Samsung $200B Broadcom Deal and Record Q2 Profit - But the Long-Term Contract Advantage Is More Complex Than BofA's Bull Case

Samsung's Q2 2026 operating profit hit 89.5 trillion won ($61.98 billion). The number alone looks like vindication for Bank of America Merrill Lynch's June thesis that Samsung's long-term contracts provide downside protection with no cap on upside, and that memory demand will run through 2028.

But that headline number is not a single story. It is a story in which Samsung's total quarterly profit surpassed its combined operating profit for the three years 2023–2025 - while its smartphone business posts an operating loss on the same component price surge. The boom and the squeeze happen simultaneously. That structural tension is the part of the BofA thesis worth scrutinizing.

Samsung's Q2 revenue reached 171.5 trillion won, up 130% year-on-year and 28% quarter-on-quarter. Operating profit jumped 19-fold to 89.5 trillion won. The Device Solutions division - which houses memory and foundry - accounted for 127.5 trillion won in revenue and 89.2 trillion won in operating profit. Nearly all of Samsung's consolidated profit came from one unit.

The other side of the same trade: the MX and Networks unit (smartphones) recorded an operating loss of 700 billion won. Revenue grew year-on-year on Galaxy S26 demand, but surging memory and display component costs erased margin. Samsung is simultaneously the biggest beneficiary and the biggest casualty of the memory price spike. The irony is internal: the company's most iconic brand is being squeezed by its own chip division's pricing power.

This is not a minor footnote. Samsung raised handset prices already and analysts expect further increases. The memory supercycle is cannibalizing Samsung's consumer business the way no external shock has in years.

Decomposition: The Memory Supercycle and the Mobile Sacrifice

Samsung's Q2 revenue reached 171.5 trillion won, up 130% year-on-year and 28% quarter-on-quarter. Operating profit jumped 19-fold to 89.5 trillion won. The Device Solutions division - which houses memory and foundry - accounted for 127.5 trillion won in revenue and 89.2 trillion won in operating profit. Nearly all of Samsung's consolidated profit came from one unit.

The other side of the same trade: the MX and Networks unit (smartphones) recorded an operating loss of 700 billion won. Revenue grew year-on-year on Galaxy S26 demand, but surging memory and display component costs erased margin. Samsung is simultaneously the biggest beneficiary and the biggest casualty of the memory price spike. The irony is internal: the company's most iconic brand is being squeezed by its own chip division's pricing power.

This is not a minor footnote. Samsung raised handset prices already and analysts expect further increases. The memory supercycle is cannibalizing Samsung's consumer business the way no external shock has in years.

The Contract Structure: Forward Purchase, Not Price Insurance

BofA's core thesis rests on Samsung's multi-year supply agreements limiting downside while leaving upside uncapped. The contracts exist, but their mechanics do not match the simple "floor without a ceiling" framing.

The $200 billion MOU Samsung signed with Broadcom in July 2026 - and the separate long-term deals with five of the world's largest data center customers - are structured as forward purchase commitments. As South Korean presidential adviser Kim Yong-beom clarified at the post-summit briefing: "These are advance contracts - advance orders for memory chips on that scale. This is not an investment."

The structure says: if Samsung produces the chips at contracted volumes, the buyer commits to purchasing them. For Samsung, this provides demand certainty to justify building new HBM-capable fabs that cost $10 billion or more each and take three-plus years to bring online. For Broadcom, it converts a volatile spot market - where contract DRAM prices rose 93–98% quarter-over-quarter in Q1 2026, the largest quarterly spike in memory history - into a managed multi-year pipeline.

But the contract terms have not been disclosed. There is no public evidence of price floors, minimum-margin guarantees, or downside protection clauses. The "uncapped upside" BofA describes is real only to the extent that spot-market prices continue to rise above contract pricing - and that Samsung's non-contracted capacity can capture the difference. Given that Samsung's HBM capacity remains constrained by yield and qualification issues rather than installed wafer starts, the upside capture is not automatic.

HBM Market Position: The Number the Bull Case Skips

Here is the data point the BofA thesis doesn't address with the specificity it deserves: SK Hynix overtook Samsung in full-year operating profit for 2025 - the first time in history. SK Hynix posted 47.2 trillion won in annual operating profit to Samsung's 43.6 trillion won. SK Hynix's entire business is memory. Samsung's memory segment alone generated 24.9 trillion won. The gap comes from Samsung's diversified structure, but the HBM story is the real differentiator.

