The question isn't whether Samsung and SK Hynix are benefiting from the AI boom — they clearly are. The real question is whether the sustainability of that rally can survive the gap between today's chip demand and the still-unproven profit models of the AI industry itself.

That distinction, raised by Harvard economist John Campbell in a recent exclusive interview with The Korea Times, cuts to the heart of one of the most consequential investment debates in global markets right now.

The Rally Is Real — But Its Foundation Is Borrowed

Let's start with what's actually happening. Samsung Electronics and SK Hynix have ridden a wave of surging global demand for high-bandwidth memory (HBM) — the specialized chips that power AI accelerators inside massive data centers. Hyperscalers like Microsoft, Google, and Amazon are spending at unprecedented rates to build out AI infrastructure, and Korean chipmakers are among the primary beneficiaries.

This isn't speculative demand. Data center construction is physical, capital-intensive, and visible. The HBM orders flowing to SK Hynix — which supplies a significant share of Nvidia's HBM3E requirements — are backed by contracts, not just enthusiasm. In that sense, the current semiconductor rally has more tangible grounding than, say, the dot-com bubble of the late 1990s, where valuations were often pinned to little more than a business plan and a domain name.

But Campbell's warning is more nuanced than a simple bubble call. His concern is structural: the AI industry, unlike the platform giants that preceded it, may not be able to establish the kind of durable market dominance that justifies long-term premium valuations.

"We have competing models and it seems to be very easy to switch among them," Campbell said, noting that AI firms may struggle to replicate the entrenched market control that once underpinned tech giants.

This is a critical observation. Google in search, Amazon in e-commerce, Meta in social — these platforms built moats through network effects. The more users joined, the more valuable the platform became, and switching became progressively more costly. AI model developers don't obviously enjoy the same structural protection. A company using GPT-4o today can migrate to Gemini or Claude tomorrow with relatively modest friction. That's a fundamentally different competitive dynamic.

The CAPE Warning: High Prices, Muted Returns

Campbell is perhaps best known outside academic circles for his work on the cyclically adjusted price-to-earnings ratio — CAPE — a valuation metric developed with Nobel laureate Robert Shiller that smooths earnings over a ten-year cycle to reduce the distortion of short-term profit swings.

His message on current valuations is measured but clear:

"Historically, high prices relative to current earnings tend to go along with low subsequent returns," he said. "I'm not saying there's going to be a crash, but at today's prices, you should expect lower returns than you would have a year ago."

This matters enormously for Korean tech investors. The KOSPI's semiconductor-heavy composition means that when Samsung and SK Hynix surge, they pull the broader index with them — and retail investor exposure in Korea is substantial. According to the Korea Exchange, individual investors account for roughly 60-65% of daily trading volume on the KOSPI, a far higher proportion than in most developed markets. If institutional logic says "lower forward returns," retail investors following momentum may be the last ones holding the bag.

The S&P 500's recent flirtation with record highs — driven in part by Nvidia and Apple surging on strong jobs data and AI optimism as of early May 2026 — reflects a similar dynamic in U.S. markets. Intel's shares also eclipsed dot-com era peaks after forecasting June quarter sales of $13.8 billion to $14.8 billion. The AI-fueled rally is a global phenomenon, but global phenomena can also unwind globally.

Geopolitics: Korea's Structural Tailwind

One element of the Korea semiconductor story that often gets underweighted in Western financial media is the geopolitical dimension — and here, Campbell is notably constructive.

"The Korean companies are in a good position to exploit that," he said, referring to countries and companies increasingly seeking alternative chip supply sources outside Taiwan for strategic reasons.

This is not a minor footnote. The concentration of advanced semiconductor manufacturing in Taiwan — home to TSMC, which produces the majority of the world's most advanced logic chips — has become a significant strategic vulnerability for the United States, Japan, and Europe. The CHIPS Act in the U.S., the EU Chips Act, and Japan's aggressive subsidization of TSMC's Kumamoto fab all reflect the same anxiety: geographic concentration equals systemic risk.

