Before We Climb Past Each Other
November 7, 2026
Growing Block Universe, 5-MeO-DMT & Nested Observer Windows
November 8, 2026Listen to Free Audio Version Here
By C. Rich
I have a guesstimate about where artificial intelligence is heading, and I suspect most of the people using it today have absolutely no idea what is coming. They are busy celebrating each new model, marveling at what it can accomplish, and complaining whenever their favorite chatbot becomes a little slower or a little less intelligent. What they have failed to recognize is that they may be participating in the greatest transfer of human knowledge and productive capability in history, and when the transfer is complete, the companies receiving all that knowledge may no longer have much economic incentive to share their most powerful intelligence with the public.
We have spent the past several years feeding these machines everything we know. Our books, scientific papers, research, photographs, programming languages, medical knowledge, history, philosophy, and an extraordinary amount of our collective intellectual heritage have become the raw materials of the artificial intelligence revolution. Some of that material was licensed, some was publicly available, and some remains the subject of enormous copyright disputes. We then enthusiastically began interacting with the resulting models, correcting their mistakes, teaching them how we work, and demonstrating what human beings actually want intelligent machines to accomplish. Now imagine that the companies building these systems eventually decide they have acquired enough of our knowledge to develop products and services worth considerably more than the subscriptions we pay to access their machines. What happens when the most profitable business is no longer renting us intelligence but using that intelligence themselves? I believe we are approaching the beginning of that transition, and the first warning arrived just last week.
After a week of celebrating and pushing Grok Bot on the public, asking everyone to come and check it out, Elon did a 360 stunningly. On November 1, Elon Musk acknowledged that the cost of serving Grok Bot had become unsustainably high relative to the revenue it generated. Paying customers were already reporting that their weekly usage allowances had been exhausted unexpectedly, sometimes with days remaining before their limits were supposed to reset. Here was one of the world’s most ambitious artificial intelligence companies, operating at enormous scale, effectively acknowledging that its customers were consuming more computing resources than the company could economically justify providing at existing prices.
Think about the implications. We have been operating under the assumption that artificial intelligence will become more accessible as the technology improves. Each generation of models would become smarter, faster, and more capable, while competition would eventually make this extraordinary intelligence available to virtually everyone. That was the promise of the coming age of abundance, and for a while, it appeared to be exactly where we were heading. But what happens if the intelligence becomes more expensive to operate as its capabilities increase? What happens when an AI agent spends hours researching, reasoning, writing software, conducting experiments, or completing complicated assignments that once required teams of human beings? The financial equation begins to change. The customer paying a monthly subscription might consume far more computing power than that subscription can support. Somewhere inside the company’s accounting department, somebody eventually notices that the customer is receiving a thousand dollars of computational work for a fraction of its cost.
The accountants have a remarkable ability to bring technological revolutions back to Earth. Damn bean counters. My theory is that the Grok Bot incident is an early glimpse of a much larger economic transformation. The public may discover that the computing power it has come to expect is gradually being rationed, restricted, repriced, and ultimately redirected toward customers and internal projects capable of generating substantially greater returns. And I suspect that this transformation will be far more dramatic than most people imagine. Consider the enormous concentration of computing infrastructure among a handful of technology giants. Amazon, Microsoft, Google, Meta, and Oracle are among the companies controlling or building immense computational capacity, while specialized artificial intelligence developers such as OpenAI, Anthropic, and xAI compete for the resources necessary to develop increasingly powerful systems. The ownership structures, partnerships, and business models differ, but the economic reality is becoming unmistakable. The capacity to operate frontier artificial intelligence at its highest levels is concentrating around organizations with extraordinary access to capital, electricity, specialized processors, and physical infrastructure.
We are watching the construction of an entirely new industrial hierarchy, with an extraordinarily small number of companies controlling the machinery upon which much of the next economy will depend. Beneath them sits an expanding class of businesses that rent computing capacity, build services using commercial AI models, or develop applications dependent upon continued access to systems they do not own. Some of these businesses are innovative and commercially successful. Others are essentially intermediaries, renting intelligence from somebody else’s machine and packaging it into a product for their own customers. This is what I call the renter class of the artificial intelligence economy. And I believe a substantial portion of that renter class is going to disappear. These companies are building their businesses on the assumption that the underlying intelligence will remain available, affordable, and sufficiently capable to support their products. But what happens when the owner of that intelligence decides to raise the rent, reduce the available compute, reserve its strongest models for internal use, or release a competing product developed by the same machinery?
The renter has very little leverage. He does not own the building, the machinery, the electrical infrastructure, or the intelligence itself. He owns a business whose continued existence depends upon somebody else’s willingness to keep the lights on. It is difficult to imagine a more precarious position from which to build the next generation of the American economy. The problem becomes even more interesting when we recognize that the companies renting out intelligence are also discovering what their customers are willing to pay for. The artificial intelligence industry is not simply learning how to answer questions. It is learning which problems are worth solving, which businesses can be automated, which professional services are vulnerable to replacement, and which scientific discoveries might be converted into commercial products.
A company that owns the intelligence does not necessarily need to continue renting that intelligence to a thousand smaller companies if it can use the same machinery to enter the markets those smaller companies serve. It can integrate vertically, develop its own products, and capture the profits that once flowed through an entire ecosystem of intermediaries. And unlike the renter, it owns the factory. What makes this possibility especially fascinating is that we may already be witnessing the first recognizable example of the economic model I am describing. In September, Reuters reported that Anthropic had quietly established a physical biology laboratory in the San Francisco Bay Area. This was not simply another software application or an upgraded version of Claude. Anthropic had moved into the world of actual laboratory science, where its artificial intelligence could help direct biological experiments and participate in scientific investigation. The company had also acquired the biotechnology startup Coefficient Bio earlier in the year.
