Analysis · Published 21 September 2026
One way to make an AI device more useful is to give it more room to work. That room is partly inside the chip: memory that holds the data a processor needs. Finding space for it can be as consequential as improving the processor itself.
On 2 September, Israel’s RAAAM Memory Technologies and GlobalFoundries announced a collaboration on GCRAM memory, with a test-chip tape-out completed on GF’s FDX platform. Design access for initial customers is planned for early 2027. Joint announcement.
Where Abu Dhabi enters the picture
The connection is in the ownership chain. Mubadala, based in Abu Dhabi, reported beneficial ownership of 72.82% of GlobalFoundries as of 30 June 2026, in a filing dated 10 August. RAAAM’s development partner therefore has a substantial Emirati shareholder. Ownership filing.
That establishes a specific commercial link worth following. It does not establish a direct Mubadala investment in RAAAM, a factory in the Emirates or a Gulf customer order. The distinction matters: readers evaluating a business opportunity need to know who is developing technology, who owns the partner and who has actually committed to buying something.
For SalamTechs, this is the interesting regional story. An Israeli company can participate in an industrial network connected to Gulf capital through a global manufacturing partner. The useful next question is what that relationship enables technically, and which milestones remain before customers can use it.
A smaller workspace can change the design
RAAAM develops memory technology that chip designers can incorporate into their own products. Its approach aims to fit more memory within a given chip area, or preserve capacity while using less space. The company also targets lower power consumption. These are supplier claims, rather than results independently tested by SalamTechs. RAAAM’s technology explanation.
Consider a hypothetical camera inspecting packages on a conveyor. Its designer has to accommodate the image-processing logic and the data that logic uses. If the same memory capacity occupies less space, the designer gains a choice: attempt a smaller chip, or allocate the freed area to another function. Neither outcome happens automatically; the whole design must still work.
The distinction between a component and a complete device is particularly important when reading efficiency figures. A memory improvement does not translate one-for-one into a smaller electricity bill or longer battery life. The sensor, processor, communications and software still contribute to the result.
The partners report 40% less memory area and up to 60% lower memory power than SRAM for this development. Those figures concern memory, not the complete chip or device. Development details.
The milestone after the headline
A tape-out means a chip design has reached the handoff for fabrication. It is an engineering milestone; it does not by itself demonstrate production yield, sustained customer demand or performance in every operating condition. A buyer still needs evidence from the version that will go into its product.
Our practical reading is to watch three things: measured results from manufactured silicon, the conditions behind those measurements, and customer integration. A useful evaluation would compare equal memory capacity and comparable operating requirements, then measure the effect on the finished system. Ask also what design work is needed to adopt the memory and how failures would be investigated.
For Israeli suppliers and Gulf technology buyers, this is a more useful conversation than treating a partnership announcement as the finished achievement. The ownership connection is documented. Turning the engineering promise into a product customers can rely on is the work that follows.