Ex-Google and Meta execs raise $100M for AI chip startup Majestic Labs to solve AI’s memory bottleneck

Three former Meta and Google chip executives are taking on one of the biggest headaches in AI infrastructure: the skyrocketing cost of building and running data centers. Their new AI startup, Majestic Labs, has raised $100 million to rethink how memory is handled inside servers — and it’s promising hardware that could shrink entire data centers.

Founded by Ofer Shacham, Sha Rabii, and Masumi Reynders, Majestic Labs has been quietly operating since late 2023. The trio worked together, leading silicon design teams at both Meta and Google, and now they’re betting that their approach to chip architecture can help hyperscalers save billions. The company’s patent-pending system packs roughly 1,000 times the memory of a typical enterprise-grade server, allowing one Majestic server to replace as many as ten racks of today’s standard setups, the founders told CNBC.

Majestic Labs aims to pack the memory of ten racks of servers into a single system, tackling the “memory wall” bottleneck in AI workloads. By breaking through this long-standing limitation between compute and memory, Majestic hopes to make large-scale AI training and inference faster, cheaper, and far more efficient.

The company recently closed a $71 million Series A led by Bow Wave Capital, with participation from Lux Capital and others. Including earlier funding, Majestic has now secured a total of $100 million.

Majestic’s pitch is simple but bold: the way AI data centers are built today doesn’t scale. Every large language model, from OpenAI’s GPT to Anthropic’s Claude, depends on high-performance GPUs. These chips, mostly made by Nvidia, have become the industry’s backbone — and its bottleneck. Majestic’s founders say they aren’t trying to replace GPUs but to solve the other half of the problem: memory.

“Nvidia makes excellent GPUs and has driven incredible AI innovation,” Shacham said. “We’re not trying to replace GPUs across the board — we’re solving for memory-intensive AI workloads where the fixed compute-to-memory ratio becomes a constraint.”

Majestic’s technology aims to break that constraint. Its servers combine compute and massive memory capacity in a single unit, collapsing multiple racks’ worth of equipment into one compact box. That means less space, less cooling, and less energy — and a lot less money spent by cloud giants.

Prototypes are expected to reach select customers in 2027, and pre-order discussions are already underway, according to the founders. The company, which has fewer than 50 employees split between Los Altos, California, and Tel Aviv, Israel, plans to scale its teams and raise additional funding in the coming year.

For the three co-founders, Majestic is the product of years of shared experience and a clear view of AI’s infrastructure limits. They met at Google, where they helped develop the company’s custom chips known as TPUs. Reynders, who joined Google in 2003 as a senior corporate counsel, later became director of product management and silicon. Rabii, who sold his earlier chip company Arda Technologies to Google in 2011, went on to lead engineering for Google’s Argos video chip, now used in YouTube’s data centers. Shacham arrived after Google acquired his startup, Chip Genesis, in 2013. His designs later powered parts of Google’s Pixel smartphones.

In 2018, the three moved to Meta, where they founded the Facebook Agile Silicon Team (FAST) inside Reality Labs. The group developed Meta’s in-house silicon before layoffs in 2023 forced Shacham to shut down parts of the operation. “Part of that was layoffs across the organization, and FAST was not excluded from that,” he said. “It’s not a good place to be, not a good feeling to do.”

That experience pushed the group to start fresh. They began brainstorming the biggest pain points in AI hardware — and kept coming back to memory. “We’ve been friends and colleagues for a long time, so this notion of working together and doing something exciting has always been in the periphery,” Reynders said.

As they prepare for growth in 2026, the team is leaning on an extended network of more than 1,500 former colleagues from Meta and Google — a deep talent pool familiar with their work and leadership. “There’s that trust they already have with us,” Rabii said.

Majestic’s timing may prove right. Cloud giants are pouring unprecedented sums into infrastructure — Alphabet, Meta, Microsoft, and Amazon are expected to spend a combined $380 billion this year on data centers alone. If Majestic can deliver on its promise to make AI computing lighter, cheaper, and less energy-hungry, those conversations about pre-orders may turn into some of the most consequential deals in the next wave of AI hardware.