A major leak just gave the tech world its first look at Intel’s upcoming artificial intelligence hardware. Images of the “Crescent Island” PCIe accelerator board hit the internet this week, courtesy of a well-known hardware leaker. These photos confirm that Intel is preparing a powerful new GPU designed specifically for AI inference tasks. The images show a massive graphics processor sitting at the center of a complex circuit board, hinting that Intel is ready to challenge its rivals in the data center market.
The most striking feature of this new card is the sheer size of the GPU itself. It looks significantly larger than Intel’s current flagship model, the Xe2-based BMG-G31. According to the leak, Intel built this new chip on the “Xe3P” architecture. This is a brand-new design that Intel only recently teased during deep-dive presentations for its Panther Lake and Xe3 mobile processors. By using the scalable Xe3P architecture, Intel aims to offer a design that works everywhere, from small laptop graphics to giant data center AI clusters.
The leaked board design reveals a clever strategy for managing memory costs. Instead of using expensive, high-bandwidth memory (HBM) modules—which currently suffer from severe supply shortages and high price tags—Intel decided to use LPDDR5X memory. The board features 12 memory modules on the front and 8 on the back, for a total of 20 modules. Each module provides 8 GB of space, giving the card a massive total capacity of 160 GB. This choice makes the card a cost-effective powerhouse compared to competitor solutions that rely on pricier alternatives.
Intel built this high-end board with a robust power system. The photos show 13 populated voltage regulator modules (VRMs), though the board design actually has room for up to 18. Power flows into the card through a single 16-pin connector located on the back, which is standard for high-performance PCIe cards today. The leak also revealed a side USB Type-C port, which engineers likely use for testing and debugging during the development phase.
The Crescent Island accelerator is purpose-built for AI “inference” workloads. In simple terms, inference is the process where a trained AI model takes information and generates a response, such as a chatbot answering a question or a system analyzing sensor data in real time. Because inference is the most common task in the AI industry, Intel focused on making this card “power-and-cost-optimized.” It is specifically designed to work in air-cooled enterprise servers, meaning companies won’t need to install expensive liquid-cooling systems just to keep the card from overheating.
Competitors like Nvidia and AMD currently dominate the AI space, and they rely heavily on top-tier HBM memory to reach peak performance. While these companies are already talking about next-generation memory standards like HBM4, they struggle to source enough inventory to meet global demand. By choosing LPDDR5X, Intel is making a strategic bet. If they can provide a card with 160 GB of memory at a much lower price than Nvidia, they could win over “tokens-as-a-service” providers who need to run thousands of AI queries every second without spending a fortune.
The software side of this launch is already well underway. Intel is currently testing its open and unified software stack, which helps different types of processors work together in a single system. The company already uses this software with its existing Arc Pro B-series lineup. By applying these optimizations early, Intel ensures that when Crescent Island finally launches, it will be ready to perform from day one.
Intel currently plans to start sending samples to customers in the second half of 2026. This gives the company several more months to refine the manufacturing process and improve the final design. As the world’s appetite for artificial intelligence continues to grow, companies are spending over $1 billion every month just on new server hardware. Intel hopes that by offering a cheaper, power-efficient alternative to the industry standard, it can grab a significant slice of this massive market and prove that it still belongs at the top of the graphics card game.