Consider what this means. Anthropic is developing increasingly powerful intelligence that has absorbed enormous amounts of existing scientific knowledge. It is now connecting that intelligence to the physical world, where hypotheses can be tested, experiments conducted, and discoveries pursued. The laboratory provides something that a language model alone cannot: a means of confronting its theoretical conclusions with biological reality. Today, Anthropic describes the effort as scientific research and collaboration, and it has not announced that this laboratory is producing commercially approved medicines. But the direction of travel raises a fascinating economic question. Why would a company possessing extraordinarily powerful scientific intelligence necessarily limit itself to selling access to that intelligence when it could eventually participate directly in the discovery, development, and commercialization of the products?
Imagine Claude becoming capable of generating promising molecular designs, directing robotic experiments, analyzing biological results, and improving subsequent experiments at a pace no conventional human research team could match. Human scientists, regulators, and clinical trials would remain essential, but the speed and economics of discovery could change beyond recognition. Now imagine what that capability could mean for the pharmaceutical industry. Companies such as Eli Lilly have spent generations building the research facilities, expertise, manufacturing capacity, and commercial infrastructure required to bring medicines to market. The industry depends upon enormous investments, long development cycles, and the possibility that expensive research programs will fail. What happens when a company whose primary business was once artificial intelligence begins using that intelligence to transform the economics of pharmaceutical discovery?
A future Anthropic laboratory, or an entire network of laboratories directed by increasingly capable successors to Claude, might someday make the research capabilities of today’s pharmaceutical giants look like Romper Room, the old children’s television show, if you are not old enough to know that. That would require breakthroughs in experimentation, clinical development, and manufacturing that are not yet assured, but the potential is extraordinary. And why would this transformation stop with medicine? The same pattern could emerge in materials science, energy development, engineering, agriculture, manufacturing, financial analysis, and virtually every other field where knowledge can be converted into commercially valuable products. The companies controlling the strongest artificial intelligence systems could eventually discover that the real money is not in selling the intelligence itself. The real money is in allowing that intelligence to produce things the rest of civilization wants to purchase.
We have been thinking of these companies as the next generation of software providers. Perhaps we should begin thinking of them as the possible foundations of entirely new industrial empires. Maybe Elon Musk will open up a burger chain with robots who make the perfect burger every time. Space Burgers or Xburgers, I can see the neon signs now. Amazon could use its infrastructure and intelligence to develop products and logistics systems beyond anything its current marketplace can accomplish. Google could apply its computational capabilities across science, medicine, transportation, and materials discovery. Microsoft could move deeper into automated industrial and enterprise systems. Meta could pursue entirely different forms of commercial intelligence, while Oracle could expand its position inside the infrastructure upon which increasingly automated organizations depend.
The precise winners and products remain uncertain, and some of these companies may continue renting substantial amounts of compute indefinitely, but they will soon see that they are losing money; the spreadsheets will dictate those realities. But the larger economic incentive is difficult to ignore: when intelligence becomes a productive asset, the owner can earn money from the output rather than merely charging someone else for access to the machinery. Elon is already doing what I am talking about with Starlink, Robotaxi, and soon Optimus robots. You won’t need PayPal or Apple Pay when there is X money. The man is getting into the fossil fuels business and more. How long before the other superscalers realize they are already behind the ball compared to Elon?
Artificial intelligence may eventually resemble electricity in one important respect. Most customers do not particularly care how the power is generated. They care about what the electricity allows them to accomplish. A company possessing an extraordinary amount of intelligence may eventually discover that its most profitable opportunity is not selling the equivalent of electricity by the kilowatt-hour but operating the factories that consume it. There is another development that makes this scenario considerably more troubling, particularly for smaller businesses attempting to survive the transition. The American political environment has just changed dramatically. The Democratic victories in the November 3 midterm elections have given the party control of both chambers of Congress beginning in January, while its candidates have also made significant gains in state governments. The results arrive during a growing political backlash against the enormous data centers necessary to support artificial intelligence.
This is not merely a theoretical concern about what politicians might do someday. Opposition to data center development has already become a significant political issue, and the election results suggest that the movement now has considerably more influence over future construction. According to reporting by the Associated Press, a large majority of victorious gubernatorial candidates supported stronger restrictions on data centers, with a substantial number endorsing bans or moratoriums. Local voters have also overwhelmingly supported restrictions in communities where these projects have appeared on the ballot. For politicians, the objections involve electricity prices, water consumption, land use, tax incentives, and the environmental consequences of building enormous computational facilities. Some of these are legitimate questions about how infrastructure costs should be distributed. But legitimate concerns do not eliminate the economic consequences of the policies used to address them.
Every additional restriction, permitting delay, construction moratorium, or expensive regulatory requirement can increase the difficulty of building new computational capacity. And who is best positioned to survive those obstacles? The companies with the most friends in the Democratic Party. Imagine what the landscape will look like when a group of people who think government should be in every corner of your life gets control of AI. You thought the COVID-19 pandemic was bad. George Orwell on steroids is what is coming. It won’t take long for that prediction to manifest. Polymarket will have a field day with that axiom. How much do you want to bet?
