FROM THE DESK OF STEVEN FELDMAN
OPEN POSITION
Let me start with something that sounds obvious but has enormous consequences once you follow it to its logical end.
The S&P 500 is not the economy. It is not even the stock market. It is a market-weighted index of the 500 best-managed, most profitable, most adaptable companies in America - a collection that is continuously curated, endlessly self-renewing, and structurally designed to represent winners rather than reality. When you buy the S&P 500, you are not buying the America economy. You are buying the cream of American companies, skimmed and concentrated into a single instrument.
That distinction has mattered enormously for investors over the last two decades, and it explains much of what I wrote about in last month's letter. But today I want to follow that distinction somewhere most investment commentary doesn't go - to a conclusion I haven't heard articulated quite this way before, and that I think deserves a name.
I'm calling it the Terminal Paradox.
| The Terminal Paradox: the greatest companies in history are on the verge of
| winning so completely that winning destroys the prize.
The Companies That Defined My Lifetime
Think about the major technological innovations of the last fifty years, and the companies that embodied them. Software gave us Microsoft. Communications gave us Apple and - more recently, more radically - SpaceX. The internet gave us Google and Amazon. Social media gave us Facebook, now Meta. Robotics gave us Tesla and Amazon again. Clean energy, in its current form, lives largely inside Tesla. Healthcare's most transformative recent chapter is being written by Eli Lilly.
I could have included defense. But nuclear weapons were invented in the 1940s, and everything that has come since, however lethal, is incrementally less consequential than that. So I'll leave war innovations off the list.
What's striking about this list is not just the companies - it's the pattern. Each innovation created a new category. Each new category created a new industry. Each new industry created new employment, new consumption, new wealth distributed across a broad enough base to sustain the next cycle. The textile worker became the factory worker. The factory worker became the office worker. The office worker became the knowledge worker. Every wave of creative destruction turned out to be creative enough to absorb its own destruction.
That pattern is about to break. And the breaking point has a name: artificial intelligence.
Why AI Is Different From Everything That Came Before
Along comes AI, and you add Nvidia, Broadcom, AMD, and - when they eventually list - OpenAI and Anthropic. More importantly, virtually every company on the list above is already all-in on AI infrastructure. Microsoft. Google. Amazon. Meta. Tesla. Apple. These are not RCA, the radio giant that couldn't fully adapt when television arrived. These are the companies building the television, the broadcast tower, the cable network, and the satellite all at once.
Just the public companies listed above represent a third of the S&P 500's total market capitalization. That concentration is historically unprecedented. And it is about to get more concentrated, not less.
Here is the claim I want to make carefully, because it is the one that everything else depends on: AI is not just another innovation. It is the last one, as it will eventually have all knowledge and computing power. In short, it is the innovation that captures all future innovations.
Think about what comes next. An energy breakthrough? It will be discovered, optimized, and deployed by AI. A healthcare revolution - beyond what Eli Lilly is already doing with GLP-1s? AI will find the molecules, run the trials, and personalize the treatment. Quantum computing? Its primary application will be to make AI faster. Every frontier that remains - materials science, climate technology, drug discovery, financial modeling - will be explored primarily by AI systems, and the economic value of those explorations will accrue primarily to the companies that own the AI infrastructure.
This is the subtle but crucial distinction. It is not that innovation ends. It is that innovation gets privatized. The upside of every future discovery flows to a closed club of companies that already exist. New entrants cannot compete, because the capital requirements for AI infrastructure - the chips, the data centers, the models, the talent - are beyond the reach of any startup that isn't already a giant. The frontier is open. The tools to explore it are not.
| It is not the end of innovation. It is the privatisation of innovation's upside into a
| club that is no longer accepting new members.
Then comes the next chapter. Companies realize they can do more with more - not more employees, but more AI-enabled capacity. They expand into adjacent markets, take share from less capitalized competitors who couldn't install and master AI quickly enough, and absorb the customers those competitors leave behind. The Radio Shack and Borders Books of the AI era - smaller, slower, less capitalized businesses across every sector - go the way of every incumbent that couldn't adapt to the previous internet wave, except faster and more completely.
Employment in those businesses collapses. The jobs don't move to a new industry. There is no new industry. There is only the AI infrastructure layer and the AI-enabled layer, and neither needs as many humans as the economy they are replacing. People either work for an AI infrastructure company, or for a company that has mastered AI - and even those positions are fewer than the ones that disappeared.
Here is where the paradox bites. Employees have another name. They are called consumers. And consumers who are unemployed, underemployed, or simply frightened about their economic future do not buy things with the enthusiasm that American corporate profit models require.
| Employees have another name. They are called consumers. And consumers who
| have been automated out of the economy are not in the mood to buy a lot of stuff.
Corporate profits surge in the short run as labor costs collapse. But the very workers whose costs were eliminated were also the customers whose spending sustained demand. Henry Ford understood this a century ago - you have to pay workers enough to buy the cars. The AI era's productivity miracle is running the Ford logic in reverse: eliminate the workers, eliminate the costs, eliminate the customers, eliminate the demand.
