One area I?d push back on a bit: your confidence AI compute shortages could last 10-20 years

I just enjoyed the new quote feature. In replying to you I just selected your text above the editor, then touched (I?m an iPhone right now, so 'clicked' if you have a mouse) the word 'Quote' to the top-right of the editor and in goes your quote at the top of the post.

Back to your question, I concur with you entirely that the hardware supply constraints have historically been short-lived, and I think the current really high memory price will work itself out over merely 3 years or so (though I don?t know for sure - it could be 5 or 6 also), and nothing like the 15-20 general infrastructure supply constraint that I'm speculating.

Where my ludicrous 15-20 figure comes from - and it is just a conceptual frame, I don?t mean 17 years but rather I mean there is a much larger overall demand cycle that is going to continue far beyond the individual category cycles currently in the press particularly memory supply.

Both copper supply and electrical grid transmission line expansion are facing structural shortages expected to last 15 to 20 years.
Copper mine development and regulatory approvals for transmission grids both commonly take up to two decades, creating deep deficits extending toward 2040.

But my vision is that application development will ramp up way more than everyone is modelling. This happened in the 80s with software building putting huge pressure on simple serial compute demand, which it couldn?t match continually and led to companies like Motorola and Intel appearing as chip giants for the first time. Their huge earnings didn?t come from businesses and the public abstractly wanting chips actually, but demand arising from the huge and relentless software buildout over 20 or so years creating that demand.

What I predict is that application development involving AI has an insane runway ahead for productive improvements that is really, really early now; and the market isn?t really correctly modelling it. It is looking at the application world as we see now and understanding that - correctly - supply will work itself out over a just few years.

What is instead going to happen is that there will be business after business - small medium and large - and even new software industries starting that don?t exist now such as increasingly productive manufacturing robotics - that will take a decade or two, rather than the 2-3 years on everyone?s minds, to even conceivably start to reach a point where the middle software-hardware could composite layers are failing to 'find new ideas' that demand much more compute to give more productive results for the end-business.

In a sense we had a rally long-term lull in software advancement ideas the last 10 years or so, smaller than the relentless pushes that put demand on compute through tje 80s and 90s - and what is going to happen is a massive new demand surge at the software-middle-hardware buildout layer.
Not being properly modelled.

So that’s about the software layer and how that changes over time to create new and ongoing demands not really being modeled now.

On top of that there is energy constraints - specifically the generation of raw electrical power - which are projected to be a multi-decade challenge that will fundamentally reshape industrial planning and tech development through 2040 and up to 2050.

Firms like Brookfield Corporation will benefit from that too with the fees from their BEP partnership sub.

About Meta and Google and those firms with massive social moats (I'm putting search under that category ‘social moat') in sense it is the combination of seriously gigantic data owned and social habit formed), they will reinforce the social moats with infrastructure scale moats - and doing a bit of a Coke on that regard actually. The infrastruee can have huge incremental return on capital as we saw from Amazon's buildout - with a 5-10 year delay before the huge profits are reaped - but the buildout has multifaceted benefits in that it it helps solidify with thick moat-concrete-plus-alligator the social moats.

There is really nothing like it in history I?m not afraid to say (knowing how stupid that sounds) because on the past there were not the same possibilities for social/tech strangulation as the past (such as News Corp's distribution dominance, which had a moat and still does but which everyone walked away from so their revenue *under that* moat disappears). It’s very important to understand that the revenue generation is always independent to the moat itself for example Google's ads generate the revenue but the moat is search dominance - data ownership - social habbit).

So where the real moats are there is going to be a lot of winner takes all effects, such as infrastructure leasing woth only the very largest scale providers (a few must exist tougher with dominance in different niches) and for advertising generated from social moats.

In short as far, far more is done in software say in 2045, than people reason today. That, with its ramifications, is my central insight. One can project several variables forward but keep other variables the same ("software will be something like today") and make the projection error on that basis. there are going to have massively more ways to suck 'fees' out of the whole system (*for those with the real moats*) without the public really able to escape. So many things will change but the social habits and data ownership alone are really really hard to do anything about for competitors, and the infrastructure just makes the existing moats far thicker concrete laid, along with all the new giant alligator species place in one by one. Meanwhile new revenue streams inside those moats are going to be produced, as my prediction playing out ahead of time with Google's compute accelerating toward the 30% of total in a crazy short amount of time. Meta will definitely have new revenue streams say 8 years away, which will put to halt Wall St laughing at them for only being able to protrude ads. They are barely getting started with how they can utilize their obscene data ownership / social habit / compute infrastructure moat-conglomerate.

- Manlobbi