The Storage and Memory Signal

Direction: Worsening.

This period: Samsung reportedly asked customers for 2027 HBM4 pricing above three times current HBM3E levels, around $4.50 to $4.90 per gigabit versus roughly $1.50 today, according to Digitimes and other supply-chain reporting October 6-7. That follows Micron's own pricing reset disclosed on its earnings call last issue, meaning two of the three major HBM makers are now reported to be setting 2027 contract pricing well above 2026 levels, not just one. On the conventional side, Toshiba said October 2 it will invest about ¥60 billion (roughly $380 million) to double AI-data-center HDD production capacity by fiscal 2027, the first real capacity response this newsletter has tracked from any of the three major hard-drive makers to a market Seagate and Western Digital have both called sold out. That capacity won't ship for at least a year, so it changes nothing about near-term pricing power yet.

Driver: Unchanged since Issue 03: HBM production is still eating disproportionately into wafer capacity that would otherwise make conventional DRAM, and now a second maker besides Micron is reportedly pricing 2027 HBM output on that basis rather than treating this year's squeeze as temporary.

Watch for: SK hynix's Q3 2026 earnings call, expected around October 27, and Samsung's, expected around October 29, the first chance to see whether either company's own guidance language confirms the pricing-reset reporting now circulating about both of them.

On October 8, Solidigm, SK hynix's NAND flash subsidiary, named Goldman Sachs and Morgan Stanley as lead underwriters for a planned US initial public offering, according to Bloomberg and multiple Korean outlets. The reported target is roughly $10 billion raised, at a valuation of up to $100 billion, with a listing penciled in for 2027. It's the first concrete, bank-specific detail this newsletter has been able to report in a story it has now tracked across four straight issues of SK hynix saying only that it's "reviewing various funding options."

That $100 billion ceiling is notably lower than the $100 billion to $150 billion range Reuters and Bloomberg floated when this newsletter covered the story in Issue 09's Deep Cut, "Solidigm's $150 Billion Question." Whether that's a sign the earlier number was always the optimistic end of a wide range, or that the bankers now actually running the process have talked it down toward something they're more confident of pricing, isn't something either side is saying publicly.

The bank announcement reopened a fight that has nothing to do with NAND flash economics and everything to do with how SK Group is structured. The Korea Times reported October 8 that SK hynix shareholders and governance critics are warning a standalone Solidigm listing would deepen SK Group's already long chain of parent-subsidiary listings, a structure Korean regulators have spent years trying to discourage because it lets the same underlying assets get valued, and traded, more than once under different tickers. Because Solidigm's revenue and operating profit are already consolidated into SK hynix's own results, a separate Solidigm stock is, in a real sense, a second claim on value SK hynix shareholders already own. Lawmaker Ahn Cheol-soo has publicly pressed SK Group chairman Chey Tae-won to clarify the company's plans, and minority shareholders are reportedly organizing to oppose the listing outright. This isn't a new argument, this newsletter flagged an earlier round of the same backlash back in August, but naming real banks for a real timeline gave it a sharper edge than "reviewing options" ever did.

SK hynix's own public position hasn't moved an inch. The company said as recently as October 1, covered in Issue 10, that "nothing specific has been decided" and that it's merely reviewing options to strengthen Solidigm's competitiveness, repeating language it has now used for roughly four months running. Solidigm itself is reportedly still without a confirmed US NAND fab, a separate thread this newsletter flagged in Issue 08, with upstate New York still mentioned as a candidate site.

A Solidigm listing matters to this beat less as a Wall Street story than as a NAND-supply story. An independently capitalized Solidigm, with its own stock to issue and its own balance sheet to borrow against, is a company that can fund a US fab a lot more easily than one waiting its turn against its parent's HBM-heavy capital budget. Naming Goldman Sachs and Morgan Stanley doesn't guarantee that fab gets built, or that the IPO happens on the timeline now being reported. It does mean the question has shifted from whether SK hynix spins Solidigm out, to how fast, and how much of SK Group's own governance baggage comes along for the ride.

Toshiba said October 2 it will spend about ¥60 billion, roughly $380 million, to double hard-disk-drive production capacity for AI data centers by fiscal 2027, its first major HDD investment in roughly five years. The expansion runs through its Philippine manufacturing plant, where new lines promise up to 40% higher per-unit capacity and more automated inspection, feeding a roadmap toward 65-terabyte-class drives by 2030 and 100-terabyte-class drives after that. Toshiba currently holds a little over 10% of the HDD market by capacity shipped; the stated goal is roughly 30% over the medium term.

The market's answer arrived within hours: Seagate shares fell roughly 10.2% and Western Digital about 10.2% the same day, wiping billions off both companies' market value. This newsletter has tracked the HDD shortage since Issue 05, when a 30TB-class enterprise drive was already pricing 146% above a year earlier and both incumbents said they were sold out through 2026 and booking into 2027 and 2028. Toshiba's move is the first real capacity response to that shortage this newsletter has seen from any of the three major HDD makers, rather than the usual story of everyone staying sold out and letting price do the rationing instead.

The investors selling Seagate and Western Digital on the news are betting on a future the capacity itself won't touch for at least a year. Toshiba's new lines target completion by fiscal 2027, and qualifying AI-grade drives at volume typically takes longer than that; the 65-terabyte roadmap doesn't land until 2030. In the near term, Seagate's and Western Digital's own sold-out guidance into 2027 and 2028 hasn't changed, and Seagate's next earnings call, scheduled October 28, is the first chance to hear whether the company sounds any less confident about pricing power now that a third major maker says it's finally building more supply.