AI Bubble? What AI Bubble? Datacenter Investors All In Despite Whispers Of A Pop

Fears that AI may be a bubble about to burst have yet to dent datacenter investment, with a handful of new developments revealed this week.

GPU cloud provider CoreWeave has struck a deal with Nordic operator Bulk Infrastructure to serve up a large-scale Nvidia compute cluster from the latter's N01 Datacenter Campus in Norway.

According to CoreWeave, its infrastructure using the GPU maker's DGX GB200 NVL72 server platform installed at the site will make it one of the largest Nvidia-based AI deployments in Europe.

Bulk has secured 400 MW of renewable energy for the facility, with the potential to expand up to 1 GW.

Expected to be operational by summer, this project is understood to be part of an initial investment of $2.2 billion announced by CoreWeave across Europe for building new datacenters and procuring available capacity.

Karianne Tung, Norway's minister of digitalization and public governance, says the Norwegian government's goal is to establish a national infrastructure for AI toward 2030.

In Japan, investment biz SoftBank plans to convert a former Sharp LCD panel factory into a datacenter to operate AI agents developed jointly with ChatGPT creator OpenAI.

The Japanese megacorp confirmed it has signed an agreement for the site in Sakai City, Osaka prefecture, of approximately ¥100 billion ($672 million), and intends to have a bit barn with a capacity of 150 MW operational by 2026, with a possible expansion to "more than" 250 MW.

SoftBank says the facility will develop generative AI and will also open up the datacenter to universities, research institutes, and enterprises.

The move is - we think - part of an agreement with OpenAI announced last month on what Softbank called "Cristal intelligence," which it said will "securely integrate the systems and data of individual enterprises in a way that is customized specifically for each company." SoftBank is also taking a $40 billion stake in OpenAI.

Over at Supermicro, chief exec Charles Liang said at the HumanX AI conference in Las Vegas earlier this week that it built the Colossus datacenter facility in Memphis, Tennessee, for Elon Musk's xAI project in just 122 days.

• AI running out of juice despite Microsoft's hard squeezing
• Microsoft tempted to hit the gas as renewables can't keep up with AI
• Frack to the future? Geothermal energy pitched as datacenter savior
• Amazon, Meta, Google sign pledge to triple nuclear power capacity by 2050

Colossus, with its 100,000 Nvidia Hopper GPUs, was aided by Nvidia pre-fabricating the racks before shipping them to the site, where they just needed to be connected to power, data, and cooling, Liang said in an interview. A funding round last year secured xAI $6 billion for the project.

According to xAI, it has since doubled the size of the infrastructure in 92 days to 200,000 GPUs, and has now acquired a second 1 million sq ft site in Memphis to expand operations. Previously it said it intends to expand to a million GPUs.

All of this investment comes amid a growing feeling that the AI market may about to implode. Many business leaders are unconvinced that AI is worth the expense, and Baidu CEO Robin Li described the AI sector as an "inevitable bubble," similar to the dotcom bubble that went pop around the turn of the millennium.

Microsoft, which has itself sunk billions of dollars into OpenAI, recently pulled back on datacenter investments in some places, according to TD Cowan.

The investment in AI infrastructure still heavily outweighs customer spending on AI licenses. ®