Samsung's record profit fails to impress on AI-chip jitters

Samsung guided to roughly 89.4 trillion won, about $58.4 billion, in second-quarter operating profit - a 19-fold jump from a year earlier and the largest quarterly operating profit the company has ever reported, beating the consensus LSEG SmartEstimate of 87.3 trillion won by about 6% ^[1]. On paper it was a blowout. In the market it read as a warning: Samsung shares fell as much as 10.1% and closed down about 6.9%, erasing more than $100 billion in market value in a single session ^[2].

The disconnect has a simple mechanic behind it. Samsung stock had already climbed roughly 150% over the year, pricing in exactly this kind of AI-memory windfall ^[1]. When a stock is priced for perfection, even a record only matches expectations, and a 6% beat is not the sort of upside surprise that pulls in new buyers. Albert Yong of Petra Capital Management said the earnings were widely expected and had largely been priced in after the pre-earnings rally ^[1]. What looked like disappointment was really textbook profit-taking on a trade that had already run.

Strip away the market drama and the source of the profit is unusually concrete: memory prices are exploding because AI has created a genuine scarcity. Average selling prices for DRAM rose 44% and for NAND flash 53% quarter-on-quarter, according to Citi Research figures cited in the earnings coverage ^[3]. Rapid expansion of high-bandwidth memory - the stacked, high-speed memory that AI accelerators depend on - has absorbed manufacturing capacity, tightening supply of the ordinary memory that goes into phones, laptops and servers.

That is why the reaction outside trading desks has been closer to anger than applause. On Reddit, the loudest threads were not about the earnings beat but about the consumer bill behind it, with builders complaining that an SSD now costs several times what it did a year ago and framing the record profit as AI-era price gouging rather than a triumph. The community mood skewed heavily toward bubble and gold-rush metaphors, and a recurring worry that margins this fat cannot survive the moment supply catches up. It is a useful counterweight to the Wall Street framing: the same scarcity that produced a historic profit is a rising cost for anyone buying computer hardware.

The most consequential reaction came from the sell side. Morgan Stanley told clients that the narrow, semiconductor-centered rally is ending and that it prefers to reduce chip exposure in favor of hyperscalers, arguing chip earnings momentum has passed its peak ^[4]. That call matters because it reframes the selloff from a one-day wobble into a possible turn: if the biggest memory names have already delivered their best numbers, the trade that carried them up all year loses its engine.

The timing amplified everything. South Korea's KOSPI fell 4.9% and tripped a circuit-breaker halt, reported as the market's sixth this year, as SK Hynix and other chip names sold off alongside Samsung ^[5]. The jitters crossed the Pacific too, with Nasdaq futures sliding after the record profit failed to reassure investors about the durability of AI capital spending ^[6]. Analysts also flagged a structural quirk: 2026 estimates are front-loaded into the first half, so any easing of the memory shortage or new discipline in hyperscaler spending could tip the second half into a correction ^[4].

There is a serious case on the other side, and it is not just retail optimism. Samsung and SK Hynix have warned that AI-driven memory shortages could persist through 2027 and beyond, with customers already reserving HBM supply years in advance ^[7]. Together with Micron, the two Korean giants control well over 90% of the global DRAM market, a triopoly now valued in the trillions - a level of pricing power that looks less like a bubble and more like a structural repricing of memory as a scarce input to AI ^[7].

The real disagreement, then, is about duration, not direction. Bulls see a multi-year supercycle in which memory makers finally capture durable margins; skeptics like Morgan Stanley see a peak already in the rear-view mirror ^[4]. Both can point to the same July print. What breaks the tie is not this quarter's profit but next year's supply - whether new fab capacity and any softening in hyperscaler capex arrive fast enough to puncture prices, or whether the shortage, as the memory makers insist, has years left to run. Samsung's detailed divisional results, due July 30, will be the next real test of which story is winning ^[3].