The Quarter Record Profits Became Bad News

On July 6, Samsung Electronics reported an operating profit of 89.4 trillion won — a 19-fold increase over the same quarter a year earlier, the kind of number that once sent shares vertical. Instead, Samsung stock plunged 10 percent, erasing $100 billion in market value, and the Korea Exchange activated a circuit breaker after the KOSPI index fell more than 8 percent intraday ^[1][2]. A mechanical safety valve built for panic had been tripped by a record earnings release. The same data point that would have been a buy signal six months earlier had become a sell signal, and the market's machinery registered the reversal before most investors could name it. In January, Samsung had forecast a record 20 trillion won in fourth-quarter operating profit, driven by AI-fueled demand for high-bandwidth memory chips, and the market treated the number as a growth runway ^[3]. By April, the company projected 57.2 trillion won for the first quarter — an eight-fold year-on-year increase — and Citigroup and KB Securities were still raising their annual forecasts, with some analysts projecting Samsung could become the world's most profitable company by 2027 ^[4]. The same numbers that meant buy in January meant sell in July. What changed was not the numbers but what the market decided they signified. Morgan Stanley analysts warned in early July that memory chips were "approaching peak rate of change in (1) pricing YoY, (2) inventory, (3) the earnings revision breadth" — reframing record profits not as a floor beneath the stock but as the top of a cyclical arc ^[1]. Toni Meadows of BRI Wealth Management put the mechanism plainly.

Valuations in semi-conductor stocks had priced near-perfect demand, for what has been a cyclical area in the past, so was always going to leave stocks vulnerable at some point in what has been a rapid rise. — Toni Meadows

Jensen Huang named the paradox the market had arrived at: if Nvidia delivered a bad quarter, it was evidence of an AI bubble; if it delivered a great quarter, it was fueling one ^[1]. Nvidia had reached a point where strong results and weak ones could both be read as reasons to sell — a bind that had nothing to do with the quality of the numbers and everything to do with the interpretive frame they were now being read through. The pattern was not confined to Samsung. Micron reported record fiscal third-quarter revenue of $41.4 billion, a 346 percent year-on-year increase, with gross margins of 84.9 percent — and its stock dropped 22 percent from its late-June peak ^[5][6]. Broadcom posted record AI semiconductor revenue of $10.8 billion, a 143 percent surge, but its Q3 AI chip forecast of $16 billion fell short of the $17.2 billion analysts expected, and CEO Hock Tan maintained rather than raised the company's $100 billion fiscal 2027 AI revenue target; shares fell 13 to 20 percent ^[7]. SK Hynix reported a six-fold increase in operating profit to 60.54 trillion won — and cratered because the number missed analyst expectations ^[8]. By late July, the Philadelphia Semiconductor Index had lost more than 25 percent in a month. SK Hynix was down 46 percent, Samsung 35 percent, ASML nearly 20 percent ^[9]. The market was not merely rotating out of expensive stocks. It was reading a set of demand-side signals that made record profits look less like a growth story and more like a peak. Meta launched "Meta Compute" in early July to sell excess GPU computing power to external customers, and CEO Mark Zuckerberg declined to provide a timeline for return on invested capital — a major hyperscaler signaling potential overcapacity in the very infrastructure whose scarcity justified the chip boom ^[10]. Lan Guan, Accenture's chief AI and data officer, described what end-users were encountering.

Clients are ready to scale AI, but then they hit this unexpected cost wall. — Lan Guan

A UBS survey found roughly 60 percent of businesses were reducing AI spending. Michael Burry opened short positions on the iShares Semiconductor ETF as of June 30, betting the gap between chipmaker valuations and the slower returns at companies buying the hardware was unsustainable ^[11]. None of this means AI demand has collapsed. Hyperscaler capital expenditure is still accelerating: Amazon projects $200 billion in data center spending for 2026, Alphabet raised its guidance to $180–190 billion with a $460 billion backlog, and the four largest hyperscalers collectively project over $650 billion this year ^[12][13]. Google Cloud grew 63 percent year-on-year, with generative AI model revenue up nearly 800 percent ^[13]. Nvidia trades at 22.9 times forward earnings, barely above the S&P 500's 21.5 times — hardly the multiple of a stock the market believes is in freefall ^[14]. The profits are real. The demand, for now, is real. What changed is the interpretation: the market has decided that $800 billion in AI infrastructure spending in 2026 is more likely to mark the top of a cycle than the middle of one. By July 29, Apple had reclaimed the title of world's most valuable company from Nvidia — by largely sitting out the infrastructure arms race ^[9]. The market rewarded the company that did not build. The firm that stayed out of the boom became more valuable than the firms that defined it.