Ep 9: Apr 25 (Q1 25) – Tariffs and Summary of main Technology Trends

• Intel got a new CEO, and the strategy will revolve around re-energizing the Intel Foundry, working as an outsourcer for brands like Nvidia and Qualcomm, and restructuring the part of the company too inefficient. The potential benefits also go in the direction of accelerating production back in the US of semiconductors, putting further pressure on TSMC, which is already aiming to produce more in the US, the 2nm, and potentially the 1.6nm, as formerly discussed.
• Q12025 Summary: The first quarter showed a big competition for semiconductor supremacy and their production returning to the US. The pitch we saw from JD Vance at the Venture Capitalist event in March is to combine production and design in the US, bringing back innovation under US supremacy. I analyze more in the section on Job Evolution because I think the pitch is quite valid in some areas, especially related to innovation vs stagnation from cheap labor, but I put in the equation also feasibility from a workforce point of view, and how that could be approached. Finally, I just remind that the AI Chipset export restriction will come from the former US Administration within the first 120 days of the year, and this will be on top of any tariffs strategy.

The most recent tariffs discussion from the new USA Administration is posing a clear topic on the dependency of each other’s markets and the synergies lost. If on the usual market (steel, coil, etc) is possible to have different sourcing, when we land to discuss around technology, there is a strong dependency on the US market on tech services, and there is today no real option to balance without a proper homework that would take many years to create autonomy.

Looking at how Europe, and a big part of the world, is depending on US technology, we can see that over the last 30 years, the US has built a strong ecosystem that, for example, Europe consumed from without, creating a proper alternative base. China, in the last 10+ years built some level of technology independence, building its own OS for mobile phones and having some other OS for computers based on forks mostly of Linux and other infrastructure services with its own technology. China also built its own semiconductors, chipsets, and AI, as well as datacenters. Most recently, Europe also spoke about the wish to have an EU Operating system, but let me say that things are more complex than just an OS.

When I wrote in October last year in my newsletter on the evolution of the market with the EU Crisis and Draghi’s receipt for transformation, I mentioned that Draghi was considering the last 30 years of US development in technology a huge advantage that could no longer be closed by the EU in some areas like Clouds, but could be worked out in AI. Looking now to the tariffs (that don’t apply today to services but only to tangible goods), there is an important element to consider from two aspects, one related to the cost of a service that could be raised in the future (i.e. applying tariffs to cloud usage) but even more relevant the risk to be “disconnected” by a service operated by a company in a country with a different relationship (i.e. recent Starlink discussion), impacting on the own sovereignty of each country.

Technologies are built on layers, making more and more levels of abstraction and making users not see the underlying complexity, and enjoying easy way to use. However, each layer is done by one or more technologies, and the dependency on a stack of technologies defines the dependency on who is producing and maintaining that stack. Without landing into doing an ISO-OSI stack comparison, let me use an easy example to make it understandable to everyone:

• Physical/Backbone: Most of the network equipment is US (i.e., Cisco) from hardware and technology, and most of the infrastructure globally runs on these technologies. Alternative vendors are often US-based or have dependency in their chipset, architecture, protocols (each network device has chipsets) on US vendors (again, we have Intel, AMD, and in the area of network, there are also decent European entities and Chinese, but still, the US is big here)
• Datacenters: Run on US Operating Systems (limited by some open source with Linux and some good EU alternatives for Linux solutions) and run on US Virtualization layers (where we can exclude some of the open source again)
• Storage and Chipset: Many US but also where we have alternatives, the OS of those storage runs not always proprietary.
• Clouds: They run on Datacenters, so even if they are in the EU and from an EU vendor, they depend on what they have inside on US technologies.
• SaaS applications: They run on Clouds, so they depend on that layer before.
• End users: Run mostly on US OS (including mobile phone OS) and devices (here we have alternatives, but the hardware chipsets are again mostly US)
• Application level: Here, I don’t want to even start. Just remind yourself of your work suite on your PC and think which provider built it. Then we have security layers, etc., with many presences from US.
• Data and AI sit on top of the others. So all the other layers influence them.

I stop to make further layers of explanation, but we could do that more deeply and more accurately. Definitely, we will have exceptions, definitely we could build a certain stack with minimal or no dependency on the US, but it would represent a specific case that does not reflect the most enterprises and even consumer reality we have today. Here, I just mentioned technology services, but any physical product using those services is also influenced by that. I take one example that is the automotive market, which is more and more cloud and tech connected, and many technologies of those cars run on the cloud of the US, and with technologies from the US. This can be applied to other types of products where there is a technology inside. I just want to make clear how tight and interconnected we are with technology that has been built and developed accurately by the US, with a lot of influence and evolution from the rest of the world, and that has brought over the last 30 years a lot of innovation. China, which worked over the last 10 years on reducing the dependency, definitely has still some areas to decouple but worked out already on chipset layers, worked on OS layer and I expect some level of virtualization and storage including OS. Also worked out to reduce security layers depending on US technologies as this would be strategic in a cyberwar. All this to say, that a fight to decouple dependency is long, big and is requiring long investments to try to reach the same level of maturity we have today. In the section on Job Evolution, I evaluate how the production back to US could be influenced by Technology to be achieved considering the equation of job labor needs and cost mitigation for inflation.

