Updates with further details, context in paragraphs 2-5

May 8 (Reuters) - SoftBank Corp 9434.T has begun discussions with U.S. chip giant Nvidia <NVDA.O> and Taiwanese contract manufacturer Foxconn 2317.TW as it weighs plans to build "made-in-Japan" artificial intelligence servers, the Nikkei newspaper reported on Friday.

The Japanese telecom company wants to build a production system by initially assembling externally sourced components by the end of the decade, before eventually taking charge of the entire server manufacturing process, Nikkei added.

The unit will focus on high-performance servers capable of running advanced graphics processing units at high speeds, the report also said.

The project will be part of SoftBank's medium-term management plan, which, according to Nikkei, could be announced as soon as Monday.

The development comes after Masayoshi Son-led SoftBank Group poured in more than $30 billion in investments in OpenAI so far, for about an 11% stake, in an "all-in" bet that it will emerge as a winner in the battle among developers of large language models.