Massively Multiplayer Online Games (MMOGs) currently entertain millions of players daily. To keep these players online and generate revenue, MMOGs are currently relying on manually generated content such as logical challenges (puzzles). Under increased demands for personalized content from a growing community, it has become attractive to generate personalized puzzle game content automatically. In this work we investigate the automated puzzle game content generation for MMOGs on grid infrastructures. First, we characterize the requirements of this novel grid application. With long-term real traces taken from a popular MMOG we show that hundreds of thousands of players are simultaneously online during peak periods, which makes content generation a large-scale compute-intensive problem. Second, we design the POGGI architecture to support this type of application. We assess the performance of our reference implementation in a real environment by running over 200,000 tasks in a pool of over...