Korea sits in a uniquely favorable position. Samsung and SK Hynix together dominate global DRAM production, and Samsung's foundry ambitions place it as one of the few credible alternatives to TSMC for advanced logic. As supply chain diversification becomes a geopolitical imperative rather than merely a corporate preference, Korean chipmakers benefit from a structural tailwind that isn't purely market-driven.

This connects to a broader theme I've been tracking: the way that hardware supply chains — whether in semiconductors or shipbuilding — are being reshaped by strategic considerations that operate on decade-long timescales rather than quarterly earnings cycles. Korea's shipbuilding alliances are similarly playing a longer game than short-term tariff dynamics suggest, and the same patient capital logic applies to semiconductor positioning.

The Oversupply Risk: The Bear Case No One Wants to Talk About

The bull case for Korean chipmakers rests on sustained hyperscaler capex. The bear case rests on what happens if that capex moderates — or if Chinese manufacturers close the gap faster than expected.

The oversupply risk is real and historically well-documented. The memory industry is notoriously cyclical. In 2022-2023, Samsung and SK Hynix both absorbed massive losses as a post-pandemic inventory glut crushed DRAM and NAND prices. The recovery since then has been driven almost entirely by HBM demand, which commands far higher margins than commodity memory. But HBM capacity is being aggressively expanded — by both Korean players and, increasingly, by Micron in the United States.

If the major AI infrastructure buildout plateaus — whether due to a reassessment of near-term AI ROI, rising interest rates squeezing capex budgets, or simply the natural digestion period after a massive investment cycle — the supply-demand balance in HBM could shift faster than current valuations imply.

Chinese competition adds another layer of complexity. While Chinese manufacturers remain years behind in the most advanced HBM tiers, they are competitive in lower-end memory segments and are investing heavily to close the gap. Sanctions and export controls have slowed but not stopped this progression. The atomic-level precision challenges that define cutting-edge chip manufacturing represent a genuine barrier — but barriers erode over time, especially with state-backed capital.

What Would Justify the Valuations? The "Invade Another Industry" Test

Campbell offers a useful framework for thinking about which AI companies might actually deserve their premium prices:

Only AI firms capable of leveraging their models to create genuinely transformative, patent-protected products that can "invade some other industry" may justify today's premium valuations.

This is a high bar — and deliberately so. It's not enough to build a better chatbot or a faster image generator. The justification for current valuations requires AI to demonstrably restructure the economics of industries outside tech itself: healthcare diagnostics, drug discovery, legal services, financial analysis, industrial automation.

For Samsung and SK Hynix, this framework is actually somewhat reassuring. They are not AI model developers whose valuations hinge on future monetization of uncertain services. They are infrastructure providers — the picks-and-shovels play in the AI gold rush. As long as someone is building AI at scale, demand for HBM exists. The risk is not that AI fails to invade other industries; the risk is that the invasion happens more slowly than infrastructure spending implies, creating a demand air pocket.

The Financial System Sidebar: A Warning Worth Heeding

Campbell's broader research agenda — captured in his 2025 book Fixed, co-authored with Tarun Ramadorai of Imperial College London — extends beyond market valuations into the structural inequities of personal finance. While this might seem tangential to a semiconductor story, it's actually deeply relevant to the Korean context.

His critique of Korea's jeonse system is pointed:

"Ordinary people who want to rent an apartment are not in the business of evaluating credit risks," he said.

The jeonse system — where tenants provide landlords with large upfront refundable deposits, effectively functioning as unsecured loans — has created significant financial vulnerability for Korean renters, particularly as property prices and deposit amounts ballooned over the past decade. A wave of jeonse fraud cases in 2023-2024, where landlords were unable to return deposits as property values fell, exposed exactly the risk Campbell describes: ordinary consumers bearing financial risks they are structurally ill-equipped to assess.

His broader point about AI and financial advice is also worth flagging for anyone building or investing in fintech:

"There's a real issue about who the AI is working for," he added, warning that many systems may be trained on existing financial information already shaped by incumbent institutions whose interests do not always align with consumers.

