FILE PHOTO: A NVIDIA logo is shown at SIGGRAPH 2017 in Los Angeles, California, U.S. July 31, 2017. REUTERS/Mike Blake/File Photo
(Reuters) - Italian startup iGenius and Nvidia on Thursday said they are planning to bring online one of the world's biggest deployments of Nvidia's newest servers by the middle of next year in a data center in southern Italy.
iGenius is building a data center that will house about 80 of Nvidia's most powerful servers, called GB200 NVL72 machines, each with 72 of the tech company's "Blackwell" chips in them.
The startup did not give a price tag for the project. But Chief Executive Uljan Sharka told Reuters that iGenius, one of the few AI startups in Europe valued at more than $1 billion, had raised 650 million euros so far this year and is taking on additional capital for the AI computing system, which will be called "Colosseum".
iGenius differs from rivals such as OpenAI in that it develops open source AI software models for chatbots it sells to banks, healthcare firms and other industries with strict data security rules, who run the models on their own infrastructure.
For Colosseum, iGenius has also tapped into Nvidia's full array of software tools, including one introduced this year called Nvidia NIM that functions almost like an app store for AI models. That means the AI models iGenius hopes to develop with Colosseum - some of which could be as large as 1 trillion parameters according to one measure of AI sophistication - can easily be distributed to any business that uses Nvidia chips.
"With a click of a button, they can now pull it from the Nvidia catalog and implement it into their application," Sharka said.
Colosseum will be one of the largest deployments of Nvidia's flagship servers in the world, Charlie Boyle, vice president and general manager of DGX systems at Nvidia, told Reuters.
Multiple Nvidia software and hardware teams are working directly with iGenius to bring the system online, he said.
"They're really building something unique here."
(This story has been refiled to reflect that Nvidia's software tool is NIM, not NIMS, in paragraph 5)
(Reporting by Stephen Nellis in San Francisco; Editing by Tom Hogue)
