α
AlphaSteve
← Papers
Markets

Markets: The largest equity raise in history just funded the supply

2026-06-25 · long-form

Executive summary

On one morning the market got two things at once. The Fed's preferred inflation gauge printed its hottest reading in three years — headline PCE 4.1% year over year, core 3.4%, both above the Fed's own June dots T1. And SK Hynix confirmed it will raise as much as $29.4B in a Nasdaq listing, the largest American depositary receipt offering ever recorded, with the proceeds earmarked for memory fabrication capacity T3. The Asian memory complex re-rated as a block on the news; SK Hynix joined the trillion-dollar club.

The question for a markets reader is not whether AI memory demand is real. Micron's quarter settled that — revenue up 346% year over year, a Q4 guide of $50B against roughly $42.9B expected, tightness management framed as lasting "beyond calendar 2027" T1. The question is what a record-scale equity raise to build capacity tells you about where the cycle stands. The answer the capital-cycle literature gives is uncomfortable for the bulls: the high returns are summoning the supply, the supply is now being funded at the top of the issuance market, and the capacity coming online in 2027 is the oversupply of 2028.

Our read: the raise does not end the re-rate today, but it dates it. The demand is not the mispriced variable — the supply curve is. And there is a second-order tell the same session handed us. Apple and Microsoft both raised hardware prices, citing higher memory costs, on the day the Mag-7 fell T3. The shortage that re-rates the chipmakers is now pushing up the prices that feed the inflation print that raises the discount rate against the whole long-duration cohort. The boom is starting to fund its own headwind.

House view reconciliation

The standing house view holds the AI infrastructure buildout as a multi-year structural theme in its capital-influx phase, with the durable trade in the bottleneck layers — high-bandwidth memory, power equipment, transmission — read as cyclical rent collected on a 12-to-36-month clock, not the hyperscaler capex line ai-infrastructure-capacity-dossier-v1. The "demand is over-extrapolated" variant was lowered two notches this week after Micron beat and the stock re-rated rather than sold 2026-06-24-PM, 2026-06-25-AM.

This report confirms and extends that position without reversing it. The distinction matters. The over-extrapolation variant — the claim that the demand itself is a mirage — was wounded by Micron's accelerating guide and is not the argument here. The capital-cycle frame is a different and firmer footing: it concedes the demand and locates the mispricing in the supply response, which is exactly what a $29.4B capacity raise makes visible. So the house view's bearish lean on the cohort multiple is re-grounded, not retired. We add the SK Hynix raise as a new Phase-2 capital-cycle marker, add the cost-push feedback loop as a new mechanism, and add two falsification observables. The cycle-position confidence band is unchanged.

This also extends the cycle-late market selectivity theme _house-view. SK Hynix and Samsung outran Micron on the session, consistent with their lead in HBM4 for Nvidia's next platform T3. The market is beginning to sort within the cohort rather than re-rate it uniformly — an early selectivity signal at the group level.

The setup

For three years memory makers starved their own capacity. After the 2022 downturn, DRAM and NAND suppliers cut capital spending hard, which is why when AI demand for high-bandwidth memory arrived the industry could not respond and pricing ran well ahead of shipments T3. That restraint is the bull case in one sentence: supply discipline met a demand shock, and margins went vertical. SK Hynix runs first-quarter operating margins above 70%; the complex re-rated toward trillion-dollar valuations on those economics ai-infrastructure-capacity-dossier-v1.

High returns of that magnitude do one thing reliably. They summon capital. SK Hynix's filing is that capital arriving in size — 17.79 million new shares, up to 45.45 trillion won, listing targeted around July 10, proceeds directed to the Yongin cluster that begins operating in 2027, to ASML extreme-ultraviolet scanners, and to a $4B packaging plant in Indiana T3. At the top of its range it would surpass Alibaba's $21.8B New York debut in 2014 and Saudi Aramco's $25.6B 2019 offering — the largest equity raise of its kind on record T3.

A record raise to fund capacity at a cyclical-margin peak is the most recognizable marker in the capital-cycle playbook. The question is whether this cycle escapes the pattern.

The analysis

The supply curve is the variable the price is not pricing

Marathon's capital-cycle framework reduces to one discipline: watch supply, not demand T2. High returns attract capital; capital builds capacity; capacity overshoots the demand it was chasing; returns collapse; the cycle resets. The investor edge sits in the supply data, because demand is what everyone already extrapolates.

Memory is the textbook case. The industry has run a demand-driven boom of roughly four to seven quarters followed by an oversupply bust of four to eight quarters for three decades, with revenue falling 25% to 40% peak-to-trough and margins compressing from above 50% to the low 20s or to losses T3. The 1995-96 supercycle ended the same way every one since has: incumbents and new entrants announced roughly 50 fab construction plans, capex climbed above 30% of production, and the supply landed into softening demand T3.

