Rocket One has signed a development partnership with Virginia Commonwealth University to build SKYMRAM, a next-generation magnetic memory aimed at AI accelerators and defense chips. The project pairs a commercial chip developer with a public research university, and its target is specific: nonvolatile memory that can sit close to logic without the standby losses of conventional alternatives.

The name tells you the ambition. MRAM, or magnetoresistive random-access memory, stores bits as magnetic states rather than electric charge. That makes the cells nonvolatile — they hold data when power drops — and, in principle, tolerant of the radiation environments that defense systems must survive.

What is SKYMRAM meant to do?

The partnership positions SKYMRAM as memory for two demanding customers at once. AI chips need fast, dense memory near their compute elements, and the data-shuttling between logic and DRAM remains a familiar bottleneck. Defense chips need storage that endures power loss, interference, and harsh conditions.

Magnetic memory addresses both problems in one technology. It does not need constant refresh the way DRAM does, and it does not lose its contents when the supply is cut, the way SRAM and DRAM both do. For edge AI devices and mission hardware, that combination is the entire sales pitch.

Why a university partnership?

Rocket One is not going alone. VCU brings research capacity in materials and device physics, areas where magnetic memory still has open questions — cell stability, write energy, and integration with standard CMOS process flows.

The startup-university structure is a common route for memory technologies that have not yet reached high-volume manufacturing. MRAM in its embedded form has shipped from foundries for several years, but standalone and high-performance variants remain development programs rather than products. A university partner supplies the patient research; the company supplies the path to commercialization.

Where does magnetic memory stand today?

Embedded MRAM is real and shipping; the broader MRAM family remains a challenger. The incumbents are entrenched. SRAM owns on-chip caches, DRAM owns main memory, and NAND owns storage. Each defends a mature manufacturing base measured in decades.

MRAM's case has always been the middle ground: denser than SRAM, faster than NAND, nonvolatile unlike both DRAM and SRAM. Its critics answer that it has been the technology of the future for a long time. SKYMRAM is, in effect, another attempt to close that gap for two niches where the incumbents are weakest — power-constrained AI inference and radiation-hardened defense electronics.

Who benefits if it works?

The immediate beneficiaries would be designers of edge AI hardware, where every milliwatt of standby power matters, and defense system builders, who must qualify components for conditions that commercial parts do not face.

For Rocket One, a working SKYMRAM would be a differentiated product in a memory market with few new entrants. For VCU, the partnership is a route from laboratory research toward a named commercial technology. For the wider industry, it is another data point on whether magnetic memory can finally move from promising to production.

What happens next?

The partnership is at the development stage. There are no announced products, no disclosed capacity plans, and no published timeline for commercialization — the normal condition of a memory project at this phase, and a reminder that the distance from lab cell to qualified chip is measured in years.

The open questions are the standard ones for MRAM: write speed, endurance, density, and cost at scale. Rocket One and VCU will have to answer all four before SKYMRAM appears in a shipping AI accelerator or a defense platform. Until then, the partnership signals intent rather than delivery — a common starting position in memory, and rarely an easy one to leave.