AI, tariffs and a pause (Weekly summary 01/02)

There have been quite a few interesting events to analyze this week, and below I list the most noteworthy news. Let’s get started:

• Trump pressures again with his tariff strategy, this time threatening a 25% levy on oil from Canada and Mexico, effective today. While no final decision has been made, he has made it clear that it will depend on how these countries “treat the U.S.” and on oil price developments. His official argument cites the fentanyl crisis and the lack of control his neighboring countries exert over the border.

  The political background is evident: Trump is using tariffs not only as a trade tool but also as leverage on immigration and drug-related issues. Additionally, there is the possibility of new tariffs on China, accusing it of facilitating the trafficking of this drug, which adds a new front of tension to the trade war.

  The implications could be significant. If the tariffs are imposed, they would directly affect the flow of Canadian crude into the U.S., potentially increasing costs for American refineries (which, notably, are designed to process heavy crude, very different from domestic production) and, paradoxically, raising fuel prices in his first presidential year. In Mexico’s case, the impact would depend on its ability to redirect exports and its negotiation margin with Washington.

  So far, neither Canada nor Mexico has issued an official response, but any retaliation could escalate tensions within the USMCA and destabilize the region. It remains to be seen whether this is merely a pressure tactic or if Trump is truly willing to sacrifice energy stability for short-term political goals.

• The Federal Reserve kept interest rates unchanged, and Jerome Powell made it clear there is no rush to cut them until inflation and employment data justify it. With a stable yet uncertain macroeconomic outlook due to potential policy shifts under a Trump administration, the Fed remains in wait-and-see mode to assess the impact of possible changes in tariffs, taxes, and immigration policies. Powell emphasized that unemployment has been stable for six months and that inflation has shown signs of moderation, though it remains above the 2% target.

  

   

  Since his return to power, Trump has intensified his rhetoric against the Fed, blaming it for inflation and accusing it of focusing on diversity and climate change policies rather than its economic mandate. While Powell avoided responding to the president’s comments, he reiterated that monetary policy is based solely on economic data.

  The FOMC statement reaffirmed that economic activity continues to expand at a solid pace and that the labor market remains strong, though inflation “remains elevated.” Interest rate futures indicate that the market does not expect cuts until June.

• X (formerly Twitter) has taken another step in its ambition to become a super app by partnering with Visa to offer a real-time payment system. This service, called “X Money Account,” will allow U.S. users to make peer-to-peer payments, link their debit cards, and transfer funds to their bank accounts through Visa Direct. While there is no official launch date yet, X’s CEO, Linda Yaccarino, has hinted that this is just the first of many new features in the platform’s payment ecosystem this year.

  Elon Musk has spent decades trying to create an integrated services platform. In the 90s, he launched X.com, which later merged into what is now PayPal. Since then, he has maintained the vision of developing an app similar to China’s WeChat, combining messaging, payments, and entertainment within a single ecosystem. However, the current landscape is far more competitive: Meta has already added shopping, gaming, and dating features to its platforms, while Apple has been accused of blocking the development of super apps on the iPhone to maintain control over its services.

  Since acquiring Twitter in 2022, X has experienced an exodus of advertisers and users due to the rise in controversial content on the platform. Now, with its entry into digital payments, Musk aims to give the business a new direction, but he will have to contend with competition from tech giants and potential regulatory hurdles. The big question is whether X can gain traction in this space or if, as with PayPal in the past, Musk will once again run into insurmountable barriers. We certainly like the new direction and think it has the potential of becoming a true superapp.

• We have seen the largest single-session stock market drop in history, with $NVDA, up to that point the largest company in the world, losing nearly $400B in market cap in a single day. DeepSeek, a Chinese AI company (spun off from a hedge fund) has released an open source LLM model, priced well below the competition and performing at a world-class level (it was, in the main comparative rankings, on par with ChatGPT). Moreover, they have managed to train this model for ~$6M, vs. the $500M that a development of this level can cost in the West. DeepSeek has not only achieved technical parity with OpenAI in terms of reasoning, but has done so with astonishing computational efficiency, challenging the traditional model of training with high-end hardware. Its focus has been on inference (the use of the model to find answers) rather than training, adding a particularly novel chain of reasoning.

  The key breakthroughs of DeepSeek-MoE optimized, MLA for memory reduction, and extreme optimizations on H800s-demonstrate that the U.S.-imposed chip shortage has not stopped development in China (that is if we are to believe that they have not used top-tier chips, of course, which has been questioned by many authoritative voices in the industry, including Elon Musk himself), but has incentivized more aggressive innovation. This point is critical: the idea that technology sanctions will slow China down seems more based on political wishful thinking than technical realities.

  From an investment perspective, Nvidia is the most exposed. DeepSeek’s optimization suggests that hardware spending could slow significantly, affecting demand for high-end GPUs. However, inference remains a growth market, and the question is whether the company will be able to pivot in this new scenario. In the meantime, Meta and Apple are obvious winners: the former because of its bet on open models, the latter because inference in devices makes even more sense with lower memory requirements. The biggest losers really are the companies that develop the models (e.g OpenAI), which have seen their monopoly assaulted and, once China has entered the market, will eventually see their margins reduced to 0.

  It has become fashionable in most analyses of these hypotheses to use the Jevons Paradox, which we discussed in detail a year ago as applied to oil consumption. I agree that his thesis is the most likely path that the GPU and data center market will follow: the paradox stipulates that the more efficient an industry becomes in the consumption of a good (e.g., engines in gasoline consumption), the more the demand for the underlying good increases, since the increase in demand caused by its affordability more than offsets the reduction in consumption due to greater efficiency. The same could happen with GPUs: the fact that models can now be trained more efficiently only encourages greater use and proliferation, both by established players and new entrants, as the monetary barriers to entry for acquiring the hardware have been greatly reduced.

  

   

  What is clear is that AI development has accelerated beyond the regulatory and commercial limits that the US was trying to impose. If the U.S. response is more chip protectionism or attempts to slow down the opening up of models, it will only succeed in weakening its own innovative ecosystem. The alternative is to accept that competition is here to stay and that the only viable response is to keep innovating faster than others. The biggest opportunity I identified with these downturns is related to nuclear power: the market logic (flawed, in my opinion), was that, with more efficient models, much less electricity consumption would be needed, and therefore the forecast increase in nuclear power would be affected; again, I encourage you to read Jevons.

• Oil inventories in the US (and also globally) continue to fall (-16Mb in January, that is, 0.5Mb/d), and it is increasingly difficult to justify forecasts of an oil surplus in Q125. Some banks, such as BoFA are starting to pivot, hiding behind the imposition of sanctions to justify their erroneous forecast, and no longer foresee a surplus in 2025, while others, such as JP Morgan continue to insist on their idea, leading them to an increasingly ridiculous position.

  

   

  Pressure and criticism are already beginning to mount against the IEA, which has become a propagandistic body and whose predictions, once a benchmark, are becoming increasingly fanciful. In this regard, articles have begun to appear asking the agency to abandon its political agenda of energy transition and return to its original function, providing quality and useful information.

  At these underlying prices, we have commented several times how there is no incentive for increased production, and we have had confirmation from several US oilfield service companies this week. Below is an excerpt from one of their conferences, where they state that they expect most companies to focus on maintaining production, and that exploration and development will remain flat with respect to 2024.

  

   

  The majors are not even able to cover their current shareholder return programs through the cash flow they generate.

  

   

  Higher.