Stargate struggles with the details.
- Now that Stargate is at the stage where it has to start putting money into the ground, problems are emerging which have delayed the project and mean that its ultimate scope and scale may be a tiny fraction of that which was originally envisaged.
- Stargate was launched at the White House in January of 2025 with SoftBank, Oracle and OpenAI all touting that, in combination with UAE investor MGX, $100bn would be invested “immediately”, with $500bn in total to be invested over the next few years.
- Stargate is stretching the definition of “immediately” as six months have rolled by and there is barely anything to show for all of the hyperbole rolled out in January.
- However, there are plenty of signs that the project is in difficulty.
- First, Oracle: which stated last month that the Stargate entity “is not formed yet,” meaning that in practical terms, almost nothing has happened.
- Furthermore, OpenAI, clearly frustrated with twiddling its thumbs, has gone ahead and signed a $30bn deal with Oracle for compute capacity over several years, which has nothing to do with Stargate or SoftBank.
- Second, Disagreements: where OpenAI and SoftBank do not seem to agree on where the data centres should be built or how the energy should be sourced or whether SB Energy, an entity tied to SoftBank, should be involved.
- None of this is conducive to a fruitful partnership and makes SoftBank’s and OpenAI’s recent statements at SoftBank World 2025 difficult to believe.
- Furthermore, I suspect that MGX will be looking at this with some concern, which may prompt it to reconsider whether this venture is a prudent use of its capital.
- OpenAI is referring to the huge data centre being built in Abilene, Texas, as “Stargate”, but given that the venture has not yet even been formed, I find this statement very difficult to believe.
- Other sources also indicate that SoftBank is not involved here either.
- Taken in conjunction with the recent $30bn deal with Oracle, which obviously, should have been part of Stargate, leads one to be concerned that all is not well with the venture and the relationship.
- This puts both the venture and the further $30bn investment that SoftBank has committed to make (on top of $10bn earlier this year) in some doubt. ‘
- There is no reason why Stargate could not be formed and then take over the deal with Oracle and the Texas data centre, but I am starting to have concerns about the viability of a partnership between OpenAI and SoftBank as they appear to have difficulty in working together.
- Take this in context with the falling out with Microsoft and OpenAI’s precarious corporate structure, and one has to begin to wonder whether partnering with OpenAI is worth the bother.
- The net result is that in this climate, a failure of Stargate to launch is unlikely to cause it any real problems as OpenAI seems to have no difficulty in finding another $20bn or $30bn down the back of the sofa whenever its bank balance gets low.
- However, it does cast doubts on Softbank’s AI ambitions as well as the viability of long-term plans to invest hundreds of billions in data centres in competition with Microsoft, Google, Meta, Amazon, Alibaba, Baidu, Tencent and so on, who are all spending as much and as fast as they can on AI capacity.
- This looks an awful lot like overbuild to me in the same way that the Internet was overbuilt in 2000, meaning that some form of correction is likely in the next 1 to 2 years.
- This won’t be as bad as the Internet bubble when it bursts, but it will cause a reordering of the top table in this sector.
- This will cause a significant correction, especially in the valuations being applied to the model makers like Anthropic and OpenAI, and I do not want to be anywhere near them when this happens.