• So far, I have seen mostly a decent level of marketing against meme coins. Many people experienced losing money in such a type of game and stopped investing in it. This is, for me, a good result as I care to have small investors not end up in such traps. Other crypto coins are suffering as we mentioned last month and will most probably continue for some time, such as this trend. There is an up and down as more investments are promised in the direction of Bitcoin, and so far, it seems this will be the only real crypto to be sponsored by the US government. I still have no clarity around the stablecoins.
• Q12025 Summary: The first quarter has shown a big change in the crypto market, from last year ramping up (until mid-January) to the highest levels and then sinking really fast in February, together with the many new memes that accelerated the negative hype versus the entire crypto market. However, the real cryptocurrencies have a certain meaning even if, at the moment, it seems that investment from the US government is mostly oriented in the Bitcoin-only direction. I always recommend diversifying and using crypto only in small parts of investments.

• On the micro-reactor energy production, we see some progress ongoing, for example, also in the US. The need for distributed energy near data centers without impacting the grid distribution is going to accelerate.
• Similarly, in the area of nuclear energy by fusion, we see startups ramping up also with the focus to refactor former fission nuclear plants, like happening in Germany.
• Q12025 Summary: Some progress in the area of fusion nuclear and micro-nuclear reactors for fission in very small reactors. This quarter saw the acceleration of alternative energy production sources to compensate for the big demand from different areas. Fusion seems to be progressing over the quarter, with 10x times the ability to keep plasma controlled to support fusion, and China is leading the record after France’s former achievement. This seems promising to get real in the next 10 years. Meanwhile, the production of micro-reactors with minimal time to switch off and reduced risk of fissile drop is growing. There is a clear race for energy production lead, considering this as the requirement to lead the AI revolution. The US is highly focused on this lead.

• Some companies like Apple have kept their focus and support on DEI so far and are giving a certain stability in their organization. However, they recently showed some openings on the privacy aspects that were a key reference of their brand. I just reflect if these will influence the internal opinion on the enterprise, as happened with Google (as I mentioned in the former newsletter)

• So what? Microsoft announced in March that they are delaying building some AI datacenters, as they are also announcing they will adopt the improvements coming from DeepSeek. This was my prediction in the newsletter of two months ago related to January, and is going to influence the overall Tech investments as the market is getting more efficient in delivering AI. This is going to democratize AI.
• The most recent Deepseek V3 runs also on consumer hardware. This is going to definitely impact the need for stronger hardware and clouds, and could decelerate the Nvidia race even if they were having a pipeline of demands already filled for years.
• Costs Optimization: I keep reminding you to have a continuous check on the AI pricing and avoid too long a commitment, as this pricing will drop.
• Q1 2025 Summary: We saw a ramp-up around discussion of agentic as they are developing more as independent solutions. At the same time, the Trump administration is cutting many jobs in the government and announced they want to build up more high-salaried jobs in the US, combining design and manufacturing. A first insight from me is below, but then in the Job Evolution section, I analyze in depth how the AI could influence the Job Evolution market with tariffs.
• My Thoughts: I predict that the jobs gap in the US (already open jobs), extended by the need for more advanced jobs to bring back a full integrated products lifecycle in the US, will create a challenge on the availability of resources and how to keep labor costs low to keep the overall products affordable. The way I see this formula could fit is to use agentic and robotics to make manufacturing much more autonomous (not just automatic but Industry 5.0) and higher-salaried jobs at the top of the chain to grow this structure. So my view here is that a big part of the gap of many jobs expected could be covered by autonomous agents (physical and not). I make a deep dive reflection on this in the section on Job Evolution.

• There is still an acceleration in the conversation around humanoid robots, trained with AI and learning to walk and move more like humans. Even a few tests for early adopters to have a humanoid at home, like the Swedish brand 1X is trying to do. Still to clarify how people like to have such a presence in their home, either for privacy reasons (camera watching) or for human safety confidence.
• My Thoughts: I just don’t challenge AI safety at this stage, but I challenge the capability of a tech environment to not be hacked. What would happen if a hacker took control of a remote robot and forced it to do actions that should not be done? We always reflect on how to guarantee that an AI is not acting wrongly, and there is a consensus on its way to operate, but there is still the Byzantine fault, as a bad actor that could influence the way such autonomous agents operate in a thousand ways. The AI governance and, in general, the freedom of such equipment to walk and be autonomous in actions on the premises of houses or fabrics need to be properly thought out. Today we have automatic actions, robots with specific tasks and areas of coverage, and they are predictable and with a specific purpose, not a generic, open way.

