How Former Google & Meta Execs are Tackling AI Memory Limits

Three former executives from Google and Meta have joined hands to tackle the most stubborn bottleneck in AI development: the memory wall.

Ofer Shacham, Sha Rabii and Masumi Reynders, industry leaders and co-founders of the AI infrastructure startup Majestic labs, have raised US$100m, to pioneer technology designed to reduce the cost of building data centres by overcoming the memory problem.

This comes as welcome respite for tech giants increasing their investment in data centre infrastructure to fuel the AI race.

Majestic Labs’ mission to democratise access to advanced AI capabilities while reducing environmental impact is already evident in its innovative architecture.

“Majestic is built on a simple and powerful insight: AI’s next leap forward will come from access to more powerful AI infrastructure and more powerful AI infrastructure requires a reimagination of the memory system,” says Co-founder and CEO, Ofer Shacham.

“Majestic servers will have all the compute of state-of-the-art GPU/TPU-based systems coupled with 1000x the memory. 

“Our breakthrough technology packs the memory capacity and bandwidth of 10 racks of today’s most advanced servers into a single server, providing our customers with unprecedented gains in performance and efficiency while slashing power consumption.”

Customers will be able to access prototypes of the box servers by 2027.

What is the memory wall in AI infrastructure?

AI systems require reading and writing massive quantities of data between processors and memory blocks.

Even as companies like Nvidia release increasingly powerful GPUs, improvements in memory technology have not kept pace.

As processor speed improves faster than memory bandwidth, processors are forced to wait for data to arrive, causing performance stagnation.

This becomes a major bottleneck in performance-intensive AI workloads.

Majestic’s patent-pending technology enables the memory of 10 or more racks to be packed into a single server, unlocking up to 50-fold performance gains.

Majestic Labs sets out to change the face of AI infrastructure

“AI infrastructure is scaling at unprecedented speed, but the industry has not solved key fundamental architectural inefficiencies,” says Co-founder and COO Masumi Reynders. 

“Majestic addresses this by delivering immediate operational gains on today’s workloads while maintaining full programmability and flexibility to adapt as AI evolves beyond transformer-based models.”

Lowering the AI infrastructure cost will make the technology accessible to more people, particularly in the developing countries.

“Majestic allows for a level of scalability and operational efficiency that simply isn’t possible with traditional GPU based systems,” says Co-founder and President Sha Rabii. 

“Our systems support vastly more users per server and shorten training time, lifting AI workloads to new heights both on-premises and in the cloud. 

“Our customers benefit from tremendous improvements in performance, power consumption and total cost of ownership.”

By removing the memory wall, the company is not just designing faster machines, but redefining what is computationally possible, opening the door to larger models and broader global access.