Capital expenses (the famous capex) linked to the development ofTO THE In the United States they reached $ 250 billion in 2024 (+48% on 2023) and in 2025 they were expected to reach $ 320 billion (without counting Stargate the “launched” project from Trump who alone would apply 500 billion). We speak above all about datacenters and other infrastructures necessary to manage, preserve and process the billion billions of data from here the world seems to be unable to do without, to chips Super quick and performing NVIDIA style. These investments are similar as an order of magnitude to the total of the Draghi Agenda Plan for all product sectors in Europe, just to have a comparison.
Until a couple of weeks ago everything is fine, the American domain on the theme was obvious and the push to use theTO THE, In the workplace and not, it would have pushed all us humans, willingly or nolent, to spend and invest in the so -called lids and in everything that will apparently make our work and our lives much simpler, more simple and efficient.
Then she came, the very chinete deepseek and everything has changed.
The writer is certainly not a computer expert and less than less one Heavy User of technology and therefore will avoid any comparison between her and chatgpt or other similar systems.
Will report some facts that have come out in these two weeks:
- Deepseek democratizes access to the development of AI, opening the way to a proliferation of new models directly on smartphones, PCs, cars. It is in fact opensourceavailable and downloadable from everyone unlike American competitors.
- It has some important competitive advantages, also acclaimed by members of the Silicon Valley, especially in terms of reducing the amount of memory / chips used, ability to use much less input / parameters for the same result obtained, optimization of inferential processes, memory compression and reward functions (and here the writer raises his hands and refers to the Bain studio called Deepseek a game changer in ai efficiencyjust published).
- It would cost much less. And here the advantage would be twofold because it could further accelerate the development of the AI, multiplying investments on a numerical level, because it is accessible to many more people, companies, institutions. Here a nice graphic designer, just by Bain who shows you the current situation.
So all right? I wouldn’t say. The reaction of the markets was first very worried; The question is: there was a excess of investments in AI especially by the Mag7 or Hyperscalershow are they called in the graph by JPM AM? Are we facing a second case similar to that recorded only 3-4 years ago on Metaverse? Soon to say it.
Without a doubt, however, the calculations say that the 250 billion of capex spent in 2024 they require revenues for 500 billion dollars in the field of AI in order to maintain 50% of the average gross margin with which these companies live today, under penalty of reducing their marginality, the multiples to which they treat and consequently the scholarships. The declarations made so far by them project 100 billion as you can see below. When and how will the other 400 arrive?
We close with a curiosity dedicated to those who were amazed to see Deepsek appear from China. You know how many were the Chinese who participated in the drafting of the 2017 Paper who laid the foundations for the subsequent birth of Chatgpt? 15. Out of 25 experts in total who contributed.
