Samsung heads for an 18-fold profit jump as AI memory demand runs hot

Analysts expect around 86 trillion won in second-quarter operating profit, a third straight record, as DRAM and NAND prices climb.

Samsung Electronics is on course for one of the strangest earnings comparisons in its history. Analysts expect the company to report second-quarter operating profit of around 86 trillion won, roughly $56bn, when it publishes preliminary guidance this week.

That would be about 18 times the 4.6 trillion won it managed in the same quarter a year ago, back when the memory business was still a drag rather than an engine.

The gap says less about a sudden turnaround than about how far the AI memory cycle has run since Samsung crossed $1tn in market value earlier this year.

The figure is an estimate rather than a company number, drawn from analysts polled ahead of the guidance, and the usual caveats apply. Preliminary Korean earnings rarely miss the mark by much, though, and the direction is not in dispute.

If it lands, it would be Samsung’s third consecutive quarter of record operating profit, each one propelled by the same thing: memory chips selling for far more than they did a year ago.

Prices are the story. Citi Research put the quarter-on-quarter increase in DRAM contract prices at about 44 percent, with NAND flash up roughly 53 percent, and HSBC reached broadly similar conclusions.

Those are not the incremental moves the memory market is used to. They reflect a shortage that has stopped feeling cyclical and started to feel structural, as demand for AI infrastructure keeps outpacing what the three big memory makers can supply.

Micron’s quadrupled revenue told the same tale from the American side of the industry.

What has changed in this cycle is breadth. High-bandwidth memory, the stacked DRAM that sits beside Nvidia’s accelerators, remains the most profitable corner of the business, and Samsung has spent the past year trying to close the gap on SK Hynix there.

But the price surge now extends well beyond HBM. Standard DRAM and conventional NAND are climbing too, as AI workloads spill out of training clusters and into the ordinary server and storage build-outs that support inference at scale.

That has consequences a long way from the data centre. Memory makers have largely reallocated existing capacity toward AI rather than adding new lines, which is why the same shortage that is inflating Samsung’s profit is also squeezing the parts bin for phones and laptops.

Apple felt it directly when it pulled the $599 Mac mini amid the DRAM crunch. Samsung sits on both sides of that trade, selling the scarce memory while its own devices division pays more for it.

Samsung supplies memory to most of the companies building large AI systems, among them Nvidia, Google, and Apple, which places it near the centre of the infrastructure chain even where it is not the market leader.

The competitive picture is not entirely comfortable, all the same. SK Hynix has been the faster mover on the newest HBM, locking in a multi-year supply agreement with Nvidia, and the two Korean giants now trade the title of the country’s most valuable company between them.

There is also the question of what a number this large does inside the company. Samsung’s semiconductor division has been generating the overwhelming share of group profit, a lopsidedness that has already fed a bruising dispute over how the windfall should be split with workers.

Record earnings tend to sharpen those arguments rather than settle them.

The preliminary guidance will give the headline profit and revenue figures but not the divisional breakdown, which comes with the full results later in the month.

Until then, the estimate stands as a marker of how completely the memory business has been reshaped by a single demand cycle.

Eighteen times last year’s profit is not a recovery. It is a different company operating in a different market.