"AI Chip Mania Sows Seeds of Its Own Destruction" is a Streetwise analysis column by James Mackintosh, published in the Wall Street Journal (Markets / Stocks) on May 16, 2026. It argues that the 2026 boom in high-bandwidth-memory (HBM) chip stocks — Micron, Samsung Electronics, and SK Hynix — fits the pattern of a cyclical commodity industry, and that the low forward price-to-earnings (P/E) ratios investors read as cheap have historically coincided with cycle peaks rather than bargains.
Bibliographic record
- Author: James Mackintosh (WSJ Streetwise columnist)
- Outlet / venue: Wall Street Journal — Markets / Stocks
- Published: May 16, 2026
- URL: wsj.com
- Form: Streetwise analysis column with two embedded LSEG charts (share-price change since 2024 for Micron / Samsung / SK Hynix; Micron forward P/E by cycle).
Summary of argument
Mackintosh's column treats memory-chip stocks as the most exposed case of the 2026 AI-chip rally and frames the central question as one of valuation rather than AI demand. He accepts that AI demand is strong — his framing in adjacent WSJ pieces ("the AI frenzy is back") is largely accepting of it — but argues that memory is a commodity whose cycle mechanics AI demand does not change, and that AI-bullish investors have historically misread that cycle at the moments when reading it correctly would have mattered most. The column does not commit to a directional call ("Micron is a sell") but to a valuation-skepticism call ("the low P/E is not a bargain"). It closes: "As with all commodities, success sows the seeds of its own destruction — even if AI hopes are fulfilled."
Memory-chip stocks in 2026
Micron Technology recorded its biggest-ever loss three years before the column and is now forecast to become the sixth-most-profitable U.S. stock, making just under $100 billion over the next 12 months — more than Meta or Berkshire Hathaway. The three HBM makers — Micron, Samsung Electronics, and SK Hynix — are, in Mackintosh's words, "in the sweet spot of the chip cycle," boosting prices, profits, and share prices. The two Korean stocks have made Korea's market the best-performing in the world in 2026, and Micron is contributing materially to Wall Street's upgrades of the S&P 500 earnings outlook.
The cyclical-industry mechanism
Mackintosh's central frame describes the memory business as a textbook commodity cycle: heavy investment is required to build a fab; when demand rises, supply takes years to catch up; during that lag, prices and profits jump; high profits encourage CEOs to expand supply; high fixed costs encourage running fabs at full capacity even after demand peaks; and the cycle turns when excess supply pushes prices and profits down, as in the 2022–23 memory bust. His contribution to this familiar story is the argument that the market already knows the cycle exists but has historically misread its timing.
Capital spending in the current cycle is already committed. Micron is spending roughly $150 billion to build or expand fabs in New York, Idaho, and Virginia, and new Korean fabs (Samsung and SK Hynix domestic capacity) are opening. As Mackintosh puts it, "Already the high profitability has encouraged heavy capital spending."
The cycle-peak P/E pattern
The claim that gives the column its title is that Micron's prior cycle peaks have each occurred at a forward P/E low enough that the market would otherwise call it cheap. Two weeks before May 16, 2026, Micron's forward P/E was under 10×, making it the S&P's third-cheapest stock by that measure. Mackintosh reads a sub-10× forward P/E as consistent with the market pricing in cyclicality, but notes that historically this is also the pattern at cycle peaks, so a cheap-looking valuation cannot be used as a contrarian buy signal: "It just means investors recognize that the boom times in memory chips never last."
| Cycle peak | Forward P/E at peak | Subsequent path |
|---|---|---|
| 1984 | 15× | Stock took 9 years to surpass that level |
| 2018 | 5.5× | "Losses for investors who were fooled into thinking they were buying a bargain were vast" |
| 2022 | 9× | Stock halved that year, then doubled after the loss was priced in |
| Two weeks before May 16, 2026 | Sub-10× (Micron was S&P's third-cheapest stock by forward P/E) | "It just means investors recognize that the boom times in memory chips never last." |
Risks Mackintosh names
Mackintosh identifies the largest risk as memory-efficiency breakthroughs, which he says are "impossible to quantify": "AI technology could become far more efficient in its use of memory, meaning data centers need less of it." He notes that memory stocks took a hit in March 2026 when Alphabet researchers published a paper showing dramatic improvements in memory efficiency, but recovered, and that "engineering improvements for specialized data centers should be expected — but how big they are and when they come is unknowable in advance."
On the demand side, he lists risks to the whole AI supply chain — data-center plans scaled back, AI uptake slower than hoped, and political backlash hindering expansion — observing that "all are plausible; none are considered that serious by the AI bulls driving stock prices."
On the supply side, he points to new entrants in non-memory chips. "Fat margins on Nvidia's chips have persuaded Alphabet to develop … TPUs, dedicated to training AI." Amazon's Graviton chips provide the central processing unit (CPU) for the inference involved in using AI models, demand for which has boosted Intel. Recent entrant Cerebras, which launched the first of its giant chips for both training and inference only in 2019, raised $5.55 billion in its IPO on Thursday (May 14, 2026), and its shares more than doubled immediately.
