The weekly download: No 17
Bubble or no bubble, we will be building companies on top of the AI infrastructure boom for at least a decade. Some charts and quotes to think about on that theme...
The noise around a potential bubble centred on AI continues unabated. I have no idea whether it is a bubble or not.
If there is a bubble, it is fuelled by the high level of investment in data centre infrastructure, amplified by a web of cross investments and deals amongst a handful of major companies.
Underpinning all the speculation and the eye watering numbers, real assets are being built. Some of those trillions may evaporate. Nonetheless, founders and leaders will be building businesses on top of this infrastructure for at least the next decade.
So this week I want to share a handful of charts that pick out some interesting points about the story:
A 10 year view
Starting with this which shows the 10 year capex trend for Meta, Microsoft Amazon and Alphabet. Seen from this perspective, AI is simply an extension of an existing trend.
The next three charts are all from the same source - Understanding AI, 16 charts that explain the AI boom
Location, location, location
US data centres are heavily concentrated by geography. A key driver of this appears to be electricity costs.
Amongst all the talk of sovereign AI and nationalism. Countries in Europe and the Middle East are desperate to attract some of the investment boom for their own economies. Unless the deal includes sustainably low costs of power, few if any of these investments will prove durable.
Rents and capacity
Rents are only just starting to rise. This suggests capacity build has been running ahead of demand until now.
As a UK taxpayer, this makes me nervous. Together with the electricity cost point, I can see the ideal conditions for a whole herd of white elephants.
Big tech has the cash
Operating cash flow for Google, Amazon, Meta, Microsoft, Oracle is greater than capex. This level of investment is sustainable for now. Oracle is the weakest position and needs to borrow to keep up. Intuitively this makes sense.
OpenAI, Anthropic, and xAI do not have this luxury. They need investment or loan capital to keep up.
Overall, it does look as the level of investment is sustainable for now. The revenue may take some time to emerge but the main players can afford to wait.
Outside the mainstream
Let me finish with two quotes:
“Chinese companies are pioneering a different path to global AI supremacy: combining world-leading open-source models with domestically produced inference chips to create a scalable, cost-effective deployment ecosystem that could capture the majority of global AI usage.”
When I posted this online, I had a comment from the CEO of Verso Industries, Michael Zimmerman:
“We have an architecture that runs on regular cpus instead of clusters of graphics cards. Drove a 157.5x reduction in memory and compute.”
Perhaps the real challenge for big tech will not be an AI bubble but tech innovation that overtakes their model?
Thanks for reading
Reasonable protection. Good question at the end. It's whether people want to be active in answering it. Or just observing as others answer it with their actions, with big tech calling the shots.