Newsletter · 2026-09-19 · Ashutosh Agarwal

SK Hynix Eyes Intel's Ohio Fab as an AI Slowdown Scare Sinks Memory Stocks - HBM & The Memory Supercycle - Week of September 19, 2026

HBM & The Memory Supercycle for the week of September 12 to September 19, 2026: SK Hynix is in early talks to make memory in the U.S. for the first time, most likely by leasing part of Intel's delayed Ohio fab under a threat of up to 100% tariffs, while memory stocks were crushed by an AI slowdown scare that bulls on the podcasts call a gift and analysts flagged the memory tax squeezing hyperscaler margins.

HBM & The Memory Supercycle

Week of September 19, 2026: SK Hynix Eyes Intel's Ohio Fab as an AI Slowdown Scare Sinks Memory Stocks

The world's biggest maker of AI memory just floated the idea of building it in America, while its stock, and every other memory name, got knocked down by a scare that has nothing to do with how much memory the world needs.

That's the week in one sentence. Most of the commentary below comes from pundits rather than insiders, and each section flags which is which.

TL;DR

• SK Hynix, the leading supplier of the memory that sits next to Nvidia's chips, is in early talks to make memory in the U.S. for the first time ever, most likely by leasing part of Intel's delayed Ohio factory. The trigger: a Washington threat of up to 100% tariffs on Korean chipmakers who don't build stateside.
• Memory stocks got crushed this week, not on any bad memory news, but on a broader "let's slow down AI" scare started by the CEOs of Anthropic, OpenAI and others. The memory bulls on the podcasts say the underlying shortage is worse, not better, and are treating the dip as a gift.
• The "memory tax" keeps showing up in other people's margins. One analyst flagged that memory from Micron is now "hitting peak headwinds" and squeezing hyperscaler profits. The bottleneck is spreading, not fading.

What's New

1. SK Hynix Wants to Make Memory in America (Operator News, Secondhand)

This is the one genuinely new corporate development of the week, and it's a big one.

On The Rundown (Sept 16), the host walked through a Reuters report that Intel and SK Hynix are discussing a chip-manufacturing deal. Why it matters, in plain terms: SK Hynix is Korean, it's "one of the biggest memory chip makers in the world," and, critically, it's "the leading supplier of HBM, which is high bandwidth memory that sits next to NVIDIA GPUs inside AI servers." (HBM is the fast, stacked memory that feeds an AI chip; think of it as the AI accelerator's short-term memory.) Today SK Hynix makes essentially all of that in South Korea.

Now, per the report, it's considering U.S. production "for the first time ever." Two options are on the table:

• Lease part of Intel's massive Ohio chip factory, a plant that has "already been delayed until 2030" and badly needs a marquee tenant.
• A three-way joint venture between "Intel, SK Hynix, and a hyperscaler that wants to lock in memory supply." Read that again: a cloud giant helping fund a memory fab just to guarantee it gets chips. That tells you how tight supply is.

The stick behind the deal is political. As the host put it: "Commerce Secretary Howard Lutnik has threatened up to a 100% tariff on Korean chip makers who don't build in the U.S." Both Intel and SK Hynix rose about 3% on the report.

And the one-liner that captures the whole supercycle: "there is a huge shortage of memory as AI data center buildout keeps accelerating. That's why the price of RAM costs more than rent these days."

Why it moves the thesis: For years the memory-supercycle story has been a Korea-and-Taiwan story. If SK Hynix actually breaks ground in Ohio, it reshapes the geography of memory, throws a lifeline to Intel's foundry ambitions, and signals that the biggest player expects demand durable enough to justify building on expensive new soil under political duress. Caveat: talks are early, nothing is signed, and this reached the podcasts as a syndicated news item (it echoed across a dozen aggregator feeds this week, e.g. Personal AI), not from anyone inside either company.

2. The AI Slowdown Scare Gutted Memory Stocks (the Week's Dominant Tape Story)

Here's the odd part: the biggest price action in memory this week came from news that had nothing to do with memory.

