Micron (MU) Slips as Intel-Backed Kepler Computing Takes Aim at Memory With 2,000 Wafers to Its Name
Micron slipped on Thursday because a private company that has run about 2,000 wafers in its whole life said it can help end the memory shortage. That’s the trade. Kepler Computing came out of stealth this week with a ferroelectric composite material, a 3D scheme that stacks memory on top of logic, and GlobalFoundries as its manufacturing partner. The headlines called it Intel-backed. It is. It’s also AMD-backed and Gates-backed, and GlobalFoundries has money in it too. Intel Capital is one line on a cap table that has raised $468 million.
Put the 2,000 wafers next to something. A DRAM fab running 45,000 wafer starts a month (the ceiling Nanya set for its planned Fab 5A) gets through that many in about a day and a third. Kepler’s lifetime output would vanish inside one working day at one mid-sized plant. Micron, worth about $1.17 trillion at Wednesday’s close, booked $41.46 billion of revenue in its last reported quarter. Kepler’s entire funding history comes to a bit over 1% of that one quarter. Add the Commerce Department’s letter of intent for up to $245 million and it’s still under 2%.
So what’s the market reacting to? The technology is real enough to take seriously. Kepler says its composite, found after 35 attempts, lets ferroelectric cells read and write at lower voltage and pushes SRAM density to levels it compares with 2nm and 3nm logic, on GlobalFoundries’ 28nm process and without EUV. The two companies say they converted a fab line to next-generation status in eight months, against a usual 24. First HBM samples are promised for later this year, a Singapore ramp in 2027, US production in 2028. GlobalFoundries’ own executive said the physics is solved and what’s left is good results on thousands of wafers and millions of devices. That last part is where memory companies live or die. Yield at volume is the product.
What Kepler threatens is the calendar. Micron’s valuation rests on a shortage that lasts, and its operating margin above 80% exists because the only accepted way to add DRAM supply is a new fab: $20 billion to $40 billion and years of tool installs. Every model that runs the shortage through 2027 and beyond has that assumption buried in it. Kepler’s pitch goes straight at it. Retrofit old trailing-edge lines, get more memory bits out of existing silicon, skip the greenfield wait. If that worked at scale, new supply would show up on a retrofit schedule, and the length of the shortage (which is what Micron’s multiple actually pays for) would shrink. The market doesn’t need Kepler to win. It only needs the timeline to look a little less certain, and on Thursday it did.
The retrofit story has a hole, though, and GlobalFoundries pointed at it. Kepler’s composite contains iron. Fabs work hard to keep iron out, so the process has to run on dedicated tools or be fully encapsulated. Dedicated tools are capex. Not $20 billion of it, but “no new fab” quietly turns into “new equipment inside an old fab”, and the cost advantage narrows with each converted line. It also limits how fast the method can spread to foundries that don’t want iron anywhere near other customers’ wafers.
Then there’s the product. Kepler’s density gains are strongest in SRAM, the cache that sits on or beside a processor. That business belongs to logic foundries and chip designers; Micron doesn’t sell it. In HBM, where Micron makes its money, Kepler hasn’t shipped a sample. HBM is sold after months of qualification with Nvidia, AMD and the custom-silicon houses, and 2026 supply was spoken for long ago. A new entrant needs a qualified part and a customer willing to name it before it takes a single bit of share. Kepler has neither yet.
The Intel label is doing work in the headline, so it’s worth a closer look. Intel’s own memory ambitions are loud right now. Lip-Bu Tan calls new memory architectures a pet project, Intel’s XBM patent for an interposer-free HBM alternative is public, it co-develops Z-Angle Memory with SoftBank’s Saimemory, and it hired former SK Hynix chief Seok-Hee Lee in June. Its CFO has since said Intel will go after DRAM through partners instead of its own fabs. Read that way, Kepler is one more name on the partner list, and “Intel-backed” lets a start-up borrow the aura of a company that just raised $20 billion. Intel Capital’s check is a venture stake. It doesn’t come with Intel’s fabs.
Washington’s part is more concrete. The July 29 letter of intent is one slice of $874 million across seven companies, and each recipient hands the government a minority, non-controlling equity stake. The award still has to clear diligence. It’s the same grant-for-equity approach that left the government holding 10% of Intel, and it tells you Washington wants a second American path to memory supply. It doesn’t make that path arrive sooner.
For Micron holders the dip changes nothing in the numbers due on September 30. The stock trades at about 23 times trailing earnings, sits below its $1,255 52-week high, and carries an average street target near $1,515. The fiscal fourth quarter will be judged on HBM4 pricing and how far out the long-term customer agreements now run. A company that ships its first samples after the print can’t touch either. What Thursday did show is how thin the market’s confidence in shortage duration has become: a stealth reveal with 2,000 wafers behind it moved a trillion-dollar stock.
Watch for the first customer to qualify a Kepler HBM sample. Until one does, Micron’s shortage is intact.