Reno-based AI hardware startup Positron AI closed an $875 million Series C financing round, arguing commodity LPDDR5X memory can match Nvidia’s high‑bandwidth memory for inference workloads.
Reno‑based AI hardware startup Positron AI has closed an $875 million Series C financing round, betting that commodity LPDDR5X memory can rival Nvidia’s high‑bandwidth memory (HBM) in inference workloads.
Funding round details
The round was led by a consortium of venture firms, with participation from existing investors. The capital will be used to scale Positron’s proprietary memory‑controller technology and to accelerate product development for data‑center AI inference servers.
Technical premise
Positron AI argues that the cost‑effective LPDDR5X memory, when paired with its custom controller, can deliver comparable bandwidth to HBM while reducing overall system cost and power consumption. The company claims its solution can achieve up to 1.5 TB/s of memory bandwidth, a figure traditionally associated with HBM‑2e stacks.
The startup’s approach focuses on optimizing data pathways and reducing latency through software‑defined memory scheduling, allowing inference models to run efficiently without the need for expensive HBM modules.
Market implications
If Positron’s claims hold up, AI hardware manufacturers could see a shift toward more affordable memory configurations, potentially lowering the price of inference‑focused servers and expanding access for smaller enterprises.
- Lower total cost of ownership for AI inference clusters
- Reduced power draw compared with HBM‑based designs
- Simplified system integration using standard LPDDR5X modules
We believe commodity memory, when engineered correctly, can democratize high‑performance AI inference.
The company plans to begin sampling its first inference accelerator boards later this year, targeting customers in cloud services, edge computing, and enterprise AI deployments.