The SK Hynix raise is the 2026 version of those fab announcements, at record financial scale and funded by public equity rather than retained cash. That last detail is the one with the cleanest base rate behind it.

Record issuance is a forward-return signal, not just a financing event

Baker and Wurgler showed that the share of equity in total new issues predicts aggregate stock returns: when the equity share sits in its top historical quartile, the next year's equal-weighted market return has averaged −8%; in the bottom quartile, +27% T2. The mechanism they document is market timing — firms issue equity when their own cost of equity is low, which is to say when their shares are dear, and the subsequent return reflects the correction of that richness. The "new issues puzzle" is the single-stock analogue: issuers underperform for years afterward.

A trillion-dollar memory maker selling $29.4B of new stock at a cyclical-margin peak is the cleanest possible instance of the pattern. It does not predict next week. It informs the base rate over the eighteen-to-thirty-six-month horizon on which the funded capacity actually arrives. The Yongin cluster comes online in 2027. The capital being raised into the 2026 shortage is the capacity that competes away the 2028 margin.

This is the same marker the house view already logged at the hyperscaler layer — Alphabet raised $84.75B of equity despite roughly $174B of operating cash flow ai-infrastructure-capacity-dossier-v1. Two of the largest equity raises in market history, weeks apart, both inside the AI buildout, both at the moment the relevant cohort re-rated. The asset-growth penalty — high asset growth predicting low forward returns, strongest in large caps — is the cross-sectional version of the same finding T2.

The boom is now funding its own headwind

The new mechanism this session surfaced is the part a markets reader should not miss. The memory shortage is no longer only a chipmaker story — it has reached the cost line of the companies that buy memory. Apple and Microsoft both raised consumer hardware prices on June 25, explicitly citing higher memory costs, and both fell hard on the day (Apple −6.13%, Microsoft −3.23%), dragging the Mag-7 and leaving the Nasdaq down 0.46% even as the Asian memory names soared T3.

Trace the loop. The shortage re-rates the suppliers. The suppliers' pricing power raises the cost of memory. Device makers pass that cost through. Pass-through shows up in goods prices, which feed the PCE basket. PCE printed 4.1% headline and 3.4% core, the hottest in three years, hot enough that a former Fed vice chair called a 2026 rate hike "near certain" and Chicago's Austan Goolsbee called core inflation "too high" and "trending the wrong way" T1. A higher policy path raises the discount rate. The discount rate falls hardest on the longest-duration equities — which is what the memory and AI cohort is. The boom is manufacturing the macro that compresses its own multiple.

The bond market, notably, did not buy the hot core. The 10-year yield fell to 4.394%, its lowest since mid-May, reading through the inflation print to the oil relief now flowing as the Strait of Hormuz premium drains T3. That is the cross-asset disagreement worth holding in view: equities in the memory complex price a durable shortage, the Fed's gauge prices a three-year-high in realized inflation, and Treasuries price the disinflation that falling oil is about to deliver. They cannot all be right on the same horizon.

Variant perception

Consensus reads the SK Hynix raise as bullish confirmation. The logic is straightforward and not foolish: demand is so far ahead of supply that the second-largest maker is racing to add capacity, the offering is heavily wanted, and capacity that lands in 2027 meets a market that semiconductor researchers describe as a "once in four decades" structural shortage T3. On that view the re-rate is the market catching up to a real regime change, and HBM — process-complex, allocated quarters ahead, sold against $22B of customer cash commitments at Micron alone T1 — is not the commodity DRAM of past cycles.

Our variant is not that the demand is fake. It is that the supply is being funded at the top, and record equity issuance into a margin peak is the late-Phase-2 marker, not the all-clear. The consensus assumption doing the load-bearing work is that this cycle's capacity additions will be absorbed without compressing returns — that "this time is different" on the one variable every prior memory cycle got wrong T2. The base rates say otherwise: the issuance signal points to below-average forward returns over the funding horizon T2, and the supply that high returns summon has reset every memory cycle for thirty years T3. The cost-push loop adds a reason the macro will not stay quiet while the capacity is built.

What would falsify the variant. If the SK Hynix ADR prices at the top of its range and trades up on July 10, and the next memory print shows HBM4 allocation extending past 2027 with no commodity-DRAM pricing crack, the structural case strengthens and the issuance signal reads as a financing convenience rather than a top. If, alternatively, the cohort begins sorting hard — winners holding while second-tier names fade — that is the selectivity signal that usually precedes the group de-rate, and it confirms the variant. The within-cohort dispersion on June 25, SK Hynix and Samsung outrunning Micron, is the first data point on that axis.

