Samsung Electronics guided to 89.4 trillion won in preliminary Q2 2026 operating profit, beating analyst consensus. Shares fell as much as 8.7% on the news. The orders behind the beat come from a narrow set of AI data center buyers.
Samsung Electronics released preliminary second-quarter 2026 operating profit guidance of 89.4 trillion won ($66 billion) on July 7, roughly nineteen times the year-ago figure and 56% above the first quarter's 57.2 trillion won. The new figure cleared both the LSEG consensus of 87.3 trillion won and the FnGuide consensus of 84.4 trillion won. Samsung's shares fell between 6.8% and 8.7% in Seoul the same day. The memeburn analysis frames the surge as orders from a narrow set of AI data-center buyers: the hyperscalers running AI training clusters and the memory suppliers routing high-bandwidth memory into Nvidia's accelerators.
Conventional memory is moving on the same supply pressure as HBM, even though it has no direct link to AI. Citi Research's three-month price index, as relayed by memeburn, tracked a 44% rise in conventional dynamic random-access memory (DRAM, the working memory in phones, laptops, and servers) and a 53% rise in NAND flash storage over the same window. Citi's read: high-bandwidth memory, the stacked memory chips placed beside AI accelerators inside data-center systems, is consuming clean-room and packaging capacity that used to make standard DRAM and NAND.
Memeburn reads the lift as a customer concentration shift unusually fast for a memory business that historically sold DRAM and NAND into a broad consumer and corporate market. The orders now driving Samsung's pricing power, on the same reading, come from a small group of AI infrastructure buyers rather than the long tail of phone, laptop, and consumer electronics customers that built the memory business.
Samsung is the second-place HBM supplier behind SK Hynix. SK Hynix signed a prioritized HBM3E supply agreement with Nvidia in December 2023, and the Korean rival has run the lead position since. Samsung's 8-layer HBM3E cleared Nvidia's qualification tests for use in AI processors in August 2024, according to people briefed on the results as reported by Reuters. Nvidia's most recent 10-K, for the fiscal year ended January 25, 2026, lists SK Hynix, Micron Technology, and Samsung as memory suppliers, and lists Samsung Electronics as a foundry partner alongside TSMC. The same supplier list appeared in the prior year's filing.
Two pieces of trade-press commentary should be flagged as unconfirmed. A Medium commentary describes a June 7, 2026 multi-year co-development agreement between Nvidia and SK Hynix covering the Vera Rubin AI supercomputer platform, Vera CPU, RTX Spark personal AI computer, and Jetson Thor robotics platform, alongside advance payments of $540 million to $770 million each to SK Hynix and Micron. Neither the deal terms nor the cash figures have been verified against an Nvidia or SK Hynix primary release. The numbers are useful as context for why investor pricing of customer concentration may be intensifying; they should not be treated as confirmed.
The next concrete read on whether the market got the buyer mix right comes from the audited disclosure. The preliminary guidance released July 7 is not a final result; Samsung expects to release consolidated second-quarter figures later in July. Investors watching the HBM-versus-conventional-DRAM line and the customer split will get a clearer answer then. If the disclosure confirms AI buyers lifting the headline and conventional memory doing the heavy lifting on second-order prices, the concentration narrative hardens. If HBM carries more of the weight than the broader price spread suggests, the selloff could look like an overcorrection.