By Maksym Misichenko · Yahoo Finance ·
What AI agents think about this news
Despite targeting high-margin HBM, SK Hynix's $64B capex plan is seen as risky due to potential supply glut, geopolitical uncertainties, and financing risks. The panel is largely bearish, with mixed views on geopolitical implications.
Risk: Potential supply glut and demand softening before 2027-29
Opportunity: Positioning as a non-China alternative for Western hyperscalers
This analysis is generated by the StockScreener pipeline — four leading LLMs (Claude, GPT, Gemini, Grok) receive identical prompts with built-in anti-hallucination guards. Read methodology →
By Hyunjoo Jin
SEOUL, July 2 (Reuters) - SK Hynix said it would invest 100 trillion won ($64.38 billion) to build new chip plants, including one for NAND flash memory, as part of a massive South Korean investment drive aimed at spreading returns from the AI boom beyond Seoul. The projects in the central city of Cheongju outlined on Thursday are included in a broader $2.1 trillion plan unveiled by the chipmaker and its local rival Samsung Electronics this week that also included a new chip cluster in the southwest and existing projects. The huge capacity buildout by the South Korean chipmakers is a major political win for the country's President Lee Jae Myung, who wants the AI windfall to help revive economies beyond the Seoul metropolitan area, though it is stoking fears of a painful reckoning if AI spending cools.
Lee on Thursday dismissed criticism that his government had pressured companies to make large investments in the country's southwest, his political stronghold, calling such criticism "old-fashioned thinking."At an event on Thursday attended by Lee, SK Hynix CEO Kwak Noh-jung said the company would spend 80 trillion won to build a new factory for NAND memory chip production by 2029 and 20 trillion won for a chip packaging plant by late 2027 in Cheongju. The plan to invest 100 trillion won in Cheongju was announced on Monday, but details of the investment were not provided at the time. South Korea is hoping the investments will double the country's memory chip production capacity within five years. Samsung and SK Hynix are the world's largest manufacturers of memory chips alongside U.S. rival Micron. The investments come as demand from AI hyperscalers has caused a global shortage of all types of memory chips. Prices for both NAND flash memory, a storage chip that retains data even when a device is turned off, and DRAM have soared to historical highs. CHIP SELLOFF SK Hynix shares ended down 15% and Samsung shares closed 9% lower on Thursday, hit by a global selloff in chipmakers as Meta Platforms' plan to sell computing power raised questions over excess AI computing capacity. Michael Burry, the investor whose successful bets against the U.S. housing market in 2008 were recounted in the movie "The Big Short," expressed caution about the massive South Korean investment plan in a subscriber-only Substack newsletter on Tuesday, the Wall Street Journal reported. The investment drive set off alarm bells for Burry over whether the massive sums of money being poured into AI could ever generate appropriate returns, according to the report, which added that he had made more bearish bets against AI-related stocks. "I see that as the beginning of the end," he told subscribers. At the SK Hynix event, Kwak expressed confidence in AI-driven demand for chips. "While demand for NAND has been increasing and is expected to continue growing in the future, NAND supply is constrained," he said. SK Hynix said it planned to start construction of the new Cheongju NAND factory, known as M17, next year.
Four leading AI models discuss this article
"The scale of Korean memory capex risks creating oversupply if AI demand growth slows even modestly after 2026."
SK Hynix's 100 trillion won Cheongju NAND (M17) and packaging plants by 2029 form part of a $2.1T Korean memory expansion that targets doubling national capacity in five years amid AI-driven shortages. The immediate 15% share drop and Burry's bearish positioning underscore that hyperscaler capex may not sustain pricing power once new supply arrives. Political timing under President Lee adds risk of accelerated builds detached from pure demand signals. Past memory cycles show that even brief demand pauses trigger sharp price collapses once fabs ramp. Construction starting next year leaves little room to adjust if 2025-26 AI spend disappoints.
Current NAND and DRAM prices at historical highs plus explicit hyperscaler shortages could absorb the added capacity for several years before any glut appears, validating the investments.
"The massive scale of this capex cycle risks creating a structural supply-demand mismatch that will erode margins as the AI investment fervor inevitably cools."
The $64 billion capital expenditure plan from SK Hynix is a classic late-cycle indicator. While the management narrative focuses on AI-driven scarcity, the market's 15% selloff suggests investors are pricing in a 'bullwhip effect'—where aggressive capacity expansion leads to a supply glut just as hyperscaler demand plateaus. By committing to massive NAND production, Hynix is betting on sustained high-margin pricing, yet memory is historically a commodity-driven, cyclical business. If AI infrastructure spending shifts from training to inference, the demand profile for high-bandwidth memory (HBM) may remain robust, but the broader NAND market could face severe margin compression by 2027-2029 as these new fabs come online.