• Recent training of OpenAI ChatGPT with Hayao Miyazaki, building Ghibli-style AI images, went viral and pushed OpenAI to suspend as copyright concerns came out. The solution was really brilliant and came out as really nice models, but the point of copyright goes back to something I’ve been saying for many months in the newsletter on the training of AI engines and how to guarantee proper respect for the sources. This is definitely not a case of only OpenAI, as we see many other AI engines and LLMs built also with techniques of distilling that are losing the source of training data references. I reported these concerns for many months, also referred to the regulations that need to be in place, and on which Europe, with the AI Act, started to define formally, in the governance of how and what has been used to train an AI. This is even more sensible as we do that in the boundary of an enterprise, keeping attention also on reputation.

• Q12025 Summary: We saw an acceleration in the overall Quantum Computing through the approach to reduce/mitigate error creation rather than trying to build bigger engines. The strategy to reduce errors by modularly combining and extending power is going in the direction of enlarging power and progressively distributing and reducing the risk of error generation in the results. Quantum Computing is preparing for a real revolution in power to compute (billion times faster than the most powerful supercomputer), giving capabilities in areas like complex medicine and biology computations, potentially solving NP-Complex algorithms, but also posing a threat to overall security, making it much easier to crack most advanced encryption techniques unless properly adjusted. In the next 3 years, Gartner, as I mentioned in the former newsletter, is highly recommending investing in updating applications to PQC (Post Quantum Cryptography) to be in compliance with future security requirements. This is a big investment that is not necessarily on the radar yet for each enterprise, as it should.

This month I want to reflect on the Job Evolution due to Technology, taking into consideration a few base documents for the conversation:

• The Future of Jobs (World Economic Forum 2025)
• A New Future of Work (McKinsey)
• The future of European competitiveness
• The population pyramids by 2030 and beyond
• The JD Vance recent pitch at the American Dynamist Summit

Take this as a short analysis, and it is not planned to be comprehensive. I recommend taking the time to read the links I shared, as there is much knowledge available in those references.

So what are my takeaways from the read as I cross those figures? Please accept that I focus only on the technological and not the political component, as that is my area of interest.

Starting from the pitch of JD Vance a few weeks ago is clear that the US Administration wants to bring back production to the US, as this is going to bring back more control on the US-driven innovation. The pitch is quite effective in explaining the recent history effects of globalization from the innovation point of view, mentioning that cheap labor from outside the US has been a way to make products more attractive rather than really innovating. So the effect of production outside the US brought progressively more design also outside the US, reducing the innovation supremacy. Listening to the overall presentation, there is a clear focus on bringing back stability, also in the workforce, as deindustrialization is seen as a risk for the security of the nation. The tariffs are a reflection of a way to protect the US workforce from the cheap labor, and there are considerations that productivity didn’t increase in those countries that pushed for cheap labor in the past.

As the presentation progresses, the cheap labor is seen as a problem, also with illegal immigration, which is accelerating it. The way to operate effectively for enterprises that want to innovate, avoiding cheap labor, is through automation as he describes. The formula to bring back innovation in the US is making manufacturing come back, using automation instead of cheap labor, and making energy costs lower to make more competitive local production.

Looking at the trend we see from the population pyramids, we see that Europe is in a trend of already shrinking population, as well as China, where the trend is more mitigated in the US.

Combining the WEF analysis by 2030 and the data from McKinsey by 2030, we can see some commonality and a few minor divergences.

Focusing on the main similarities, there is a clear trend speaking about more jobs in the future than today, while the actual jobs are different than the future ones. Looking at the low-wage jobs and the mid and high-wage jobs, there is a clear reduction of the first ones and an increase of the last two. This means that cheap labor in the EU and US, as well as locally, will not grow but rather reduce, and there will be a clear need for upskilling people to be able to take jobs with higher complexity. Basically, low-wage jobs that are quite manual and repetitive are in focus to be automated, especially if they are those jobs that today are done by outside as cheap labor.

Looking at the McKinsey report is also clear that Europe could increase 2-3% of productivity each year if they would have a more consistent technology and digitalization adoption, where the US has already embraced that more systematically in the last 30 years.

Literally McKinsey says “European companies lag behind US peers on multiple key metrics, such as return on invested capital, revenue growth, capital expenditure, and R&D. Initial delays in Europe in technology development and adoption help explain this gap, as Europe did not benefit from the information communications and technology–driven productivity advancements that have occurred in the United States since the 1990s”.

