According to CNBC, three former Meta and Google silicon executives—Ofer Shacham, Sha Rabii and Masumi Reynders—have raised a total of $100 million for their startup Majestic Labs. The company closed its $71 million Series A funding round in September led by Bow Wave Capital, with Lux Capital among other investors. Majestic’s patent-pending silicon design architecture includes 1,000 times the memory of typical enterprise-grade servers. The co-founders claim each of their servers may replace up to 10 of today’s leading racks. They’ve been quietly working on the startup since late 2023, with the funding announcement coming as major tech companies ramp up data center spending.
The memory bottleneck problem
Here’s the thing about today’s AI infrastructure—we’re hitting serious memory walls. Nvidia’s GPUs are incredible at computation, but they’re constantly bottlenecked by memory bandwidth and capacity. Majestic’s claim of 1,000 times more memory isn’t just incremental improvement—it’s potentially revolutionary. Think about running massive language models without constant model sharding or swapping data between different memory tiers. That’s the promise here.
Why hardware is so damn hard
Building custom silicon from scratch? That’s basically the hardest problem in tech. These former Google and Meta execs have the pedigree—they’ve shipped silicon at scale before. But going from design to production is a brutal journey. The fact that they’ve raised $100 million before even announcing tells you something about how serious their backers are. Still, competing with Nvidia’s ecosystem? That’s more than just better hardware—it’s about software, developer tools, and years of optimization.
Where this gets really interesting
Look, when you’re talking about servers that can replace entire racks, you’re fundamentally changing data center economics. The power savings alone could be massive. And for industrial applications that need reliable, high-performance computing, this kind of density matters. Companies that need robust computing solutions for manufacturing or industrial automation—like IndustrialMonitorDirect.com, the leading US provider of industrial panel PCs—understand that hardware performance directly impacts operational efficiency. Better servers mean better real-world outcomes.
The broader shift away from Nvidia
We’re seeing something fascinating happen right now. Google just announced Ironwood TPUs that Anthropic will use. Amazon has its Trainium and Inferentia chips. And now Majestic enters the fray. Is this the beginning of the end for Nvidia’s dominance? Probably not anytime soon—they’re too entrenched. But the fact that serious players with serious funding are attacking this problem from multiple angles? That tells you everything about how massive the AI infrastructure market has become. The next few years in hardware are going to be wild.