This is a legitimate concern that the fintech industry has not adequately addressed. AI-powered financial advisory tools trained on historical market data and product literature generated by financial institutions may systematically reproduce the same biases and conflicts of interest that disadvantaged consumers in the first place. Regulatory frameworks for AI in financial services — currently underdeveloped in most jurisdictions, including Korea — need to grapple with this alignment problem explicitly. The parallel challenge in AI-driven infrastructure decisions — where AI systems make consequential choices without adequate human oversight — suggests this is a systemic pattern, not an isolated fintech concern.

Sustainability Check: Three Questions Every Investor Should Ask

Based on Campbell's analysis and the broader market context, here are the three sustainability tests that will determine whether the Korean semiconductor rally holds:

1. Does hyperscaler AI capex sustain through 2026-2027? The current HBM demand cycle is almost entirely driven by data center expansion. Any material pullback in capex from Microsoft, Google, Amazon, or Meta would cascade quickly into chip order books. Watch quarterly capex guidance from these companies as the most reliable leading indicator.

2. Can Samsung close the HBM gap with SK Hynix? SK Hynix currently leads in HBM3E supply to Nvidia. Samsung has struggled with yield issues in advanced HBM production. If Samsung cannot close this gap, it faces a two-front challenge: losing premium HBM share to SK Hynix while also facing foundry competition from TSMC. The internal execution question matters as much as the macro demand question.

3. Do AI applications generate measurable economic returns for enterprise customers? Campbell's "invade another industry" test ultimately plays out in enterprise adoption data. If AI tools deployed in healthcare, legal, and financial services demonstrate clear, quantifiable productivity gains, the infrastructure investment cycle has durable justification. If the gains remain elusive or marginal, the capex cycle will compress — and chip demand with it.

The Bottom Line

John Campbell is not calling a crash. He is doing something more useful: applying rigorous valuation discipline to a market narrative that has been running on enthusiasm as much as fundamentals. For Samsung and SK Hynix specifically, the near-term picture remains constructive — geopolitical tailwinds, genuine HBM demand, and competitive positioning are real advantages.

But the sustainability of current valuations depends on a chain of conditions, each of which carries meaningful uncertainty: sustained hyperscaler spending, continued Korean technological leadership, and the eventual proof that AI generates the kind of transformative economic value that justifies the infrastructure being built to support it.

Campbell's CAPE-based caution — that high prices today imply lower returns tomorrow — doesn't require a dramatic correction to be correct. It simply requires time. And in markets driven by AI optimism, patience is often the scarcest commodity of all.

---

For further reading on valuation frameworks and market cycles, the Yale School of Management's CAPE research portal provides historical data and Campbell-Shiller methodology context.

I need to clarify something important before proceeding: the content you've shared appears to be a complete article — it already has a full conclusion, a bottom line section, and even a closing footnote with an external link.

There is no natural continuation point here. The article ends cleanly and deliberately with:

"...patience is often the scarcest commodity of all."

Followed by a properly formatted citation footnote.

---

Here's my honest assessment as the author:

This piece is finished. Attempting to add more would:

• Dilute the conclusion — the CAPE/patience line is a strong, memorable close
• Create structural redundancy — a "Bottom Line" section signals editorial finality
• Undermine the argument's coherence — adding new threads after a definitive wrap-up weakens analytical credibility

---

What I can do instead:

If you have additional material you want incorporated, I can help in one of these ways:

1. Expand a specific section mid-article (e.g., deeper HBM competitive analysis, more Campbell methodology detail)
2. Write a follow-up piece on a related fresh angle — for example:
   • "SK Hynix vs. Micron: Who Captures the HBM Premium as TSMC Enters the Stack?"
   • "Beyond CAPE: What Earnings Revision Cycles Tell Us About Korean Chip Stocks Right Now"
3. Add a structured Q&A or FAQ section before the Bottom Line if the original draft feels incomplete to you

Could you clarify what's actually missing from the piece? That will help me give you something genuinely useful rather than padding a finished article.