On The MoneyFlows Show (Ep. 42, Sept 17), hosts Jason Bodner and Lucas Downey opened with it: "We're hearing from CEOs like Anthropic, OpenAI, and Grok that we need to slow down on AI development. This caused stocks to get absolutely crushed, including memory stocks."

Their take (and they're pundits, not insiders) is that Wall Street misread it badly. Bodner's read: "this call to slow down is ultimately a call for regulation, which benefits these massive AI companies... What it's not doing is cutting back the demand for AI. This bottleneck is way worse than Wall Street realizes."

Why it matters: it explains the disconnect between falling stocks and a tightening physical market. If you believe the demand is unchanged and the sellers were spooked by a talking point, the swoon is an entry. If you think the AI leaders know something the rest of us don't, it's a warning shot. That's the whole debate, and it played out live this week (see below).

3. The Memory Tax Keeps Eating Other People's Margins (Analyst)

On Investing Experts (Sept 13), Joe Albano of TechCache laid out how rising memory prices are quietly squeezing the companies that buy memory. His words: "we're getting headwinds from the cost of this compute. DRAM and NAND memory from Micron is now hitting peak headwinds. And we're seeing compressing margins. So not only are we investing more, we're getting less out of it margin-wise."

(DRAM is the everyday working memory in phones and servers; NAND is the flash storage that keeps your data when the power's off. Both are made by the same handful of companies feeding the AI boom.)

He tied it to a broader worry: hyperscaler spending is now "substantially outpacing the growth of the AI," pushing some giants into negative free cash flow "for the first time in some of these companies' history," with Google and Meta "taking out debt." His memorable line on the circular financing propping up custom AI chips: "I think it's a house of cards in the financing. And as long as nobody blows on it, it's going to be fine."

Why it matters: this is the bear-adjacent read on the same fact the bulls love. Memory pricing power is real (you can see it in Micron's numbers), but it's landing as a cost on the hyperscalers, and that cost is now big enough to bend their cash flows. Memory's gain is somebody else's pain, and that tension is the crux of the whole trade.

4. The MAG-3 Framing, 965 Billion Dollars in Profit (Pundit)

A nice bit of framing from Follow the Money with Jerry Robinson (FTM 518, Sept 16). The guest dubbed TSMC, Samsung and SK Hynix the "MAG-3" ("the picks and shovels of AI") and put a number on it: "in between this year and next year, the MAG-3 are expected to make $965 billion of profits, not revenue, but profits... It's more than Berkshire Hathaway's made ever combined. It's three times as much as Amazon's ever made." Those three now make up about 33% of the entire emerging-market index.

Why it matters: it's a reminder that the memory supercycle is, in profit terms, one of the largest wealth-creation events happening anywhere, and most of it is being booked in Korea and Taiwan, not the U.S. That's exactly the imbalance the SK Hynix-Ohio talks (above) are starting to poke at.

The Debate

Same fault line as ever: is this a multi-year structural shortage, or a classic memory cycle about to roll over? This week it played out almost entirely through pundits. Here's the steel-man on each side.

The structural bull case (shortage is real, selloff is a gift):

• On The Compound and Friends (Sept 18), Dan Ives gave the sharpest number of the week: "Demand to supply right now for chips is called 13 to 1... There are 13 orders for every one chip that can be produced." His timeline: "you're not going to have true equilibrium probably till early 2029 at this pace." On our beat specifically: "memory stocks are kind of front and center there. If you look at Micron... quarter by quarter these companies are going to continue to prove out that this." Co-host Tom Lee's frame: "you're never at a top when people want PE to be low and earnings are going up, unless it's a true deep cyclical."
• The MoneyFlows hosts pointed to the physics of the squeeze: making HBM devours DRAM factories. As Bodner explained it, HBM accounts for "about 30% of global DRAM" capacity but "only outputs about 13% of DRAM's capabilities," "a crunch within a crunch." And building more takes years: "It takes three years to make a plan [a fab]. And then next-gen AI arrives during that time and it has more and more needs." His color anecdote (secondhand, but telling): Jensen Huang reportedly walked up to the SK Hynix booth at a recent event and signed a wafer that said "please make more."
• Two demand data points that keep circulating (both secondhand quotes of operators): Elon Musk, per MoneyFlows, said on a SpaceX call that "memory output is rising 20% a year while demand is rising 200%." And Tim Cook's "hundred-year flood" line got a full replay on The Minority Mindset Show (Sept 15), including the exchange where Cook is asked whether laptop prices could double and answers "a hundred percent... Not enough memory."

