A new memory chip company backed by SoftBank just signed a deal with Intel to build next-generation computer memory.
Saimemory, which SoftBank only set up two months ago in December 2024, announced the partnership with Intel on Tuesday. They’re calling it the Z-Angle Memory program, or ZAM.
Investors liked the news. SoftBank shares went up 3.13%, while Intel’s stock jumped 5% in after-hours trading on Robinhood.
Memory chips that run AI systems and high-speed computers simply can’t keep up with demand. And it’s not getting better anytime soon. Industry experts figure this shortage will last at least through 2026 or 2027.
“Standard memory architectures aren’t meeting AI needs,” said Dr. Joshua Fryman, an Intel Fellow who runs technology development at Intel Government Technologies. Intel’s built a new way to make memory chips that are faster and use less power while costing less to produce. They think this approach will take off over the next decade.
Major makers sold out through 2026
It’s bad out there. Major memory makers Micron and SK Hynix? They’ve already sold out everything they can make through 2026. Data centers are expected to gobble up 70% of all memory chips made worldwide in 2026.
The tight supply has sent prices through the roof. DRAM memory used in computers and servers saw prices spike 172% throughout 2025. Some quarters hit price jumps between 50% and 55%, according to TrendForce.
Other parts of tech are getting squeezed too. IDC warned that PC sales could drop anywhere from 4.9% to 8.9% in 2026 because manufacturers can’t get enough chips. Smartphone sales might also slip by 2.1%.
Government-funded memory tech could solve global chip shortage
Samsung and SK Hynix bumped up prices on their advanced HBM3E memory by nearly 20% for 2026 deliveries. Both are also pouring money into boosting production. Samsung’s planning to expand HBM production by 50% in 2026. SK Hynix is spending four times more on new facilities and equipment.
SK Hynix controls over half the market for high-bandwidth memory right now. They’re expected to grab 70% of the next generation market for HBM4, which Nvidia’s upcoming Rubin platform will use. At CES in January 2026, SK Hynix showed off the world’s first 16-layer HBM4 module—48GB of data that moves at 11.7 gigabits per second. Both Samsung and SK Hynix moved up their schedules to start making HBM4 in February 2026.
Meanwhile, Intel’s bringing in technology from work they did with the U.S. Department of Energy’s Advanced Memory Technology program. That program was all about building better memory chips. Intel specifically focused on making DRAM run better while using less power.
The partnership also tackles another big concern: how much energy AI computing consumes.
As AI systems get bigger and more powerful, they’re eating up enormous amounts of electricity. The ZAM program has energy savings baked in from the start.
