JPMorgan's new Kospi bull case is a reminder that the AI trade is no longer just about Nvidia and U.S. cloud giants. The money is also moving into the memory suppliers that make the hardware boom possible.
South Korea's stock market has become one of the clearest public gauges of AI infrastructure demand, and JPMorgan is now putting a bigger number on that story. The bank raised its bull-case target for the Kospi to 10,000, with a base target of 9,000 and a bear case of 6,000, according to Bloomberg, as investors keep repricing the earnings power of Samsung Electronics and SK Hynix.
That is a striking call because the Kospi has already done plenty of work. The index recently pushed toward the 7,900 level after breaking through 7,000, extending a rally that has turned Korea into one of the world's hottest equity markets this year. In ordinary circumstances, that kind of move would invite caution. In this case, the argument from bulls is that earnings estimates are moving faster than share prices.
The reason is memory. AI accelerators need far more high-bandwidth memory than traditional servers, and the supply chain for that product is unusually concentrated. Nvidia-class GPUs may get the headlines, but they depend on stacks of advanced DRAM that sit close to the processor and feed it data at high speed. SK Hynix has been a leader in that market, while Samsung is fighting to turn its scale in memory into a larger share of HBM demand.
For years, memory stocks were treated as brutally cyclical businesses. Prices rose when supply tightened, companies overbuilt, prices fell, and investors moved on. AI has not repealed that cycle, but it has changed what investors are watching. HBM is more complex to make, harder to qualify, and more closely tied to long-term orders from chip designers and hyperscalers than commodity memory used in smartphones and PCs.
That matters because AI capital spending is still difficult to read from the outside. Microsoft, Amazon, Google and Meta give investors broad capex numbers, but the real pressure points show up further down the chain. When memory pricing rises, HBM capacity gets locked up, and Korean chipmakers start guiding higher, it tells the market that AI server demand is not just a spending plan. It is hitting physical supply limits.
This is why Korea is suddenly acting like a leveraged AI infrastructure market. Samsung Electronics and SK Hynix are not side characters in the AI buildout. They sit in the middle of it. If cloud companies and model labs keep demanding more compute, memory suppliers gain pricing power. If accelerator demand slows, that same concentration can work in the other direction very quickly.
Recent market commentary shows how much of the Kospi's upside is now tied to that one question. Samsung and SK Hynix make up a large share of the index and have been credited with roughly 70 percent of its gains this year, helped by tightening memory supply and rising chip prices. That does not mean the rest of the market is irrelevant. Defense, shipbuilding, power equipment and financials have also helped. But the main engine is clear.
Investors are buying earnings, not just a theme
The strongest version of the bull case is not that AI is exciting. It is that Korea still looks cheap if the new profit forecasts are real. Goldman Sachs also lifted its Kospi target to 9,000 last week, citing stronger chip earnings and reforms that could narrow Korea's long-running valuation discount. A market that looks stretched on price alone can look less demanding when profits are being marked up every quarter.
That is the part startup founders should pay attention to as well. Hardware costs are shaped by the same forces driving these stocks. If HBM remains scarce, the cost of building or renting advanced AI infrastructure stays high. That favors well-funded labs, hyperscalers and startups with privileged cloud access. Smaller AI companies may find that the bottleneck is not talent or model design, but the availability and pricing of the hardware stack underneath them.
There is also a concentration risk that investors may be too comfortable ignoring. A chip-heavy equity market can rise quickly when the cycle is working, but it becomes vulnerable to any sign of order delays, customer digestion, yield problems, or a faster-than-expected capacity response. Korea's rally is not only a bet on AI demand. It is a bet that Samsung and SK Hynix can keep converting that demand into profits without triggering the usual memory oversupply problem.
The next test is whether earnings revisions keep arriving. A Kospi move toward 10,000 would likely require more than enthusiasm for AI. It would need durable HBM pricing, visible capacity commitments, continued demand from accelerator makers, and enough strength outside semiconductors to keep the index from becoming a two-stock story.
For now, JPMorgan's call sends a clear market signal. AI infrastructure is being priced not just in Silicon Valley, but in Seoul. The next phase of the AI trade may be decided as much by memory supply, packaging capacity and component pricing as by the next model release.
Also read: Founders need to know when AI feels fast enough • OpenAI faces a new liability test after the FSU shooting • MTP benchmarks show AI speed gains depend on the job.