My Thoughts: From my angle, it is quite visible that US businesses and now also the Middle East are more advanced in incorporating technology-driven functions in key positions in organizations, where Europe is often still seeing some of these as support functions unless related to the end products, and even in those cases, with strong limitations. The acceleration of automation and the next progress on the autonomous agents and AI is simply making this gap even bigger from my perspective, as enterprises in Europe are stalling on the old organizational and operating models, posing also on technology investments, with an attention not systematically strategic. However, what I see as unrealistic in the US administration’s tariffs approach is the massive automation in the real short term that would be needed to achieve that type of production back in the US and accelerated by the tariffs strategy. Here is where I see a risk to simple build a wave of missed jobs coverage, while autonomous operations are not yet mature enough to scale up in a real short time.

Looking to the actual US trends, the illegal immigration restriction and the production back, will accelerate the need for resources on the US borders, paying taxes in the US and producing for the US. Having today already a gap of more than 2M jobs not covered, this means that there will be an even higher need for jobs covered in the US in the future. The rise of low jobs to medium jobs instead of automation will not be an option because it would dramatically ramp up the cost of producing goods. The gap is going to be covered by automation, as also written clearly in all the reports. It’s about 27-30% of activities that are going to be automated, mainly the most repetitive per industry or the ones more commonly across different industries (i.e., office admin). It’s at the end, the productivity gain to make more with less, but needs to be achieved in a structured way.

In the WEF report, looking at skills and job less needed in the future, manual jobs remain only on those cases that are really specific and vertical, not justifying automation due to the specialization, but those are normally not cheap labor.

My Thoughts: The trend we saw recently accelerating on humanoid robots and the acceleration on Industry 5.0, with collaboration between robots and people, with robots becoming more autonomous, seems to be the formula to match the need to have more resources to produce locally in the US autonomously without impacting productivity and using cheap labor.

This could be a formula to allow to grow still in an economy that would need to get many more jobs covered as they would move production back. However, there are considerations to be made on the fact that reducing low wages, basically automating, and pushing for higher wage jobs require accelerating upskilling with all the restrictions and limits of those. The reskilling (so changing completely roles and competences) could be even more complex in that sense. McKinsey speaks about 60% of upskill where WEF, in some cases up to 80%.

Seeking the Industries, as I mentioned in the last few months, taking the example from the WEF, some businesses with a high level of intrinsic digitalization (like banks, insurance) are set to reach a high level of automation (only AI) and a small amount of manual (People) or augmented (People with AI) jobs. Jobs in chemical and oil and gas are set to reach a high level of automation, removing even augmented roles (so either some manual or automated). Jobs in healthcare, in government, in advanced manufacturing, and energy result in the lowest rate of automation (still considerable) and keep the highest level of either manual (with complex, specialized tasks) or augmented roles (people with AI). In the middle, there are many industries that are clearly automating, and the more that are on low margin in the value chain, the more push for automation. In this sense, we see transportation where digital supply chain gets key, as well as mining, automotive, tech companies, good manufacturing, and agriculture. For details, look for the WEF and for my former newsletters.

What is quite impressive is the prediction on the timing of this change. As all the reports mention, it’s about the next 5 years until 2030 to achieve this level of automation and job upskilling for a considerable number of jobs (170 million jobs projected to be created and 92 million jobs to be displaced). Some jobs are set to reduce considerable like office support and manual jobs and agriculture is one of those with more disruption in the way to work with many jobs created but also displaced, meaning a change of generation and way to work. In McKinsey, the sales rep roles get much more reduced versus the WEF report (which sees displacements but also bigger creation as a change in the way of working), but still means a considerable change in some of those roles that are more repetitive and can be automated.

McKinsey writes rightly from my point of view, “Occupations with lower wages are likely to see reductions in demand, and workers will need to acquire new skills to transition to better-paying work. If that doesn’t happen, there is a risk of a more polarized labor market, with more higher-wage jobs than workers and too many workers for existing lower-wage jobs.”

The overall acceleration of AI is seen as a way to match the gap of the needed automation to compensate for the growing demands. From this comes the demand for a low-complex and regulated AI to quickly develop further.

My Thoughts: An AI Governance should exist and be consistent, not blocking but not missing to be fast. Copyrights on data training is one element but I’m more focused on the behavior of autonomous engines and the difficulty to deduct the origin of behavior. We speak and support ideas of AI agents for autonomous decisions, but the monitoring of the process from automation to autonomous needs to be properly industrialized. We took years to raise RPA, so robotics process automation that is indeed quite easy automation on a predictable process, eventually learned. Autonomous is bringing a level of autonomy in adjusting the automation processes and can be powerful also in the progress of deep and reinforcement learning, but also requires control at least until a certain maturity is in place. If not properly governed, wrong behavior can lead to a stop and loss of trust in the overall technology. At the end, an AI is a stochastic engine, it’s not empathic, it can be probabilistically reasonable, and we need to learn to use them properly as we do, for example, when we make decisions based on statistics.