SK Hynix holds an estimated 53–60% share of the HBM market, compared to Samsung's ~22%. SK Hynix secured over two-thirds of HBM supply orders for Nvidia's next-generation Vera Rubin platform. Samsung's HBM4 mass production came after documented yield delays that pushed the timeline back from an original target. Samsung is now targeting ~50% HBM capacity growth in 2026, but the binding constraint is qualification timelines, not installed wafer capacity.

All three major HBM suppliers - Samsung, SK Hynix, and Micron - are sold out through 2026. New meaningful capacity does not arrive until 2027 or 2028. That is why BofA projects the HBM market growing from $35 billion in 2025 to $54.6 billion in 2026, potentially reaching $100 billion by 2028. The demand trajectory is clear. The question is who captures it.

Samsung's Broadcom MOU positions it to supply HBM4 and HBM4E to Broadcom's custom AI accelerator lineup. That is a major account win. But SK Hynix locked in a $750 billion supply partnership with Nvidia and a broader US coalition at the same San Francisco summit. On the Nvidia side - which remains the largest single driver of HBM demand - SK Hynix still holds the structural lead.

Supply Chain Impact: The Memory Shortage Squeezes Everyone Else

The memory supercycle is not confined to the three chipmakers. Conventional DRAM and NAND prices have surged because HBM production absorbs leading-edge DRAM capacity, tightening supply across the board. Citi Research reported average selling prices for DRAM and NAND rose 44% and 53% quarter-on-quarter in Q2 2026.

Samsung's own smartphone division is the most visible casualty. But the pressure radiates through the entire supply chain: PC manufacturers, automotive electronics, IoT device makers, and second-tier smartphone brands all face elevated component costs. Samsung has absorbed part of the shock internally, but external buyers are feeling it too. Memory chip costs account for an estimated 52% of public cloud providers' capex this year, potentially reaching 70% in 2027. That is why cloud companies are the ones signing the multi-year contracts - not to get cheaper chips, but to guarantee access.

Samsung's foundry business, meanwhile, benefited from HBM base-die demand and strong orders from US customers, though earnings were reduced by incentive-related provisions. The foundry unit is trying to build momentum on 2nm HPC engagements while its 4nm products serve LPU and base-die demand.

The BofA Thesis: Where It Holds, Where It Thins

The parts of the BofA case that stand up to the data are straightforward. Demand visibility through 2028 is real: Samsung itself has said supply constraints will tighten in 2027 as AI token generation drives exponential demand growth. The contract pipeline provides genuine demand certainty. Memory is now structural infrastructure, not a cyclical commodity - and that regime shift is the core of the bull thesis.

The parts that thin under decomposition are the contract mechanics and Samsung's competitive positioning. The forward purchase agreements are demand guarantees, not margin insurance. Samsung's HBM market share has not yet converged with its conventional DRAM share, despite management's stated goal of alignment. And the internal cannibalization between memory profits and mobile losses shows that even the biggest beneficiary of a supercycle is not immune to its side effects.

What to watch next:

• Samsung's HBM4 qualification timeline with Nvidia. The Broadcom MOU is significant, but Nvidia remains the dominant HBM demand driver. Samsung needs to close the qualification gap to compete for the lion's share of Vera Rubin and beyond.
• Q3 earnings breakdown. Samsung's detailed divisional results for Q2 were released on July 30. The memory vs. foundry profit split, and the size of the mobile loss, will tell you how durable the internal cost pressure is.
• 1c DRAM yield maturation for both Samsung and SK Hynix. Yield curves on leading-edge DRAM nodes typically take 2–4 quarters to stabilize. If Samsung's 1c node accelerates past SK Hynix's yield curve, the HBM share gap narrows faster than consensus assumes.
• Contract pricing terms disclosure. Samsung and its counterparties have not revealed pricing mechanics. Any future disclosure of contract vs. spot pricing spreads will confirm or undermine the "uncapped upside" framing.
• Cloud provider capex guidance at Q3 earnings. Memory represents half or more of hyperscaler capex. Any material slowdown in AI infrastructure spending at Google, Microsoft, Meta, or Amazon would propagate directly back to Samsung's revenue assumptions.