By Stephen Leeb, Ph.D., and Megan Davis
Despite massive evidence to the contrary, investors seem more willing than ever to bet that AI in the U.S. has limitless potential. A slew of incoming Initial Public Offerings (IPOs) involving AI are about to land, the most fantastic of which is SpaceX, a company completely controlled and captained by Elon Musk. The IPO for SpaceX is set to hit on June 12, and the market is expected to value the company at the stunning sum of around $1.75 trillion.
When you think of SpaceX, you are likely to imagine rocket ships designed for interplanetary travel or spacecraft designed to carry us to the moon and beyond. The more sensible and level-headed among us is likely to think primarily of Starlink, their constellation of satellites that provide internet across the globe.
But shockingly enough, if you read the prospectus that SpaceX filed with the U.S. Securities and Exchange Commission (SEC), you find that SpaceX is marketing itself as an AI company. A prospectus is a formal, usually very boring, document that companies are legally required to present to provide details on their investment offerings to the public.
In looking over the SpaceX prospectus, it is not hard to understand why Starlink is the only subsidiary making money. The other subsidiaries—AI and the space businesses—have combined losses to date totaling $37 billion. But even Starlink shows per-user revenues declining, so it does not take a financial genius to assume that Starlink alone would have a tough, if not impossible, time making up for that $37 billion loss over the foreseeable future. (Patrick Boyle broke the situation down hilariously in his video “Nice Try Though.” We especially enjoyed the comment about the missing money on the accounting sheets falling off the back of a Cybertruck.)
How, then, can SpaceX possibly be poised to garner the highest-ever valuation for an IPO?
There is no answer that makes sense. Unless, of course, you believe, as the company claims in its prospectus, that SpaceX really is an AI company whose potential market is close to $30 trillion—higher than the GDP of the entire U.S.—with AI accounting for 90%. That’s an incredible claim to swallow.
But it gets even more incredible. The company’s only current AI product is Grok, a large language model (LLM) that by almost all measures ranks at the bottom. Yet the valuation of SpaceX will be 50% higher than the combined value of all IPOs issued in 2021, which, until 2026, was the best year for IPOs ever recorded. Moreover, once you count forthcoming IPOs in other AI companies like Anthropic and OpenAI, it is a good bet that total IPO valuations this year will exceed the total of the past 45 years combined.
Another way to put it is that AI currently accounts for about 45% of the market’s total capitalization. The new AI IPOs will raise the share to about 50%. This will result in a market already setting records for its dependence on one group of stocks becoming a market in which that one group is really the only market driver—and its foremost danger.
The so-called “Buffett indicator” is a valuation metric that compares the total value of all publicly traded stocks in a country to the size of that country’s actual economy, or gross domestic product (GDP). The current AI group alone would be enough to turn the indicator red. Even without these extra AI valuations, our stock market is more than twice our GDP, which means that for a lot of Americans, their retirement security is riding on the success of AI.
We have never been in a situation like this before, in which one industry is equivalent to the country’s entire GDP. It’s the epitome of being too big to fail.
If we can win the AI race, which SpaceX and its prospective investors are counting on, and AI proves to increase productivity by as much as analysts expect, is it a bet worth staking the entire American economy on?
The odds aren’t stacking up well.
First, to compete at the level once common in America, where we would win international competitions without breaking a sweat, we would need the kind of high-level talent that once populated the halls of our universities and research establishments. If you take a cursory glance at the CWTS Leiden Ranking based purely on bibliometric data, eight of the top 10 universities and institutions are Chinese. This matches up to Nature’s list as well. And if you adjust the list to reflect only mathematics and computer sciences, the top 43 out of 50 are in China. You have to scroll all the way down to No. 42 to find the first American institution, the Georgia Institute of Technology.
The fact that America has not been investing in the education systems that will decide our future is an entire blog in and of itself. But suffice it to say, we were going to have to face the consequences of those actions one day. And that day may have arrived.
In terms of China’s own technology companies, Huawei, at one point, was their shining light. In 2019, Huawei was placed on the U.S. Entity List, which cut it off from critical U.S. technologies. The following year, sanctions were expanded to block semiconductors. This kneecapping sent the company into survival mode almost overnight, and many analysts assumed Huawei had no way forward. But instead of causing the company to fold, the sanctions actually pushed them to lean more into their own culture and expand their own operating system and chip manufacturing—even currently stepping into the AI gap left by U.S. restrictions on companies like NVIDIA.
The sanctions denied China access to ASML’s extreme ultraviolet (EUV) lithography machines, which are the only machines capable of shrinking transistors below 14 nanometers (nm). Still, in 2023 Huawei astonished Western regulators by introducing a new phone that used 7nm chips. They achieved this by using a 14nm lithography machine in conjunction with an elaborate workaround termed multi-patterning. Imagine that you need to draw 100 vertical lines on a piece of paper, but you only have a very fat crayon. So, you use a stencil and do every other line, shift that stencil slightly and fill in the gaps. This would permit you to pack the lines twice as close together as the size of the crayon would allow. But multi-patterning was a stopgap. And so, the West breathed a sigh of relief, knowing 3nm and even smaller chip sizes were on the way, which would allow the West to retain their advanced lead over China.
