75% cheaper compute costs and a launch 3 months ahead of schedule. It seems almost impossible for an AI startup. Until you look beyond hyperscalers. @wonderaofficial needed more than half a million GPU hours to train AI models for music generation, voice conversion, and audio separation. Using
news·2026年6月1日·2件のソース
記事を読むすべてのソース
75% cheaper compute costs and a launch 3 months ahead of schedule. It seems almost impossible for an AI startup. Until you look beyond hyperscalers. @wonderaofficial needed more than half a million GPU hours to train AI models for music generation, voice conversion, and audio separation. Using
@ionet
2026年6月1日
75% cheaper compute costs and a launch 3 months ahead of schedule. It seems almost impossible for an AI startup. Until you look beyond hyperscalers. @wonderaofficial needed more than half a million GPU hours to train AI models for music generation, voice conversion, and audio separation. Using
news
2026年6月1日