This paper reviews and outlines the future of Urban Foundation Models (UFMs), which are large-scale models pre-trained on extensive, diverse urban data. UFMs aim to understand, interpret, and manage complex urban systems and environments, enabling autonomous performance of intellectual tasks related to urban contexts. Machine learning technologies have become pivotal in transforming urban landscapes, underpinning the development of various smart city services. These technologies enhance urban intelligence by enabling efficient resource allocation, improved public services, and elevated quality of life for city dwellers. The recent emergence of foundation models, such as Large Language Models (LLMs) and Vision Foundation Models (VFM), has significantly reshaped the research landscape in machine learning and artificial intelligence. These models are characterized by their extensive pre-training on large-scale datasets, which imbues them with unparalleled emergent abilities, including contextual reasoning, complex problem solving, and zero-shot adaptability across diverse tasks. UFMs, pre-trained on multi-source, multi-granularity, and multimodal urban data, exhibit a deep understanding of various urban data types and remarkable adaptability to a wide array of urban tasks. They can offer comprehensive insights, uncover intricate spatiotemporal patterns, and enhance decision-making across various urban tasks. Despite the burgeoning interest and potential of UFMs, the field faces several challenges, including the absence of clear definitions, the lack of systematic reviews of existing literature, and the need for universally applicable solutions. This paper addresses these gaps by presenting a comprehensive survey on UFMs. It defines UFMs, discusses their unique challenges, proposes a data-centric taxonomy to categorize existing research, and introduces a new framework for constructing UFMs. The paper also explores the application landscape of UFMs, detailing their potential impact in various urban contexts. The major contributions of this paper include a comprehensive and systematic review of Urban Foundation Models, introducing the concept of UFMs, proposing a data-centric taxonomy, and presenting a novel framework for the development of future UFMs. The paper is organized into sections discussing the basics of UFMs, existing studies on UFMs, a prospective framework for building a general UFM, promising applications of UFMs in different urban domains, and a conclusion.This paper reviews and outlines the future of Urban Foundation Models (UFMs), which are large-scale models pre-trained on extensive, diverse urban data. UFMs aim to understand, interpret, and manage complex urban systems and environments, enabling autonomous performance of intellectual tasks related to urban contexts. Machine learning technologies have become pivotal in transforming urban landscapes, underpinning the development of various smart city services. These technologies enhance urban intelligence by enabling efficient resource allocation, improved public services, and elevated quality of life for city dwellers. The recent emergence of foundation models, such as Large Language Models (LLMs) and Vision Foundation Models (VFM), has significantly reshaped the research landscape in machine learning and artificial intelligence. These models are characterized by their extensive pre-training on large-scale datasets, which imbues them with unparalleled emergent abilities, including contextual reasoning, complex problem solving, and zero-shot adaptability across diverse tasks. UFMs, pre-trained on multi-source, multi-granularity, and multimodal urban data, exhibit a deep understanding of various urban data types and remarkable adaptability to a wide array of urban tasks. They can offer comprehensive insights, uncover intricate spatiotemporal patterns, and enhance decision-making across various urban tasks. Despite the burgeoning interest and potential of UFMs, the field faces several challenges, including the absence of clear definitions, the lack of systematic reviews of existing literature, and the need for universally applicable solutions. This paper addresses these gaps by presenting a comprehensive survey on UFMs. It defines UFMs, discusses their unique challenges, proposes a data-centric taxonomy to categorize existing research, and introduces a new framework for constructing UFMs. The paper also explores the application landscape of UFMs, detailing their potential impact in various urban contexts. The major contributions of this paper include a comprehensive and systematic review of Urban Foundation Models, introducing the concept of UFMs, proposing a data-centric taxonomy, and presenting a novel framework for the development of future UFMs. The paper is organized into sections discussing the basics of UFMs, existing studies on UFMs, a prospective framework for building a general UFM, promising applications of UFMs in different urban domains, and a conclusion.