Samsung's AI chip profits hit record as foundry bleeds share

Key Takeaways

• Samsung's semiconductor division posted its strongest Q2 ever, driven by AI server memory demand
• The foundry business holds single-digit market share versus TSMC's 70%+, though 2nm projects are progressing
• Samsung's mobile division swung to a 700 billion won loss as memory costs surged

Samsung's semiconductor division just delivered its best second quarter on record, with AI server memory pushing both revenue and operating profit to all-time highs. The win, however, exposes a deeper problem: the same AI boom lifting Samsung's memory business is crushing its mobile unit and doing little to close the gap with TSMC in foundry.

The results, reported for Q2 2026, show Samsung riding the AI infrastructure wave while struggling to compete in contract chipmaking. TSMC controls over 70% of the global foundry market. Samsung holds single digits. That gap widened during a painful 2025 marked by low yields and weak utilization in Samsung's 3-nanometer process. The company is now betting its integrated approach, spanning memory, foundry, and advanced packaging, can carve out a defensible niche as AI chips demand tighter component integration.

What drove Samsung's record semiconductor quarter?

AI server memory. The surge in demand for high-bandwidth memory (HBM) used in training and inference infrastructure pushed Samsung's chip revenue and profits past previous records. The foundry side contributed too, with higher utilization rates and increased orders for HBM base dies and logic chips from North American clients.

Samsung did not break out standalone foundry financials, but division officials signaled improving economics. Work on 2-nanometer high-performance computing processes is progressing with major customer projects. The company expects profitability to return to foundry "in the near future," though executives declined to specify timing.

“While it is difficult to specify the exact timing of a turnaround due to the nature of the foundry business, we expect profitability to return in the near future.”

— Samsung foundry division, Q2 2026 earnings call

The qualifier matters. Last year's yield problems on 3nm damaged Samsung's credibility with fabless chip designers who cannot afford production delays. Recovering trust takes quarters, not press releases.

Why Samsung's integrated chip strategy might work

TSMC's dominance rests on pure-play contract manufacturing. It does not compete with customers by selling its own chips. Samsung does. That conflict has historically pushed fabless designers toward TSMC. But the architecture of AI accelerators is changing the calculus.

Modern AI chips require tight coupling between logic processors, HBM stacks, and advanced packaging. Nvidia's H100 and H200 accelerators, for instance, use CoWoS packaging that places HBM directly adjacent to compute dies. The companies that can offer memory, logic fabrication, and packaging under one roof have a structural advantage, at least in theory.

Samsung is one of only two companies with all three capabilities. The other is SK Hynix plus its partnership ecosystem. An industry analyst quoted in The Korea Herald noted this positioning: "Samsung's combination of memory, foundry and advanced packaging capabilities could provide a potential advantage as AI chip demand grows."

The analyst immediately added the caveat that matters: "However, improving 2-nanometer yields and securing major customers remain key challenges for a meaningful recovery in the foundry business."

The mobile division's margin collapse

Samsung's Mobile eXperience and Networks division posted 33.2 trillion won ($23.1 billion) in Q2 revenue, up 13.7% year-over-year. Galaxy S26 and Galaxy A series sales drove the top line. But the bottom line went negative: a 700 billion won operating loss compared to a 3.1 trillion won profit in Q2 2025.

The culprit is memory pricing. The same AI-driven demand that boosted Samsung's semiconductor profits created shortages in mobile-grade memory, pushing component costs sharply higher. Smartphone makers buy memory at market rates. Samsung's vertical integration helps at the margins but does not eliminate the exposure.

“The massive demand for AI servers is creating shortages and driving up prices for mobile memory. We already saw this in the second quarter as memory prices rose quarter-on-quarter, leading to pressure on our profitability, and we expect that cost burden to continue through the second half of the year.”

— Daniel Araujo, VP of Samsung Mobile eXperience and Networks

Araujo said Samsung is "tightening up how we allocate resources across areas like procurement, sales and research and development to minimise the impact on earnings." Translation: cost cuts across the mobile business while they wait for memory supply to catch up with AI demand.

TSMC's 70% versus Samsung's single digits

The foundry market share gap deserves scrutiny. TSMC controls over 70% of global contract chipmaking revenue. Samsung sits in a distant second with single-digit share. Intel Foundry Services and GlobalFoundries trail further behind.

TSMC's lead is not just about capacity. It is about yields, customer trust, and a 30-year track record of never competing with the companies it manufactures for. Apple, AMD, Nvidia, Qualcomm, and Broadcom all rely on TSMC for their most advanced chips. Samsung's dual role as a chip seller and chip maker creates structural tension that no marketing can fully resolve.

Samsung's path forward likely depends on two things: hitting commercially viable yields on 2nm before 2027, and landing a flagship customer willing to bet on Samsung for leading-edge logic. Neither is guaranteed. Both are necessary.

What this means for AI infrastructure buyers

If you are sourcing AI infrastructure, Samsung's results confirm what hyperscalers already know: HBM supply is tight, and it will stay tight through at least H2 2026. Memory pricing pressure will flow through to server costs, which will flow through to cloud compute pricing. Budget accordingly.

For enterprises building AI workflows, the underlying hardware economics favor patience. Rushing to deploy AI inference at scale in late 2026 means paying peak prices for constrained components. Teams evaluating AI automation tools like Zapier or n8n for workflow orchestration may find better ROI in software-layer efficiency gains while waiting for hardware costs to normalize.

The second-half outlook

Samsung expects memory cost pressures on the mobile business to persist through year-end. Foundry recovery depends on 2nm yield improvements and customer wins that have not yet been announced. The company's integrated semiconductor strategy is coherent but unproven at scale.

The AI demand tailwind is real. Whether Samsung can translate it into foundry market share gains, or merely ride the memory cycle while TSMC consolidates logic fabrication, remains the open question for 2027 and beyond.

Frequently Asked Questions

Why did Samsung's mobile division post a loss despite higher revenue?

Rising mobile memory costs, driven by AI server demand creating component shortages, increased production costs faster than Samsung could raise smartphone prices. Revenue rose 13.7% YoY but margins collapsed, swinging from 3.1 trillion won profit to 700 billion won loss.

How far behind TSMC is Samsung in foundry market share?

TSMC holds over 70% of global foundry revenue. Samsung is in distant second place with single-digit share, a gap that widened during Samsung's 3nm yield struggles in 2025.

When does Samsung expect foundry profitability to return?

Samsung executives said profitability should return "in the near future" but declined to specify timing, citing the nature of the foundry business and ongoing work on 2nm process yields.

Source: Tech-Economic Times / ET