Intel Formally Launches GPU D270 With 480GB of Memory for AI Inference
Intel has formally launched its GPU D270, an AI inference-focused graphics card offering up to 480GB of LPDDR5X memory, as the company works to establish a stronger position in the competitive AI accelerator market dominated by Nvidia and AMD.
Built for Inference, Not Training
Like Microsoft’s recently unveiled Maia 300, the D270 is positioned primarily for inference workloads, running already-trained AI models to serve user queries, rather than the computationally heavier task of training new models from scratch. This focus reflects a broader industry trend of chipmakers designing specialized hardware for inference, as the cost of running AI models at scale becomes a larger share of overall AI infrastructure spending.
The Memory Specification
The D270’s standout specification is its memory capacity, offering up to 480GB of LPDDR5X, a type of memory more commonly associated with mobile and laptop devices but adapted here for large-scale AI workloads. High memory capacity allows a single card to hold larger AI models or serve more simultaneous requests without needing to split workloads across multiple chips, which can improve both cost efficiency and response speed for AI service providers.
Intel’s Position in the AI Chip Market
Intel has struggled in recent years to match Nvidia’s dominance in AI accelerators, while also facing growing competition from AMD and custom chips developed by hyperscalers like Microsoft, Google, and Amazon. The D270 launch represents a continued effort by Intel to carve out a role in AI infrastructure, with memory capacity and inference efficiency positioned as differentiators against more training-focused competitor products.
Availability Timeline
Intel has indicated that broader availability of the D270 is targeted for 2027, suggesting the current launch marks an initial rollout stage rather than immediate mass-market supply. The announcement adds to a wave of new AI chip launches in recent days, including Nvidia’s open-sourced Vera Rubin rack designs and AMD’s new Helion rack, as chipmakers race to offer differentiated hardware for the next phase of AI infrastructure buildout.
