AI is the biggest industrial revolution in human history (whereby humans create digital intelligence through token generation), but at the same time, contrary some of the recent paradigm shifts, it is also the most capital intensive – AI is driving massive innovation, but its potential increasingly depends on a massive upgrade of the installed base of infrastructure. Nvidia CEO Jensen sees $ 3 – 4 trillion in AI infrastructure spend by the end of this decade.

This massive upgrade to installed IT infrastructure (especially for compute) is a multi-year process and is also the reason why we remain so bullish on Nvidia and TSMC.

Below is the progression of the total capex spent by the top-5 hyperscalers in the world (including OCI), which is on track to go from $ 97bn in 2020 to $ 368bn in 2026e – that’s a 4.5x increase in total capex in 6 years.

By the way, total Oracle capex should be around $ 35bn this year versus total RPO (total order backlog) of $ 455bn by end-FQ1 which will necessitate a massive uplift in capex towards compute if the company wants to service these new OpenAI related contracts… So, this chart will only continue to move up to the right if you believe OpenAI can raise the necessary funding to pay Oracle…