Majestic Labs представила сервер Prometheus для замены стойки GPU Nvidia

Israeli startup Majestic Labs unveiled the Prometheus server in early August 2026, claiming it can do the work of an entire Nvidia GPU rack. The company's key thesis: the main bottleneck in AI infrastructure is not a shortage of computing power, but memory limitations.

Why Memory Matters More Than GPU?

Majestic Labs is attacking the "memory wall" — a phenomenon well known to computing system architects: processors sit idle waiting for data to load from memory instead of performing useful computations. The company's founders — alumni from Google and Meta who founded Majestic Labs in 2023 — are convinced that this barrier limits the real performance of AI systems more than the number of computing cores.

Key facts about the company and product:

• Majestic Labs was founded in 2023, headquartered in Tel Aviv
• The team consists of former Google and Meta engineers
• The Prometheus server is positioned as a replacement for an entire Nvidia GPU rack
• Architectural bet: priority on memory speed, not computation

What's Wrong with Nvidia GPUs?

Nvidia accelerators — H100 and B200 — dominate AI infrastructure, but when working with large language models they face a fundamental limitation. High-speed HBM memory on each chip is limited in capacity: when running inference on a model with tens of billions of parameters, compute cores often sit idle waiting for data. The typical industry response — clustering GPUs and scaling interconnect bandwidth — Majestic Labs considers a half-measure, not a solution to the problem.

Prometheus rethinks the system's balance: the server architecture is built around memory, not GPU. According to the company, this allows a single device to do the work that usually requires an entire rack of accelerators.

"The conversation about AI hardware has been stuck on one word — compute.

We want to switch it to another word: memory," — the company stated its position; according to The Next Web, this thesis is at the core of the Prometheus architecture.

How Serious Is the Challenge to Nvidia?

The AI accelerator market already has companies offering alternatives to traditional GPUs — Groq, Cerebras, SambaNova. Majestic Labs occupies a special place in this field: the bet is not on matrix operation speed, but on eliminating the memory wall as a systemic constraint. This approach is particularly relevant for inference tasks, where large models require fast weight loading — and this is precisely where GPUs most often lose in the performance-to-cost ratio.

What This Means

If Prometheus withstands independent testing, Majestic Labs could change the logic of AI infrastructure procurement: engineers from Google and Meta may mount a real challenge to Nvidia's dominance in the AI server segment. The main question now is not "is an alternative possible," but "when will it be cheaper and faster than GPU clusters."

*Meta is recognized as an extremist organization and is banned in Russia.