Liqid Unveils Massive Scale-Up AI Platform Powered by AMD Instinct™ MI350P GPUs
Liqid and AMD unveiled a GPU pooling platform combining 30 MI350P GPUs per server, achieving 3.7x faster inference and 2.1x token economics. The 65% lower deployment cost and 50% efficiency gains enable on-premises agentic assistants and RAG at enterprise scale without datacenter redesign.
Targets both frontier AI and supercomputing
AMD is making a serious play for Nvidia's crown with its new Helios rack-scale AI platform, Venice CPUs, and MI455X chips.
AMD reveals MI455X (CDNA 5, Helios 72-GPU): MXFP4 peak 4x Nvidia Rubin, AMD's first truly competitive rack-scale accelerator. Microsoft/Anthropic deals confirm momentum, but real-world performance requires substantial software optimization—the gap remains.
AMD launches MI455X and MI430X GPUs with 432GB HBM4 (2.9x bandwidth increase) and up to 40.26 petaflops—CDNA 5 at 2nm. Open-stack alternative to NVIDIA's H-series reshapes capex efficiency for LLM training infrastructure and sovereign AI deployments.
AMD launched MI455X accelerators and Helios racks backed by OpenAI and Cerebras; $2 trillion AI market by 2030. Enterprise GPU capex now has credible Nvidia competitor, yet execution timing creates risk—Vera Rubin ships today while Helios reaches volume in 2027.
AMD unveils Helios platform: 72-GPU rack with MI455X (432GB HBM4, 2.9 exaflops) and Venice CPUs for reasoning models and agentic workflows. Signals viable NVIDIA challenger for hyperscalers; open UALink networking vs Vera Rubin's proprietary stack.
AMD launched the MI455X accelerator, Helios server rack, and Venice Epyc processor at its Advancing AI event, predicting a $2 trillion market by 2030
Massive performance gains, higher memory capacity, and a large scale-up domain make for AMD’s strongest AI chip yet