Two weeks, two memory makers heading to public markets on opposite sides of the world, and a price split that is finally showing up in the numbers. None of it broke the thesis. Here is what actually happened.

I skipped the round up last week, buried in work setting up more AI agents, so this one covers two weeks and a lot landed. The June drawdown recovered, two of the three makers made moves toward public markets, the biggest desk on the Street put a number on the whole thing, and the pricing data started splitting in a way that is exactly on script. Let me go through it the way I always do. What is noise, what is signal, and what I am actually watching.

1. SK Hynix comes to Wall Street

SK Hynix filed confidentially with the SEC back in March and is now close to clearance, possibly the week of June 22, with an American listing through ADRs potentially debuting as early as August and raising up to around 14 billion dollars. This is a secondary listing, the ADRs covering only about 2 to 3 percent of shares, not a move away from its Korean home.

Here is why it matters more than a normal listing. A huge pool of US money, pension funds and index ETFs, is mandated to hold only US listed stocks. That money literally could not own the best memory maker in the world while it traded only in Seoul. An ADR opens that door, and it hands American investors the actual HBM leader, 57 percent of the market, as something they can buy directly for the first time.

The honest wrinkle is for Micron. Micron has been the only clean US listed pure play in memory, so it carried a premium as the default way for American money to bet on the crunch. When the real number one shows up on the same exchange, and does it trading around 6 to 7 times forward earnings versus Micron near double that, some of that premium has somewhere better to go. That is not memory weakening. It is the menu getting better. Mild headwind for Micron's multiple, bullish for the sector. Worth watching, not a thesis breaker.

2. CXMT goes public

On June 12, China's securities regulator approved CXMT's IPO. The company will raise about 29.5 billion yuan, roughly 4.33 billion dollars, on Shanghai's STAR market, the largest A share listing in China this year. CXMT is the country's largest DRAM maker, founded in 2016, and this raise is the war chest for its expansion.

This is the "when not if" fourth player getting funded, and it is the one slow moving variable I actually respect. But size it correctly. Money buys fabs and wafers. It does not buy EUV lithography, which China still cannot get, and it does not buy the HBM4 process that the high end runs on. CXMT is at rough parity on mainstream desktop DDR5 and only at baby steps on HBM. So the raise accelerates the commodity tier, the bottom of the makers' revenue, not the HBM apex where the real margin lives.

You can already see the commodity tier moving in actual product. Two Chinese brands, Gloway and KingBank, launched 48GB DDR5 kits built entirely on domestic 24Gb chips, almost certainly CXMT, with zero Samsung, Micron, or SK Hynix content. The tell is the 24Gb density, up from 16Gb, because it means the process is progressing, not just the marketing. That is real erosion at the bottom. It is not price moving yet, the report that broke it says exactly that, and it is nowhere near HBM. The thing to watch from CXMT is density and yield climbing toward server and any genuine HBM step, not consumer kit launches, which are now routine.

3. The split gets real, and the thesis picked up a heavyweight backer

Morgan Stanley, analysts Shawn Kim and Joseph Moore, put out a note calling memory structural rather than cyclical. The framing is striking. The memory market goes from about 220 billion dollars in 2025 to roughly 890 billion in 2026 on the numbers in the note, and that 670 billion jump is larger than the entire standalone market for smartphones, or PCs, or servers. The decades long fall in the cost of a gigabyte of DRAM has ended.

The mechanism is the whole argument in one line. Bit supply is growing only about 30 percent a year, capped by EUV tools and wafer capacity, while revenue is rising roughly fourfold. So almost the entire spend surge is price, not volume. And HBM, which consumes three to four times the wafer area per usable bit, climbs from about 6 percent of leading edge memory wafers in 2023 to 34 percent by 2028, which keeps starving everything else.

Why this one counts more than the others I get sent. It is not a fund issuer selling a product, and it is not a maker talking its own book. It is a major sell side desk putting the structural case in writing with numbers behind it. That is the most credible version of this spine I have seen.