Eventually, AI reaches a ceiling. Not a technological ceiling - the models will keep improving. A market ceiling. The number of customers is finite. The products have been sold. The productivity gains have been captured. And the companies that own everything find themselves in a peculiar position: enormously profitable on paper, but operating in an economy that can no longer afford what they're selling at the scale required to justify their valuations.
This is the point where I want to be precise, because it's easy to confuse the Terminal Paradox with simpler arguments that have been made before.
This is not just the automation-kills-jobs argument. That argument has been made since the Luddites, and it has always been answered by the same historical fact: new technologies create new industries that re-employ the displaced. The weavers became factory workers. The factory workers became service workers. The response has always worked - because each new technology opened a new frontier that required human exploration.
AI closes that escape hatch. When the technology that is automating your job is also the technology automating the creation of your replacement job, the historical pattern breaks. There is no next frontier that requires a mass of human workers to explore it, because AI is already there.
And this is not just the wealth inequality argument either, though inequality is part of it. The deeper problem is structural. Markets work because they aggregate information through competition, price discovery, and the constant pressure of new entrants challenging incumbents. That mechanism - which has driven human prosperity for three centuries - requires a certain distribution of economic power to function. When one generation of companies owns the means of all future production, with no technological disruption capable of dislodging them, you no longer have a market. You have a permanent aristocracy, denominated in computer power.
The S&P 500 doesn't just become too big to fail. It becomes too big to need anyone.
| The S&P 500 doesn't just become too big to fail. It becomes too big to need
| anyone. And that is a different kind of problem entirely.
WHAT IT MEANS FOR MONEY
I recognize this is a dark picture. I want to be clear: I am describing a trajectory, not a certainty, and the timeline is measured in decades, not quarters. The short and medium run for AI infrastructure companies - and for the S&P 500 more broadly - remains constructive. The productivity gains are real. The earnings growth is real. The concentration of innovation in existing giants is real and self-reinforcing.
But I am also a believer in following logic to its conclusion, even when the conclusion is uncomfortable. And the conclusion here is this: if the Terminal Paradox plays out - if AI does concentrate economic power into a closed club, if employment does collapse faster than new categories emerge, if consumers do run out of spending capacity - the pressure on governments will be extraordinary.
Governments will not sit still. There will be retraining programs, transfer payments, some version of universal basic income. The specifics will vary, but none of it solves the problem at the scale the problem requires. The fiscal math is the same everywhere, and that is where the monetization argument begins.
Tax revenues collapse as employment falls. Social costs explode as the displaced require support. The fiscal math becomes impossible by conventional means. And the response, in every historical instance where governments have faced impossible fiscal math, is the same: monetize the debt. Print the money. Inflate the problem away.
The companies that own AI may be worth everything in nominal terms and very little in real terms, if the currency they are denominated in gets quietly destroyed in the process of managing the transition. Corporate profits that are measured in dollars that are worth half what they were are not the windfall they appear to be.
Gold doesn't have that problem. Physical scarcity cannot be automated. It cannot be inflated. It cannot be captured by the AI infrastructure layer or privatized by the companies building the models. It sits outside the Terminal Paradox entirely - not because the paradox doesn't unfold, but because gold's value is precisely its independence from the systems that the paradox threatens.
Real assets more broadly share this characteristic. Copper, uranium, agricultural land, water - the physical world that underpins the digital one. When intelligence gets cheap, what remains genuinely scarce becomes disproportionately valuable. The paradox of the AI era is that it strengthens, rather than weakens, the case for tangible assets at exactly the moment when most investors are most heavily concentrated in the digital ones.
| Gold's value is precisely its independence from the systems that the paradox
| threatens. Physical scarcity cannot be automated, inflated, or privatized by
| the companies building the models.
Imagine it is 2045. The ten largest companies in the world - all AI infrastructure giants, all descendants of today's Magnificent Seven - have combined revenues larger than the GDP of most nations. Their productivity per employee is unimaginable by today's standards. Their profit margins make today's tech margins look modest.
And they have a problem. Not a regulatory problem, not a technology problem. A customer problem. The middle class that used to buy their products, subscribe to their services, click on their ads, and fill their logistics networks has been hollowed out. Not dramatically, not all at once - gradually, over two decades, in the slow erosion way that only looks obvious in hindsight.
What do those companies do? What does the government do? What does money mean in that world?
I don't know the answers. No one does - this is genuinely uncharted territory, the kind that only science fiction writers have previously attempted to map. Some of what AI enables will be extraordinary beyond imagination. Some of it - the AI-enabled facial recognition drone, the algorithmic social control, the elimination of privacy as a concept - will not be
What I do know is that the investors who will navigate that transition most successfully are not the ones who bet everything on the nominal winners. They are the ones who understood, early enough to act, that the Terminal Paradox was coming - and built portfolios anchored to things that exist outside it.
The S&P 500 is the greatest collection of companies ever assembled. I mean that sincerely. But even the greatest collection of companies ever assembled needs someone to sell to.
Keep that in mind the next time someone tells you the index always wins.
Steven Feldman is the co-founder & CEO of GBI. This newsletter is for informational purposes only and does not constitute investment advice.