The preview edition of the Lenovo AI Host P7 has made its debut. This host is outfitted with dual domestic chips, embodying a compact and lightweight design that supports continuous, uninterrupted operation for 7×24 hours. It delivers outstanding performance, enabling the local execution of large models with parameters reaching up to 122 billion, and is capable of processing ultra-long contexts of 256K. Even when offline, its inference speed remains impressive at 50 Tokens per second. Under full load, the entire system consumes less than 30W of power and operates quietly. The P7 is pre-loaded with a diverse array of local model libraries and comes integrated with Lenovo Tianxi AI, ensuring it is ready for immediate use. It also empowers users to configure their own Agents. In Auto mode, it can automatically select an optimal solution between local and cloud models, effectively tackling challenges such as the high costs associated with cloud AI invocation, privacy leakage risks, and unstable network conditions. Lenovo is poised to officially release the P7 and initiate an early access program, inviting the first batch of users to engage in product co-creation to refine and enhance the user experience.