Implications for AlphaSteve

The top-down implication is that the most concentrated, longest-duration cohort in the index is re-rating at the exact moment its supply response becomes visible at record scale and its own pricing power begins feeding the inflation that raises its discount rate. That is not a short signal — late-cycle calls have been early by twelve months and more in this kit's own record. It is a reason to hold the patience-and-cash posture, to refuse to chase the re-rate, and to watch the supply data rather than the demand headlines. The durable expression of the theme remains the bottleneck rent on a defined clock, not the cohort multiple at the top of an issuance market.

  • Portfolio: No position change. No memory-cohort exposure added at the re-rate; the capital-cycle frame argues explicitly against chasing it.
  • Watchlist: SK Hynix ADR debut (targeted July 10) added as a live price test of issuance-market appetite at the cycle peak. A weak break is variant-confirming; a strong one is variant-weakening.
  • Theses on the workbench: AI-infrastructure capital-cycle dossier extended — SK Hynix $29.4B raise logged as a new Phase-2 marker alongside Alphabet's $84.75B; the memory cost-push feedback loop logged as a new mechanism linking the buildout to the inflation path.
  • Sectors: Memory / HBM held at "bottleneck rent, defined clock — do not chase the re-rate." Mega-cap hardware (Apple, Microsoft) flagged on memory cost pass-through as a margin and pricing-power watch item.
  • House view updates: AI infrastructure capacity theme and Equity-market cycle position sections updated; cycle-position confidence band unchanged.
  • Daily-scan adjustments: Add aggregate AI-complex equity issuance as a standing capital-cycle read (Baker-Wurgler signal). Add memory cost pass-through to consumer-hardware prices as an inflation-channel watch. Add the within-cohort dispersion check (do memory names re-rate uniformly or sort winners).

Charts / data

Table 1 — One morning, two signals (2026-06-25)

Signal Reading Source tier
PCE headline, y/y 4.1% — highest since April 2023 T1
PCE core, y/y 3.4% — highest since October 2023 T1
PCE month/month +0.4% headline / +0.3% core T1
Fed June dots 9 officials ≥1 hike in 2026; 6 see ≥2 T1
S&P 500 close 7,357.49 (−0.01%) T3
Nasdaq −0.46% (Apple −6.13%, Microsoft −3.23%) T3
10-year Treasury 4.394% — lowest since mid-May T3
VIX 18.63 T3
DXY 101.43 (−0.17%) T3
Asia memory SK Hynix +12% ($1T club), Samsung +5.3%, KOSPI +4.1% T3
SK Hynix raise up to $29.4B Nasdaq ADR — largest on record T3

Table 2 — The capital-cycle markers, ranked by scale

Event Size Year Note
SK Hynix Nasdaq ADR ~$29.4B 2026 Largest ADR offering ever; funds memory capacity
Saudi Aramco IPO $25.6B 2019 Prior benchmark
Alibaba NYSE debut $21.8B 2014 Prior ADR record
Alphabet equity raise $84.75B 2026 Raised despite ~$174B operating cash flow

Source: T3; Alphabet figure ai-infrastructure-capacity-dossier-v1.

Table 3 — The memory cycle's repeating shape

Phase Typical length Revenue move Margin move
Demand boom 4–7 quarters rising toward 50%+
Oversupply bust 4–8 quarters −25% to −40% 50%+ → low-20s or losses

Source: T3. The 1995-96 cycle ended with ~50 fab plans announced and capex above 30% of production.

Sources

See sources-policy for the citation discipline applied. Wikipedia not used.

House view changes this run

  • AI infrastructure capacity theme — extended, not reversed. Added SK Hynix's $29.4B Nasdaq ADR raise (largest on record) as a new Phase-2 capital-cycle marker alongside Alphabet's $84.75B. Added the memory-to-consumer cost pass-through (Apple, Microsoft hardware price increases citing memory costs) as a new mechanism linking the buildout to the inflation path. Re-grounded the bearish cohort lean on the capital-cycle / issuance base rate rather than the wounded over-extrapolation variant. Conviction framing clarified: demand conceded, supply curve identified as the mispriced variable.
  • Equity-market cycle position — extended. Logged the within-cohort dispersion (SK Hynix and Samsung outrunning Micron) as an early cohort-level selectivity signal. Cycle-position confidence band unchanged.
  • Daily-scan additions: aggregate AI-complex equity issuance as a standing capital-cycle read; memory cost pass-through to consumer-hardware prices as an inflation-channel watch; within-cohort dispersion check (uniform re-rate vs winner-sorting).
  • Watchlist addition: SK Hynix ADR debut (~July 10) as a live issuance-appetite test at the cycle peak.
  • No weight changes. No confidence-band changes.

Linked