Mackintosh argues that memory is less differentiated than logic, so the cycle bites harder there. GPUs (Nvidia versus AMD), TPUs (Alphabet), and Graviton (Amazon) are, in his framing, "more differentiated, so they are much less cyclical."
Key claims (confidence-rated)
| Claim | Confidence | Rationale |
|---|---|---|
| Micron's forward P/E was under 10× and was the S&P's third-cheapest stock two weeks before May 16, 2026. | high | WSJ market-data citation; observable; verifiable from market data. |
| Historical Micron cycle-peak P/Es: 1984 = 15×, 2018 = 5.5×, 2022 = 9×. | high | Mackintosh + WSJ chart; consistent with prior memory-bust accounts. |
| Micron is on track to make ~$100B over next 12 months and become the 6th-most-profitable US stock. | medium | Forward earnings forecast subject to revision; cited by Mackintosh from sell-side consensus. |
| Micron is spending ~$150B on US fab buildout (NY, ID, VA). | high | Public capex disclosures; consistent with prior Micron Technology coverage. |
| Korea's stock market is the world's best-performing in 2026 driven by Samsung + SK Hynix. | high | Market-data observable. |
| March 2026 Alphabet "memory efficiency" research paper triggered a memory-stock selloff that has since recovered. | high | Verifiable from market reaction; specifically mentioned in WSJ March 26 live-coverage. |
| Cerebras IPO (May 14, 2026): raised $5.55B; shares more than doubled on debut. | high | Public IPO data. |
| Memory is less differentiated than logic (GPU/TPU/Graviton), hence more cyclical. | high | Standard industry framing; corroborated by structural P/E differences between Micron and Nvidia/AMD. |
| Memory-efficiency breakthroughs could materially reduce data-center memory demand. | medium | The mechanism is real; the magnitude and timing are explicitly unknowable per Mackintosh. |
| Investors have historically misread the chip cycle at cycle peaks. | high | Verifiable from price-action history; the column's core empirical claim. |
| 2026 forward P/E under 10× is a cycle-peak signal rather than a bargain. | contested | Mackintosh's argument by analogy from 1984/2018/2022; the AI-demand-is-different counterclaim is the dominant market view. |
Methodological note
Mackintosh's argument rests on cycle-pattern analogy: there are only three prior cycles to draw on (1984, 2018, 2022), and the cycle-peak P/E pattern (low, not high) is informative but small-N. The column makes a valuation-skepticism call ("the low P/E is not a bargain") rather than a directional one, and stands as a formulation of the chip-cycle-skepticism position rather than a forecast.
Relevance to existing wiki content
The column engages with several existing pages. It is a narrower-bubble position within the AI Bubble Debate — not the whole AI stack, but specifically memory-chip stocks priced as if cyclicality has been suspended. Its "AI bulls driving stock prices" framing identifies the same cohort as the circular-financing thread, but locates the risk differently, in the commodity cycle rather than a financing loop. Micron Technology is the column's empirical anchor; Samsung Semiconductor and SK Hynix — HBM Leader are corroborating cases (Korea best-performing market 2026); Cerebras Systems's IPO is cited as evidence of new-entrant pressure in non-memory chips; and Nvidia & TSMC — AI Compute Infrastructure supplies Mackintosh's "differentiated chips" example, mentioned but not the focus.
The column sits in tension with the data-center buildout narrative (AI Data Centers): Mackintosh argues that a major risk to that buildout is engineering improvements in memory efficiency that the buildout's capex assumptions do not price, with the March 2026 Alphabet memory-efficiency paper as the falsifiable claim. Relative to the bubble-vs-buildout frame, the column occupies a third position: the buildout is real, but the financial expression of it in memory-chip equities reflects a cycle the market historically misreads. It is referenced alongside the WSJ "different this time" frame.
Two contemporaneous events bear on the column's framework. The May 17 Samsung labor strike (45,000 workers; roughly one-third of global DRAM share), covered in the May 17 dev-log, is a structural-supply shock that Mackintosh's framework would treat as short-term-bullish and long-term-irrelevant for the cycle (the constraint loosens once the strike resolves, while long-term capex still creates oversupply); it is tracked in Samsung Semiconductor and AI Data Centers. The Cerebras IPO (May 14, 2026; $5.55B raised, shares more than doubled) is treated by Mackintosh as evidence of new-entrant pressure in non-memory AI silicon and is tracked in Cerebras Systems.
Relationships
- supports: AI Bubble Debate (narrower bubble locus — memory equities specifically).
- contradicts: the implicit market view that AI demand has structurally suspended memory-chip cyclicality.
- related: Micron Technology, Samsung Semiconductor, SK Hynix — HBM Leader, Nvidia & TSMC — AI Compute Infrastructure, Cerebras Systems, AI Data Centers, Circular Financing in AI, AI Bubble vs. Buildout — Synthesis.