What the West did not know was that almost as soon as the 2019 sanctions were introduced, the Chinese began working on a longer-term plan that would provide a path to greater speed across the entire technology spectrum, ranging from transistors to entire data centers. This would culminate in the recent announcement of their new scaling law—Tau’s (τ) law. For the last 61 years, all semiconductor companies have been using Moore’s law as a guiding principle. It focused on space and size—making transistors smaller and smaller so as to achieve better performance. Therefore, improvement in chips was always about hitting a specific (smaller and smaller) nanometer count. But the new Tau’s law focuses on time and speed—minimizing delay across an entire computing stack. Instead of reducing the physical size of transistors, which they couldn’t do without ASML’s machines, they took a step back and tried to address the issue that smaller transistors fixed in the first place—delay. Essentially, how to increase the speed of information through the system overall without smaller transistors.
Tau’s law applied to chips was an ultra-sophisticated way of stacking complex circuits such as CPUs (and GPUs) on top of one another. But rather than a stacking that would resemble Legos, it is actually as if you built two entire Lego rooms, and rather than making the rooms smaller to decrease distance, you folded the two onto themselves. You don’t have to walk all the way down the hall if the room is right on top of you. It’s very complex as we are dealing with quantum effects and such tiny, tiny areas that have to be extremely precise.
It will not be easy to duplicate. The delicate precision required for the hybrid bonding requires a bonding accuracy measured in nanometers. If we go back to our room analogy, you need a ladder to get from the bottom of the room to the top, and if the two sides of that folded ladder are misaligned by even a microscopic hair, the entire chip fails.
This technique was dubbed logic folding and will be on full display when Huawei introduces its newest smartphone in September. If the online specifications and early tests are correct, Huawei’s phone will have speeds equivalent to 3nm chips and mark an even greater jump than occurred when they went from 14nm to 7nm chips.
They also announced specific scaling goals. In terms of complex logic chips that are critical in defining and establishing chip speed, scaling is expected to be 1.3 per year, which almost exactly matches the reduction from 7nm to 3nm (i.e., 1.3 ^3 = 2.2). Though many commentators have questioned the 1.4 nm goal for 2031, the current scaling of 1.3 points to a chip with an equivalent size of less than 1 nm. This would very likely top Western goals of making less than 1 nm at scale by the mid 2030s.
But the even bigger implications are in Chinese AI. By applying Tau scaling to their entire AI infrastructure, it potentially creates far more efficient and faster AI architecture on a systemic level. This will lead to massive leaps in AI compute. Tingbo He, who has been captaining Huawei’s chip development for over two decades, said in her paper that introduced Tau scaling: “Along this path, hardware integration is projected to increase by more than 100× by 2035, with τ reduction distributed across every layer of the stack rather than concentrated at the device level.” In this sentence buried in the paper, the Chinese are pointing to astonishing gains across the entire AI infrastructure.
Admittedly, China’s goals are extraordinarily ambitious, and they could fall short of their aspirations. But don’t forget that at the beginning of this century not a single Chinese school or institution was among the Top 10 in the world. Today, they dominate in fields like quantum mechanics that have become critical for the AI arena.
What does this mean for AI here in the U.S.? Well, Americans can apply this new scaling to our own systems, but we are undeniably behind. And the U.S. is still very focused on the questionably achievable goal of artificial general intelligence (AGI) rather than on improving AI as an aid to human creativity.
If America’s entire stock market is hinging on the success of American AI, it’s an incredibly scary spot to be in. One that could lead to a massive collapse. One that may even cause us to ask – what happens when “too big to fail” fails?
Instead, we would rather ask a different question: What if we choose to work together?In 2019, Ren Zhengfei, the founder of Huawei said, “The message to the U.S. that I want to communicate is collaboration and shared success. In our high-tech world, it is increasingly impossible for any single company or even any single country to do the whole thing.” Guo Jiakun, China’s spokesperson for the Ministry of Foreign Affairs, reiterated that message when Trump visited China last month, telling the media that the U.S. and China “should work together to promote the development and governance of AI and to ensure that AI better serves the progress of human civilization.” We think it’s a message worth heeding.
Take Your Investment Strategy to the Next Level
Affordable. Essential. By joining Turbulent Times Investor, you’ll gain full access to our Growth & Income Portfolio recommendations.
Most investors have yet to grasp the extent to which the world is changing and the profound impact it will have on financial markets. The global stock markets are rapidly approaching an era of unprecedented turbulence. Investors face enormous risks—but also some great opportunities, which we highlight and monitor in Turbulent Times Investor! Don’t miss out—join now to stay in the loop.
