From the outside, building gigawatt AI Data Centers and Power at scale, sometimes seems like just putting a whole bunch of legos together.
Investors have been aggressively placing their bets this AI Tech Wave. With the expectation that building this infrastructure at the clip of hundreds of billions of annual investments, is a straightforward execution exercise.
“Data center joint venture launched at the White House has not staffed up and is not developing OpenAI’s AI data centers.”
The TL;DR takeaways are as follows:
“OpenAI and SoftBank disagreed over how to develop data centers together.”
“OpenAI’s does not plan to build its own first-party data centers in the near future.”
“OpenAI missed its goal of contracting 10 GW of capacity through Oracle and SoftBank by the end of 2025.”
It all began with high level announcements in DC:
“When President Donald Trump announced the $500 billion Stargate AI data center project in January 2025, the three companies charged with developing it promised to move with lightning speed, spending $100 billion almost immediately to start building 10 gigawatts of computing capacity.”
“More than a year later, the Stargate joint venture has not staffed up and is not developing any of OpenAI’s data centers, according to three people involved in the shelved idea. In the weeks following the announcement, Stargate floundered, lacking leadership and coordination, these people said. Stargate’s three members—AI model developer OpenAI, cloud provider Oracle and Japanese conglomerate SoftBank—disagreed over who would do what and how the partnership would be structured.”
Remember, a Gigawatt data center runs about $50 billion today and takes two to three years to stand up with the needed equipment and power.
With all the inputs in short supply, there are few short cuts, especially on the Power infrastructure front. The bigger complications are around organizing the complicated financing and organizational priorities, and negotiating with all the key partners up and down the chain. This is starkly clear from the piece:
“Earlier that year, OpenAI had sketched plans for the project, but the Oracle negotiations had put it on hold, pausing construction. OpenAI lost its general contractor because of the delay.”
“Still, it didn’t take on the project itself. Instead, it formed a partnership with its other Stargate partner, SoftBank, and said publicly that it would work on the site with that firm’s subsidiary, SoftBank Energy.”
“Behind the scenes, the two sides clashed over control, according to two people involved in the discussions. OpenAI had hoped the Texas campus would become its first self-built data center. SoftBank, meanwhile, wanted to develop and own the project.”
“Between September and October, the Stargate team took multiple trips to Japan to work out the problems with SoftBank’s Masayoshi Son.”
The whole piece is worth a full read, if only to understand the wide delta between picture ready project announcements, and the messiness of executing through on the ground realities.
That company’s anecdotal story above is helpful in understanding the realities of execution in this AI Tech Wave, for all the companies involved in their own projects. Building it all at these scales is really hard to do in normal circumstances.
It all gets even more complicated with global geopolitical realities with complex private and government supported partnerships. Stay tuned.
(NOTE: The discussions here are for information purposes only, and not meant as investment advice at any time. Thanks for joining us here)
Podcasts Michael Batnick talks with Ian Wenik of CItywire about RIA growth. (youtube.com) Michael Sidgmore talks with Bill Crager, Co-Founder of Envestnet...
