaq ADRs |
Samsung's original shares trade on the Korea Exchange, while its GDRs trade in London.
SK hynix added another major access route in July 2026 when its ADRs began trading on Nasdaq, allowing U.S. investors to access the company without opening a Korean brokerage account.
That may gradually matter for valuation as SK hynix becomes easier to compare directly with U.S.-listed AI semiconductor companies.
The Earnings Difference Is More Important Than the Technology Headlines
HBM leadership matters because it is already changing earnings.
Samsung Electronics reported Q2 2026 consolidated revenue of KRW 171.5 trillion and operating profit of KRW 89.5 trillion. Its Device Solutions semiconductor division generated KRW 127.5 trillion of revenue and KRW 89.2 trillion of operating profit. Samsung said the Memory Business achieved record quarterly results as AI and server demand strengthened and HBM4 sales expanded.
SK hynix reported Q2 revenue of KRW 79.3 trillion and operating profit of KRW 60.5 trillion, with a reported operating margin of 76%. The company attributed the performance to strong AI memory demand and high-value products including HBM and AI-server DRAM.
These figures should not be compared as if the companies had identical business structures.
SK hynix is far more concentrated in memory.
Samsung Electronics combines semiconductors with smartphones, displays, TVs, appliances and other businesses. Even within its semiconductor division, Samsung operates memory, foundry and System LSI.
This produces a major difference for investors.
SK hynix is the more direct AI-memory investment. Samsung is the more diversified semiconductor and technology investment.
When HBM pricing and AI memory demand are extremely strong, SK hynix can show greater earnings sensitivity.
Samsung, however, has more ways to benefit if AI semiconductor value moves beyond memory into foundry, packaging and custom chips.
Which Company Has the Stronger Moat?
If the question is customer integration with NVIDIA, SK hynix currently appears stronger.
Its relationship now extends beyond product supply toward multiyear technology development and AI system roadmaps.
If the question is manufacturing integration, Samsung has an unusual strategic asset.
Its ability to combine memory, advanced logic manufacturing and packaging could become more valuable as conventional standardized HBM evolves toward customized HBM.
That makes the two moats fundamentally different.
SK hynix moat
Customer intimacy → early qualification → production visibility → HBM leadership → more customer collaboration
Samsung moat
DRAM scale → advanced process technology → internal foundry → advanced packaging → integrated AI semiconductor solution
SK hynix's challenge is maintaining its technological and execution lead as Samsung improves.
Samsung's challenge is proving that its broad semiconductor capabilities translate into consistent yields, qualification success and profitable HBM volume.
The Race Is Not Only Samsung vs SK hynix
Global investors should not ignore Micron.
Micron is the third major HBM supplier and continues to advance its HBM4 roadmap. Its current HBM4 specifications advertise speeds above 11Gbps and bandwidth above 2.8TB/s per stack.
This matters because AI accelerator companies have strong incentives to maintain multiple memory suppliers.
A multi-supplier structure can improve:
- supply security
- pricing leverage
- geographic diversification
- qualification flexibility
- accelerator production stability
For Samsung and SK hynix investors, therefore, the central question is not whether HBM demand grows.
The harder question is how much of the industry's economic profit each company can retain as capacity and competition expand.
Bull, Base and Bear Cases
| Scenario | Samsung Electronics | SK hynix |
| Bull Case | HBM4 adoption accelerates, AMD and custom ASIC customers scale, HBM4E qualification succeeds and foundry/base-die integration becomes a competitive advantage | NVIDIA AI infrastructure expands faster than expected, long-term agreements protect pricing and HBM leadership extends into HBM4E |
| Base Case | Samsung rebuilds meaningful HBM share without fully displacing SK hynix | SK hynix remains a leading NVIDIA supplier but the market becomes increasingly multi-sourced |
| Bear Case | Yield or qualification delays return while foundry investment limits returns | Customer concentration increases and HBM supply growth compresses margins faster than expected |
The base case is arguably the most important.
The AI memory market is becoming large enough for more than one supplier to grow.
Samsung does not need to eliminate SK hynix from NVIDIA to build a profitable HBM franchise. SK hynix does not need to stop Samsung from winning AMD or ASIC business to maintain an attractive AI-memory position.
A larger total market can allow both companies to win while competing for different platform economics.
What Would Cause a Valuation Rerating?
A low P/E or P/B ratio alone should not be interpreted as undervaluation.
For SK hynix, a sustained rerating would likely require investors to believe that HBM has structurally reduced the cyclicality of the memory business.
Evidence would include:
- longer customer contracts
- stable HBM pricing
- high utilization without destructive oversupply
- sustained free cash flow
- disciplined CAPEX
- continued NVIDIA platform exposure
- broader AI memory customers
For Samsung, the rerating argument is somewhat different.
