OpenAI Stargate Project - The Beginning of the End?
It's a mixed bag
Someone recently asked me what I thought about the recent announcement from OpenAI/Nvidia/Softbank/Trump about the Stargate AI data center project. I hadn’t looked into it very deeply until then, but now I have and here is my take:
Big picture: It’s complex and details are sketchy, but this may be an early indicator of the beginning of the end of the current AI boom.
I think we’re at a peak in this latest AI cycle - there have been AI winters before. Not that there aren’t potential positives to come out of AI, but I sense that the hype is outrunning the reality and that will cause a retrenchment and bursting of the bubble.
For example, there are signs that the business models for companies like OpenAI are not working. Training and running AI in these massive data centers is extremely expensive. Sam Altman recently admitted that OpenAI’s new high-end $200/month subscription to ChatGPT Pro is not profitable, even at that high price.
I’m also very skeptical of Altman’s claims of nearing the achievement of AGI. Missing that (dubious) goal will damage OpenAI. We also seem to be approaching the limit of practical uses of current AI.
Stargate levels of over-investment in emerging technologies is a pattern. I’ve been there and benefited from it. Way back in the day, Philips and Sony invested something like $1 billion in the failed Compact Disc Interactive (CDi) set top box in the heyday of the CD-ROM hype. And then, of course, there was the Dot-Com bust.
Not that it was all bad. Members of the CDi cohort referred to ourselves as graduates of “Philips University” because we learned so much about interaction design and digital content production. These skills later fed the web and many other software driven fields. And of course, the web became a major economic driver, and opened the way for mobile.
I do foresee the same kind of positive fallout for AI in the long run. But there will be many casualties along the way (sadly, my stock options in an internet radio startup are worthless…)
Impacts
So how is this deal going to play out for the organizations involved? My guess is:
Oracle, SoftBank & Nvidia: Probably a net positive. Oracle is rapidly moving towards AI and needs this infrastructure. Softbank is an experienced, well diversified Japanese financial powerhouse that can handle the risks and will reap the rewards. Nvidia is just selling shovels to the gold prospectors.
Microsoft & MGX: Probably Mixed. Microsoft is kinda getting cut out of the OpenAI train. I’m not sure the UAE based MGX has the expertise to make this work for them.
OpenAI: Probably a net negative. I fear the debt-load of this deal will hurt OpenAI. I’m not convinced their business model will work well enough to justify this kind of investment. And I think the general trend in AI is towards smaller, less resource intensive models and won’t require such ridiculous amounts of energy and data center resources. The same thing is true of AI hardware - it’s becoming smaller, cheaper and more energy efficient. Almost every day, I run effective LLMs locally on my laptop - no server or privacy concerns required!
The US Public: Probably a net negative. The major problems with the struggling national energy grid are numerous and will not be solved by the simplistic solutions coming out of this agreement (i.e. allowing power plants unconnected to the grid). So the situation will only get worse. Plus, the climate impacts of Stargate’s extra energy production/consumption will hurt to the already deteriorating environment.
I’m also skeptical of claims of hundreds of thousand of jobs magically materializing. Yes, there will be new construction and provisioning jobs, but those will go away once the data centers are up and running. Note that most data centers operate “lights-out” because they need so few personnel.
So, I think Stargate is a decidedly mixed bag.
For those wondering about how to approach AI, my recommendation is to Go Slow and Make Things Better rather than Go Fast and Break Things.