The cyclical / caution case (this always ends in a glut):

• Joe Albano (above) is the closest thing to a bear on the tape this week, not on memory demand, but on the financing holding up the buyers of memory. His "house of cards" and "at what point does that money really run out?" framing is the reminder that a shortage funded by debt and circular deals can unwind fast if sentiment cracks.
• The "AI slowdown" chorus itself (Anthropic, OpenAI and others publicly urging a slower build) is the bear's exhibit A, even if the bulls dismiss it as a regulatory play. If those calls are sincere rather than strategic, the demand curve that justifies "sold out to 2028" gets a lot wobblier.
• Even Tom Lee, a bull, named the two things that would kill this market: "a bona fide bubble... if one day everyone says there's only one AI model we want to use and then CapEx goes to zero," and Fed tightening. Both are live risks with a rate decision looming.

Where both sides still agree: nobody on this week's podcasts argued the physical shortage unwinds this year. The fight, as always, is about 2028–2029 and about what these stocks are worth after their enormous run.

Stocks in Play

Three memory names got real, by-ticker attention this week, all from the MoneyFlows stock pitch, so treat the price targets as one pundit shop's view, not gospel. Prices are as quoted on the Sept 17 episode.

Also central to the beat: SK Hynix (the Ohio story above; still Nvidia's dominant HBM supplier) and Samsung (named as an HBM/memory leader).

Read-Throughs

• GPU makers (Nvidia, AMD): Nvidia's Jensen Huang appeared on All-In (Sept 14) and discussed memory in passing, listing "memory companies" among the suppliers (Corning, Lumentum, TSMC) he "started working with long before the revolution... came." His broader message, that he hunts the whole supply chain for "bottlenecks" and "constraints," is consistent with the shortage story. The strongest Nvidia-memory line of the week was the secondhand "please make more" wafer anecdote (MoneyFlows).
• PC / handset / consumer OEMs: The clearest pass-through color came from the Minority Mindset replay of Tim Cook ("a hundred percent" that laptop prices could jump on the memory shortage) and the claim, attributed to Nvidia, that "memory costs more than the actual processors" in an AI server. A pundit pricing figure worth noting but not over-trusting: that show also cited "certain memory chip prices... up by around 90% just in the last quarter." No dated, verifiable contract sheet behind it.

A Curveball Worth a Footnote, Helium

One genuinely new angle surfaced on the Minority Mindset show (pundit, unverified, file under "monitor, don't trade"): the host argued that a bombed helium facility in Qatar during the Middle East conflict has tightened the global helium supply, and helium is "needed to keep the temperature cold when you're producing these memory chips." If real, it's another supply-side pinch layered on top of the demand boom. It has not been corroborated by any operator or specialist source, so treat it as a thread to pull, not a fact.

What Changed vs Last Week

Last week (issue #10) featured two operator interviews: Micron's Jeremy Werner on HBM4 bandwidth, and Arm's Rene Haas confirming memory prices had "nearly doubled" and were hitting phones. This week the value shifted from company voices to two things: a genuine news development (SK Hynix possibly building in Ohio, driven by the 100% tariff threat) and a sentiment shock (the "AI slowdown" call gutting memory stocks). The debate rotated back to the oldest question on this beat: is the selloff a gift or a warning?

The calendar tightens from here: Micron reports FQ4 on September 30, the first hard print in a month, and the single best test of whether the "memory tax" is still compounding or finally cresting.