Patriot Memory has announced the AI Matrix Series, a new lineup of DDR5 memory modules and PCIe M.2 2280 NVMe SSDs designed for systems running AI workloads locally. The new hardware lineup is divided into PRO, Mid, and Entry categories, with configurations targeting large language models (LLMs) ranging from 7 billion to 70 billion parameters.
The company is targeting AI developers, content creators, system builders, and small businesses that want to run AI models without depending on cloud-based services. Local AI processing can offer better data privacy, offline access, and reduced dependence on recurring API costs. However, running larger models requires sufficient memory capacity and storage bandwidth to handle model weights, context processing, and large datasets.
The Patriot AI Matrix PRO tier targets 34B to 70B parameter models and combines the MS560 PCIe Gen 5 x4 SSD with MD57N DDR5 memory. The SSD is available in 1TB, 2TB, and 4TB capacities, with the 4TB version delivering sequential read and write speeds of up to 14,000 MB/s and 11,000 MB/s, respectively. The accompanying memory configuration offers 64GB (2x32GB) capacity and speeds of up to 6000 MT/s (OC).

For users working with 14B to 34B models, the Mid tier features the MS450 PCIe Gen 4 x4 SSD and MD55N DDR5 memory. The SSD offers capacities of up to 4TB, with sequential read and write speeds reaching 5,000 MB/s and 4,500 MB/s. The memory kit provides 32GB (2x16GB) capacity and supports speeds of up to 6000 MT/s (OC).
The Entry tier is designed for smaller 7B to 14B models used for coding assistance, translation, and copywriting. It includes the MS350 PCIe Gen 3 x4 SSD, available in 512GB, 1TB, and 2TB capacities, with sequential read and write speeds of up to 3,000 MB/s and 2,000 MB/s. Patriot is pairing it with MD35N and MD34N DDR5 memory operating at 5600 MT/s in a 32GB (2x16GB) configuration.
Patriot is also planning monthly benchmark testing across open-source AI models, including Llama, Gemma, and CodeLlama. The testing will measure prompt-processing performance (pp512) and token-generation performance (tg128), alongside extended stress tests to evaluate thermal and operational stability during continuous AI workloads.
The company says the benchmark results will be updated as AI models evolve, providing system builders and integrators with performance data for selecting suitable hardware. Patriot has not yet announced pricing or availability details for the AI Matrix series.
