$1T of AI spend in the next few years. That is unsustainable. Maybe not. If in 3-5 years time, 20% of white collar workers in the OECD are "replaced" by AI agents with a 50% saving as to cost, then there is an ANNUAL saving of $1.78T to be realised. Now here's another one. That's $1.78T of savings is "net savings" based on a 50% cost. What does that mean? It's a $3.56T reduction in human opex by spending $1.78T of cost to deliver a net $1.78T of saving (that year). As highlighted by @Jason on All in Podcast this week, @AnthropicAI made an additional $10B in ARR last MONTH! This is the fastest in the history of business (at least on this galaxy). I'm going to say they will increase 1,000% in ARR this calendar year from US$9B in Dec to north of $100B in December 2026. OpenAI will launch a competing agent product soon and they should see an increase in growth too. With upcoming performance gains of the next gen after @nvidia Vera Rubin/Groq plus TPU/LPU's from others, the cost to generate tokens in 3 years will be around 1/100th of today. So the need for more infrastructure will diminish? Doubtful. I believe the consumption of tokens will increase between 2-5x that improvement (ie 50,000%). So the build out will need to increase another 2-3x in the next 5 years to support that transition. AI and AI adjacent companies along with AI native companies and companies deploying AI in an efficient and effective way will capture the the majority of the Agentic AI value transition. Other than chips, AI factories (neoclouds) will start to get the majority of funding over traditional data centres as they typically have a closer alignment on investment horizon. When silicon costs 6x the facility their housed in, you appreciate that by the time the next upgrade cycle comes the facility is probably not fit for purpose. Crypto miners understand this model well. To be clear I'm not cheerleading this. It's just what I'm seeing. The main bubble here is the assumption this is simply a mathematic slide transition. It won't be. The transition between human capex/opex (office/people) to AI agents (AI factories/AI agents)is a simple substitution algorithm with a time horizon. It won't. Or that AGI isn't going to change and possible dislocate that transition. It will. Then quantum etc. etc.. But one last thing. Even when Governments try to force companies into unproductive labour practices to retain staff, it won't work in the medium term. All it will do is erode the businesses capability to compete. Any business with significant human process that can be automated will be disrupted by AI native newcomers. All that legacy burden is simply an arbitrage opportunity for AI natives. Buckle up the speed and need to build infrastructure has only just started and it's inevitable. PS: I threw a crazy scenario in here which is not going to happen. Well not without the end of days and UBI.