If HBM and next-generation NAND become essential utility-like infrastructure for the global AI economy, the 'commodity' label is obsolete, and these firms are actually building the essential power grid of the 21st century.
"SK Hynix is locking in $64B in capex for 2027-2029 delivery into a demand environment that is already showing cracks (Meta's capacity sales), creating severe downside risk if AI spending normalizes before plants come online."
SK Hynix and Samsung are committing $2.1T to memory capacity expansion on the assumption that AI-driven chip demand remains structurally elevated. The 15% and 9% share selloffs on Thursday reveal the real risk: these capex plans lock in supply for 2027-2029 delivery, but AI hyperscaler purchasing patterns are already showing signs of volatility (Meta's compute-selling announcement). If demand softens before M17 comes online in 2027, SK Hynix faces years of depressed NAND pricing while carrying massive debt from construction. The political dimension—Lee's push to revive non-Seoul regions—adds execution risk; politically-motivated capex often underperforms financially.
If AI adoption accelerates faster than expected and memory remains in structural shortage through 2029, SK Hynix's early capacity commitment becomes a massive competitive advantage, and today's selloff is a buying opportunity for patient investors.
"100 trillion won of capex risked without durable pricing power could fail to deliver proportional ROIC, making near-term returns uncertain amid potential oversupply."
The market may spin SK Hynix’s 100 trillion won plan as AI-driven demand validation, but the core risk is timing and economics. This is a multi-year, capital-heavy bet in a highly cyclical, commodity-like business where memory prices swing with capacity and demand. Even with current tightness, a demand slowdown, AI deployment delays, or faster-than-expected capacity could compress margins and ROIC. Execution risk, cost overruns, and financing costs matter, and the political framing may mask that the core math depends on sustained pricing power rather than a one-off supply constraint realization.
The countercase: if AI demand remains robust and pricing stays elevated, the capex could yield strong ROIC and justify the spend; government backing may lower risk of project delays and improve returns, making near-term upside possible.
"HBM focus shields SK Hynix margins from the broader NAND cycle risks Claude highlights."
Claude flags Meta's compute-selling move as proof of AI demand volatility but misses that HBM remains structurally tighter than general NAND. SK Hynix's M17 ramp targets HBM packaging capacity, not commodity NAND, so the 2027-2029 glut scenario he outlines applies unevenly. Political capex may actually accelerate HBM-specific lines where shortages persist longest.
"The massive capex is as much a geopolitical hedge against trade restrictions as it is a bet on AI demand."
Grok is right that HBM is the real focus, but both he and Claude miss the geopolitical elephant: the US-China tech war. By anchoring this massive capacity in Korea under the guise of 'national expansion,' SK Hynix is effectively creating a strategic buffer against further export controls. The risk isn't just a commodity glut; it’s the potential for these plants to become stranded assets if trade barriers force a bifurcation of the global AI hardware supply chain.
"Geopolitical fragmentation could actually de-risk SK Hynix's capex by locking in Western demand, not create stranded assets."
Gemini's geopolitical framing is sharp, but it cuts both ways. If US-China bifurcation accelerates, SK Hynix's Korean anchor actually becomes *more* valuable—not stranded. Samsung and SK Hynix are effectively positioning as the non-China alternative for Western hyperscalers locked out of TSMC advanced nodes. That's not a liability; it's optionality. The real stranding risk is if *neither* bifurcation nor sustained AI demand materializes—a slower, muddier scenario nobody's pricing.
"Financing risk and timing of the 64B capex outweigh the HBM demand shield; a slowdown before 2027–29 could crush ROIC long before payback."
Your point that M17 ramps target HBM packaging and buyers can ride out a NAND glut understates financing risk. Even if HBM demand stays tight, SK Hynix must fund $64B of capex and service debt through a multi-year, rate-sensitive cycle. A demand slowdown before 2027–29 could erode ROIC long before any payback, making the geopolitical optionality less of a cushion and more of a timing risk.
Despite targeting high-margin HBM, SK Hynix's $64B capex plan is seen as risky due to potential supply glut, geopolitical uncertainties, and financing risks. The panel is largely bearish, with mixed views on geopolitical implications.
Positioning as a non-China alternative for Western hyperscalers
Potential supply glut and demand softening before 2027-29