Investors would need evidence that Samsung has converted technological progress into meaningful HBM revenue and that its semiconductor ecosystem can create higher returns.
Key evidence would include:
- continued HBM4 volume growth
- HBM4E qualification and mass production
- expansion with AMD and custom ASIC customers
- higher semiconductor margins
- improved advanced-node foundry utilization
- successful integration of memory, foundry and packaging
Samsung's HBM recovery could therefore matter far beyond the HBM business itself.
It could become proof that the company's enormous semiconductor investment base is generating better returns.
What Should Global Investors Conclude?
For investors seeking the cleaner HBM and NVIDIA exposure, SK hynix currently offers the stronger direct investment case.
Its competitive advantage is built around customer co-development, production execution and long-term AI memory relationships.
Samsung is different.
Its 2026 HBM4 progress suggests that the company should no longer be viewed simply as an HBM laggard. Commercial HBM4 production, HBM4E sampling and deeper AMD and Broadcom relationships indicate that Samsung is constructing a credible alternative growth path.
The investment debate is therefore shifting.
SK hynix is defending an established AI-memory moat. Samsung is trying to turn semiconductor scale and vertical integration into an HBM comeback.
That may be the most important distinction for investors heading into the next stage of the AI infrastructure cycle.
What to Watch
Investors should monitor five areas.
HBM4 shipment growth: Samsung must demonstrate that early production milestones translate into meaningful customer volume, while SK hynix needs to sustain its production ramp.
HBM4E qualification: Sample shipment is not the same as mass-production revenue. Qualification timing and customer adoption will matter more than headline specifications.
NVIDIA, AMD and custom ASIC platforms: NVIDIA remains strategically important for SK hynix, while AMD and hyperscaler/custom-ASIC opportunities could determine how quickly Samsung diversifies its HBM business.
Capacity and CAPEX: HBM shortages support margins today, but excessive industry capacity could eventually recreate the oversupply cycles familiar to memory investors.
Margins and cash flow: The strongest confirmation of a structural HBM moat will not be another technology announcement. It will be sustained margins, free cash flow and returns on invested capital.
Sources & Data
- Samsung Electronics — Second Quarter 2026 Results, July 30, 2026.
- Samsung Electronics — Industry-First Commercial HBM4 Shipment, February 12, 2026.
- Samsung Electronics — HBM4E Sample Shipments, May 29, 2026.
- AMD / Samsung Electronics — Strategic Collaboration on HBM4 for AMD Instinct MI455X, March 2026.
- Samsung Electronics — Samsung and Broadcom Strategic Collaboration, July 25, 2026.
- SK hynix — Second Quarter 2026 Financial Results, July 29, 2026.
- NVIDIA / SK hynix — Multiyear Technology Partnership for AI Memory, June 7, 2026.
- NVIDIA / SK Group — Expanded AI Factory and Next-Generation Memory Partnership, July 25, 2026.
- SK hynix — KRW 54 Trillion Yongin Y2 and Cheongju M17 Investment Plan, August 7, 2026.
- SK hynix — Nasdaq ADR Listing, July 10, 2026.
Data as of August 16, 2026
Investment Disclaimer
This article is for informational and educational purposes only and does not constitute investment advice. Investors should conduct their own research before making investment decisions.
FAQ
Is SK hynix still ahead of Samsung in HBM?
SK hynix retains a particularly strong strategic position through its deep relationship with NVIDIA and its long history of HBM customer co-development. However, Samsung's commercial HBM4 shipments and HBM4E progress in 2026 mean the competitive gap is considerably less straightforward than it was in previous HBM generations.
Does Samsung supply HBM to NVIDIA?
Samsung and NVIDIA maintain a broad semiconductor relationship, but investors should distinguish general cooperation from confirmed volume commitments for a specific HBM generation. Publicly confirmed 2026 developments provide clearer visibility into Samsung's HBM4 relationship with AMD and its AI semiconductor cooperation with Broadcom.
Why is SK hynix so important to NVIDIA?
HBM directly affects the bandwidth, energy efficiency and utilization of AI computing systems. SK hynix and NVIDIA have expanded their relationship into multiyear co-development of next-generation memory, making the relationship deeper than a conventional component supply arrangement.
Can U.S. investors buy SK hynix directly?
SK hynix began Nasdaq ADR trading in July 2026. ADRs allow U.S. investors to gain exposure through the U.S. market without directly opening a Korean securities account.
Which is the better AI memory investment: Samsung or SK hynix?
They provide different exposures. SK hynix is more concentrated in memory and currently offers more direct exposure to HBM and NVIDIA-related AI infrastructure. Samsung provides more diversified exposure across memory, foundry, advanced packaging and consumer technology. The better fit therefore depends on whether an investor wants concentrated AI-memory exposure or a broader semiconductor platform.
Comments
Post a Comment