Microsoft has taken over the commitment to rent data center capacity at a site in Norway, originally intended for OpenAI’s Stargate venture.
The site will provide Microsoft with access to 30,000 Nvidia Vera Rubin chips and expand on the company’s previous $6.2 billion commitment to the site. It is the latest pullback by OpenAI on its Stargate project, suggesting the startup may be looking to reduce future spending in preparation for an IPO later this year.
Just last week, OpenAI announced it was pausing the Stargate project in the United Kingdom, citing high energy costs. Both the UK and Norwegian sites are operated by AI hyperscaler Nscale, which has also found another AI operator, Google, to rent the capacity in London. It seems the old guard is more than willing to step in when startups reduce their exposure.
Last month, Microsoft moved into a facility in Texas that had been abandoned by OpenAI and Oracle and was also part of Stargate. It provided Microsoft with 700 megawatts of capacity and was located right next to the flagship Stargate data center.
While OpenAI and its two partners, Oracle and SoftBank, are still on the books for $500 billion by 2030, it has reduced its exposure, especially outside of the US. The project’s total capacity is expected to add 7 gigawatts, with 1.2 gigawatts already under development.
Backlash to the buildout
The buildout of AI data centers has worried some, especially those living near one.
OpenAI and several other developers have promised that these centers will not raise electricity prices, but local campaigners are mounting growing opposition to their construction and operation. This has led some to propose construction further afield, like in the sea or, in Elon Musk’s case, in space.
The reduction in buildout comes as OpenAI has refocused development on its key services, such as ChatGPT and enterprise GPT integrations. It recently shut down Sora, its AI video-generation app, which was reportedly costing OpenAI a lot to keep running, despite low consumer interest.
This may be partly due to the strong run of form of its key rival, Anthropic, which has had a few months of positive press for its AI services in the enterprise market. Anthropic, compared with OpenAI, has far less AI investment on its books, preferring to rent capacity from the three major cloud providers.
While OpenAI recently received an additional $122 billion in funding, there are worries that, with the current costs of day-to-day operations along with the buildout, OpenAI will likely not be profitable for many years. This is more acceptable in private markets, but if OpenAI plans to IPO this year, it will need to present a plan of action to reach profitability quickly, or its valuation is likely to plummet.
Also read: OpenAI’s Stargate pause is already evident in Europe, where the company recently put its UK project on hold over high energy costs and regulatory uncertainty.
