Warning: Last week was an unusually chaotic one, first related to a flurry of pronouncements from the White House and, second, from news that DeepSeek, a Chinese firm appeared to build AI models that rivaled the performance of the American leaders at a fraction of the cost. There’s a lot to cover. Thus, this note may be longer than usual.
Let’s start with Trump, his Executive Orders and other directives that followed the spirit of his campaign promises with the elevated level of chaos always associated with the President. Stepping back first, Trump’s overriding campaign messages were to shrink government, reduce red tape, secure our borders, and punish via tariffs those nations who Trump feels are taking advantage of America. To accomplish his objectives, he chooses to run roughshod over Congress and the courts. Already he has gotten some pushback. A poorly designed OMB directive to freeze Federal funding for a time while the administration and DOGE determine what should or shouldn’t be spent, brought immediate lawsuits and a judicial freeze. The directive was quickly rescinded. These are just a few early moves in a chess match that will play out over the coming months.
By late last week, Treasury had given Musk and DOGE access to the Federal payments system supposedly on a read only basis. Coincidentally, David Lebryk, the administrative head of the Bureau of Fiscal Service, retired suddenly Friday coincident with the announcement. Again, this is a chess move within a match that will take months to play out. While Democrats and the media predictably screamed, courts once again are likely to weigh in. From Wall Street’s perspective, however, any effort that succeeds in reducing the fiscal bureaucracy of the United States would be a net positive. Elon Musk has declared that DOGE efforts, including forcing people back to the office and early retirement offers will save $1 trillion on an annualized basis. Musk is prone to hyperbole and, thus, his proclamations have to be taken with some cynicism.
Currently, the Federal deficit is about 7% of GDP. That is an obscene and unsustainable rate. 3% would be sustainable. Government expenditures now total a bit under $7 trillion. It takes in a bit under $5 trillion. Total GDP is a bit under $29 trillion. A $2 trillion deficit is 7% of GDP. $870 billion would be 3%. Also note that as the debt burden grows (current accumulated borrowing plus each year’s deficits), debt service costs will rise. Add in Social Security, whose payments each year include a cost-of-living adjustment, and Medicare, and you see that savings of well over $1 trillion are needed to bring deficits down under $1 trillion per year. It is also a fact that every cut proposed will bring pushback from the people and programs impacted. Early shock-and-awe attempts to move quickly are likely to give way to more strategic attempts to cut with a razor blade rather than a sledge hammer. Musk promised DOGE will be done by July 4, 2026, our nation’s 250th anniversary. The bottom line is directionally, what is being attempted is correct. But not all the early efforts will prove to be legally compliant. This is going to be a two-step forward, one-step back battle. But it’s the right battle.
That brings me to the tariffs. The headlines say 25% tariffs on all non-energy imports from Canada and Mexico plus 10% tariffs on Canadian energy and 10% tariffs on everything Chinese. I would expect the final tariffs will see some carve outs upon appeals and could be reduced if Mexico and Canada take steps to Trump’s liking. Conversely, should they be implemented in a robust fashion, look for trade wars of some kind to ensue. Canada and Mexico export more oil and gas to the U.S. than the rest of the world combined. Midwestern refineries are configured to run on heavy Canadian crude. Auto parts and subassemblies move back and forth across borders and, therefore, could be subject to multiple tariffs. Given the support the UAW gave Trump during his campaign, I would expect some moderation soon.
Don’t forget that Trump’s political capital was at its maximum at the moment of inauguration. From that moment on, he can use it or fritter it away. Blaming a plane crash, for instance, on DEI regulations doesn’t improve his standing with the public. It may be a little thing overall, but all such sidebar comments lead to a nick on the popularity scale. One shouldn’t ignore the fact that Trump’s approval rating during his first term was never above 50%. Clearly, the tariffs themselves represent a significant expenditure of political capital. Polls show they are unpopular on both the left and right.
As an investor, and only looking at the numbers, the key question is whether the moves to date and the prospective moves to come increase or decrease growth, and whether they increase or decrease the pace of inflation. For all the chaos last week, the yield on 10-year Treasuries fell by 4 basis points. Besides Trump’s actions, the Fed kept short-term rates steady (absolutely no surprise), and inflation data was exactly in line with forecasts. Bottom line: Despite all the chaos, Wall Street took Trump’s actions in stride. Futures are lower this morning, reflecting modest changes downward in growth assumptions and upwards for inflation, although bond markets remain stable.
