Interesting read! I’d raise one key ‘unlike oil’ point — and that is that compute is not discovered, but made. It’s never going to be ‘found’ somewhere with strategic geopolitical interests, which makes it unique (and certainly has implications for each of the other ‘like oil’ points)
I buy your arguments that compute is the current bottleneck.
And yet, I am old enough to remember the late 90s when outrageous amounts of optic fiber and switches were installed around the world. Then technology evolved (multiplexing and so on) and we were left with massive amounts of unused overcapacity (keyword: Dark Fiber).
The analogy is not perfect, but it is not implausible a scenario where we see continuous and extraordinary investment in data centres for the next two years and then some break through innovation comes along and we realise that for 99% of the use cases, we only need 10% of the installed compute. I rarely see this scenario being discussed.
One interesting parallel with the fiber buildout -- at one point, according to wikipedia, "the amount of data that could be carried by an optical fibre was doubling every nine months". Very much like chip capabilities, and AI capabilities themselves!
It's striking how this never asks the obvious question: if compute is so strategically valuable, where are the productivity gains? Where's the economic transformation to justify these valuations? It assumes strategic importance from resource concentration rather than demonstrating it through outcomes.
High valuations justify massive compute spending, which justifies treating compute as strategically important, which justifies high valuations. The whole edifice becomes self-reinforcing until someone asks the awkward question about where the actual value creation is happening.
Interesting read! I’d raise one key ‘unlike oil’ point — and that is that compute is not discovered, but made. It’s never going to be ‘found’ somewhere with strategic geopolitical interests, which makes it unique (and certainly has implications for each of the other ‘like oil’ points)
Agreed with this framing. In light of this, what are your odds of Chinese military action in Taiwan over the next say, 3 years?
I buy your arguments that compute is the current bottleneck.
And yet, I am old enough to remember the late 90s when outrageous amounts of optic fiber and switches were installed around the world. Then technology evolved (multiplexing and so on) and we were left with massive amounts of unused overcapacity (keyword: Dark Fiber).
The analogy is not perfect, but it is not implausible a scenario where we see continuous and extraordinary investment in data centres for the next two years and then some break through innovation comes along and we realise that for 99% of the use cases, we only need 10% of the installed compute. I rarely see this scenario being discussed.
One interesting parallel with the fiber buildout -- at one point, according to wikipedia, "the amount of data that could be carried by an optical fibre was doubling every nine months". Very much like chip capabilities, and AI capabilities themselves!
It's striking how this never asks the obvious question: if compute is so strategically valuable, where are the productivity gains? Where's the economic transformation to justify these valuations? It assumes strategic importance from resource concentration rather than demonstrating it through outcomes.
High valuations justify massive compute spending, which justifies treating compute as strategically important, which justifies high valuations. The whole edifice becomes self-reinforcing until someone asks the awkward question about where the actual value creation is happening.
Thank you both, Peter and Erich Grunewald.
(All Erich for this one!)