What they make:
STT-MRAM, a memory chip that’s fast like DRAM, persistent like Flash, and nearly indestructible. No data loss on power failure. Unlimited write cycles. Works at -40°C to +125°C. Radiation resistant.
Why robots need this:
Every humanoid robot is a mobile AI computer operating in harsh environments. It can’t boot slowly. It can’t lose its state when it falls or loses power. It needs to update its AI model daily without wearing out its memory.
DRAM loses everything on power loss. NAND Flash wears out after 100k writes. MRAM does neither.
The numbers:
\- 3 million humanoid robots projected by 2030
\- \~6 MRAM chips per robot at \~$30 each
\- Everspin at just 15% market share = $81M/year, more than their entire revenue today
The chiplet angle: Everspin just joined PACE (Physical AI Chiplet Ecosystem). If Nvidia or Qualcomm builds a humanoid robot SoC, they could license Everspin’s MRAM IP directly. Every robot sold pays Everspin a royalty. ARM-style. Zero extra CapEx.
Moat:
20+ years of STT-MRAM process knowledge. 100+ patents. Only company shipping qualified high-reliability MRAM at volume today. Samsung is 15 years behind on this specific technology.
Current valuation:
MCap $213M, $44.5M cash, EV \~$168M. KGV \~17x ex-cash. You’re paying almost nothing for the robot optionality.
Risks:
Robots scale slower than expected. Samsung gets serious. Density limitations. Single analyst coverage (Needham, Buy, $14 PT).
Already profitable. Debt free. 238 design wins in 2025 ramping to revenue in 2026/2027. DoD contract. LEO satellites in production.
The robot story is early, but at $9.30 you’re buying a real business at a fair price with a free lottery ticket on the humanoid revolution.
Not financial advice. Do your own research.