Now the part I will not skip, because I do not just print the bull case. Structural is their word, and it describes the next two years, which is exactly the window I trust. It does not repeal the cycle. The same note has wafer capacity expanding about 30 percent by 2027, which is the supply side gearing up, and the price not volume math cuts both ways. It is glorious while prices hold, and it is the textbook definition of a spike that eventually mean reverts. Morgan Stanley also describes a two tier market forming, with hyperscalers locking supply through long term deals and prepayments while everyone else competes for a smaller, more volatile pool. That split is on thesis, and it is also the warning, because the soft, volatile tier is the consumer one, which is exactly where CXMT lands and where any cooling shows up first.

The demand side kept broadening

Two signals this fortnight that the "AI needs more memory everywhere, not just in data centers" pillar is real. Apple is putting a real model behind Siri, and the analyst read is iPhone DRAM content climbing toward 12GB, up from the 6 to 8GB that has been standard, which means more mobile DRAM orders for Samsung and SK Hynix, the only suppliers that clear Apple's spec and scale. And Google ordered more than three million TPUs from Intel for 2028. Every one of those custom accelerators still needs HBM from the same three makers, so a hyperscaler building its own chip is more HBM sockets, not fewer. The escalator I keep describing is now visibly running in phones and in custom silicon, not just in Nvidia GPUs.

The week ahead, Micron reports June 24

This is the one real catalyst in front of us. Consensus is high, the stock has run hard, and the options market is pricing a violent move, near term IV sitting around 125 percent. That means the results themselves, which will look enormous on a year over year basis no matter what, are not the thing that matters. The thing is the guidance, the HBM4 ramp commentary, pricing into 2027, and any hint that the PC and commodity cooling is creeping into server. When a stock is priced for perfection, the headline beat is already expected and the forward tone is what moves it. All bets are live until that print.

Memory pricing update. (2026-06 #2)

The numbers since the last roundup, and there is a real nuance hiding in them. TrendForce's second quarter contract forecast is now 58 to 63 percent quarter on quarter for conventional DRAM, and 70 to 75 percent for NAND. Still enormous, but read the change. Q1 DRAM came in at a record 90 to 95 percent, so the rate of DRAM price growth is actually slowing, even as NAND accelerates past it for the first time this cycle. Prices are still climbing fast. The acceleration is fading on the DRAM side.

Spot is firm. As of mid June, DDR5 spot is still rising with buyers accepting higher quotes, and DDR4 is so tight it is pushing some buyers down to DDR3 and lifting that too. Mainstream DDR4 spot sits around 36 dollars a chip.

The split, in the data. PC DRAM contract momentum is slowing as high prices choke PC sales, with notebook shipments now forecast down about 15 percent for 2026, while server DRAM keeps rising on depleted inventories and stays guided up into the third and fourth quarters. Meaningful new capacity does not arrive until the second half of 2027 at the earliest. One wrinkle worth flagging, SK Hynix's blended pricing in the first quarter was actually held back a touch by easing HBM contract prices, a reminder that even the apex is not a straight line up.

Net: still a shortage, still guided up, but DRAM price growth is decelerating off the Q1 peak, NAND has taken the baton, and the PC versus server split has hardened. That stratification is exactly what this thesis predicts at this stage. The scarce tiers stay tight while the commodity tier cools first.

This all makes me feel

Two weeks, thesis intact. The two headline events, both makers moving toward public markets, are confirmation, not cracks. One unlocks a wall of US capital for the best name in the group, the other funds a competitor that is still years and several process generations away from the part of the market that actually pays. The structural case picked up a heavyweight backer with numbers. Demand kept broadening into phones and custom chips. And the pricing split arrived right on cue.

My tripwire was never the stock price, a scary headline, or a China announcement. It is hyperscaler capex guidance turning down, and this fortnight it did not. So I am holding. Next week's Micron print is the thing to watch, and not for the headline number, but for what the guidance says about how long this runs.

I hold a long position in DRAM ETF (ticker: DRAM) and Micron (ticker: MU) . Not investment advice. Do your own research.