The AI trade has moved down the stack. The companies making memory, wafers and packaging are now showing the kind of earnings power software investors used to dream about.
Micron Technology did not just beat Wall Street last week. It showed you where the money in AI is actually being collected. The company reported $41.46 billion in fiscal third-quarter revenue, more than four times the year-earlier figure, and guided for about $50 billion in revenue in the current quarter, according to its results and analyst coverage after the release. That is not a story about a vague AI future. It is cash from chips already being ordered, priced and shipped.
Look at the shape of the boom. Investors spent the first phase of AI arguing about model companies, cloud platforms and who would own the application layer. Fine. Those questions still matter. But you can't train frontier models, run inference at scale or build another GPU cluster without high-bandwidth memory, advanced foundry capacity and packaging lines that are already under strain.
The market has noticed. The Guardian reported on June 29, using London Stock Exchange Group data, that South Korea's Kospi index was up 123% this year, its strongest first half since at least 1990. Samsung shares had risen 169%, while SK Hynix had climbed 303%. In the US, memory and storage names have also pulled away from the rest of the market. Micron was up nearly 270% year-to-date after its earnings, according to Investopedia, with Sandisk, Western Digital and Seagate also among the biggest winners from the memory squeeze.
This isn't only a stock chart story. SK Hynix reported first-quarter revenue of 52.58 trillion won, about $35.5 billion, up 198% from a year earlier, MarketWatch reported in April. Its operating margin reached 72%. A memory business posting that kind of margin should force you to rethink the old assumption that DRAM is just a cyclical commodity with a prettier press release during upturns.
Frankly, the old framing is now too lazy. Memory is not suddenly immune to cycles, and anyone telling you that semiconductors have become a one-way trade is selling comfort. But the AI buildout has changed the bargaining power of the companies sitting under Nvidia, AMD, Broadcom, Marvell and the hyperscalers. When supply is short and customers need guaranteed access, the contract starts to matter as much as the quarterly beat.
Micron's long-term customer agreements show the shift in plain numbers. Tom's Hardware reported that Micron has signed 16 strategic customer agreements, 14 of which represent about $100 billion of minimum contracted revenue over their remaining terms. The company also expects $22 billion in cash deposits and related commitments. MarketWatch noted that those agreements cover about 20% of Micron's DRAM volume and roughly a third of its NAND volume.
That is a different business from the one investors thought they understood two years ago. Customers are not just buying memory on the spot market and hoping the cycle behaves. They're putting money down because the shortage hurts their own AI roadmaps.
TSMC tells the same story from the foundry side. Tom's Hardware reported last week, citing Culpium, that TSMC has told customers to prepare for price increases across advanced nodes including 3nm, 5nm and 7nm, with increases generally expected in the 5% to 10% range. Advanced nodes accounted for 74% of TSMC's wafer revenue in the first quarter, when the company reported $35.9 billion in revenue and a 66.2% gross margin. Even a mid-single-digit increase on that base is real money.
C.C. Wei's message to shareholders was blunt enough: demand from AI customers remains strong, and supply will take years to catch up. You don't need to turn that into poetry. It means the companies with scarce capacity can say no, raise prices, or ask for longer commitments. The software layer may be where the product demos happen, but the hard constraint is lower down.
Goldman Sachs has also moved its attention in that direction. Business Insider reported in May that the bank called South Korea its highest-conviction equity market in Asia and forecast 300% earnings growth for the country's listed companies in 2026, driven largely by hardware and semiconductors. Samsung and SK Hynix dominate that market, so the call is really a view on memory pricing and AI infrastructure demand.
If you're long AI only through US megacap software and cloud names, you may be missing the part of the trade with the clearest reported earnings. That does not mean you chase every chip stock after a vertical move. The Kospi's run, Micron's surge and SK Hynix's triple-digit gain already price in a lot of good news. But the facts on the ground are hard to ignore: the companies selling the bottleneck are making the money.
There is still risk here. Supply can catch up, customers can over-order, and memory cycles have punished confident investors before. But this cycle has one feature the older ones did not have at this scale: AI companies are signing long agreements because compute plans fail without guaranteed chips. That's the real issue. In 2026, the picks and shovels are not a metaphor. They're the quarterly income statement.
Also read: Ardian is betting over a billion dollars that the Nordics will power Europe's AI infrastructure buildout • Proception settles Tesla's trade secret lawsuit and closes an $11 million seed round on the same day • OpenAI scrambles to fix Codex as coding agent usage blows past its own capacity models