That was hardly the case however, with news that DeepSeek, a Chinese AI firm, has built large language models that could compete with the likes of OpenAI, Meta Platforms#, Google# and others at lower cost using Nvidia# chips that were less powerful than leading edge semiconductors used by its American counterparts. Some called this a Sputnik moment. Nvidia’s market cap fell by more in one day than any company had ever experienced. In a fast-moving tech world where the disruptor can quickly become the disruptee, were investors right to run for the hills or was what we learned last week a wiggle in the emerging world of artificial intelligence?
First, let’s recognize that what DeepSeek did was ingenious in many ways. Clearly, it has built a series of competitive models that, today, put it close to leading edge. It probably isn’t there yet, but it is getting there. It is also clear that while DeepSeek is privately funded, China is pouring massive amounts of money into efforts to become leading edge in the AI race. Looking ahead, it would be foolish to assume that U.S. ingenuity will leave China and others in the dust for very long. It would also be foolish for the U.S. to assume that sanctions on advanced semiconductor chips made by U.S. companies are going to prevent China from being competitive.
Does that mean that the hundreds of billions of capital investment American companies are making to support large language models (LLM) are being wasted? This question doesn’t lead to a yes or no answer. The response is more nuanced. What businesses learn racing down the AI avenue will modify their approaches in the future. We are in the very early innings of artificial intelligence. Right now, most uses are fairly simple. At least they will seem simple compared with what’s coming in just a few years. Answering a sophisticated query almost instantly today is nice. So is using AI for the building blocks of software code. Soon AI will obsolete call centers but not until the automated responses are better than the human ones, admittedly a low bar is some cases. Over the next few years, robotics will capture our fancy whether it be Waymo taxi rides, or human-like robots performing ever more sophisticated tasks.
In 1865, an English economist named William Stanley Jevons found evidence supporting the notion that improving efficiency, even at a lower price point, would generate increased activity that would more than compensate. Jevons paradox helps to explain Moore’s law. It doesn’t just apply to technology or semiconductors. Thus, if Nvidia and others can introduce new chips that give 10x the performance for 3x the price, demand will evolve to that advantage of that gain in efficiency. But with that said, Wall Steet is all about the ability to meet or exceed expectations. Might there be a moderation in Nvidia’s growth rate, for instance, as customers adjust their cost-benefit analysis based on what they might have learned from DeepSeek’s experience? At the same time, the race to the finish line continues at a rapid pace. The big hyperscalers know that building ever more effective and accurate models won’t slow down. But what may happen is that beneath the big hyperscalers, a new class of AI firms will emerge that, rather than being soup-to-nuts for all, focus more on vertical markets where costs can be tightly focused on more limited objectives. A module McDonald’s might use to automate order taking at the takeout window has to be nimble enough to correctly interpret an order and get changed multiple times on the fly. But it doesn’t have to be trained to tell you the weather or the score of the local baseball game.
Did DeepSeek pull the rug out from under the likes of Nvidia and Microsoft#? No. While Wall Street reacted in shock last Monday, all the major players have been aware of the company and many others in China for some time. They may have been somewhat surprised that the performance gap narrowed as quickly as it did. One should also realize that DeepSeek’s model is open sourced, meaning any user can take its code and modify it for it’s specific use. Indeed, it is assumed by sophisticated observers that DeepSeek most likely fed its model building on models developed by OpenAI and others.
In conclusion, the AI race is a rapid one still evolving. The ultimate winners haven’t been determined and won’t be for some time. It wouldn’t be far-fetched to conclude that some of the current Magnificent Seven won’t be so magnificent when the dust settles. There is a big pot of gold at the end of the rainbow. It will be shared not only by some of the obvious tech leaders of today but by some new tech companies as well as the owners of data centers, and companies that provide all the plumbing and services necessary to support AI. Last week may alter the landscape a bit, but won’t change the vision that artificial intelligence is with us to stay, that it will change our lives, that oversight will be needed to contain risks, and that a lot of businesses will become hugely profitable as a result over the coming decades.
Today, Nathan Lane is 69. Fran Tarkenton turns 85.
James M. Meyer, CFA 610-260-2220