The AI bubble is getting a face: Dario Amodei, CEO of Anthropic, is calling for government regulation for the most powerful AI tools — while simultaneously announcing his company's IPO with a valuation of over $2 trillion. When was the last time a company said its product could threaten humanity — and then asked for your money? Jay Martin, economic analyst and host of the Vancouver Resource Investment Conference, dissects in a new analysis what lies behind this contradiction: no moat, a merciless price war — and a regulatory maneuver that America has seen before.
Warren Buffett made the term famous: A moat is what prevents customers from leaving a company and stops competitors from undercutting prices. Apple has one — anyone who switches loses their photos, their family chat, and the synchronization of all devices. Visa has one — every new card network would have to convince millions of merchants and millions of customers simultaneously.
AI companies like Anthropic or OpenAI have none. Their corporate customers — whether Uber or any mid-sized software provider — communicate with every AI provider in exactly the same way. A switch takes an afternoon. Platforms like OpenRouter even allow simultaneous connection to hundreds of models and switching between them in seconds. The CEO of a sales software company put it bluntly in the Wall Street Journal: "The honeymoon phase of being tied to a single model manufacturer is over."
Around 80 of the 260 pages of the Anthropic IPO prospectus deal with risks. Among them is the warning that advanced AI could lead to an absolute catastrophe and threaten the survival of humanity. To the same investors reading about this danger, Anthropic is simultaneously offering shares in a company valued at over $2 trillion. For Wall Street, this is not a contradiction — it is a business model.
The problem: Anthropic generated $4.6 billion in revenue last year while incurring an operating loss of $8 billion. The company has promised the big tech corporations — Microsoft, Amazon, Google, Oracle — a total of around $1 trillion for computing power. This promise can only be fulfilled if revenues increase sharply. And for that, Anthropic needs loyal customers who pay more than they do today — customers it does not have, according to its own analysis.
In the summer of 2026, OpenAI cut the price for one model by 80 percent and for another by 20 percent. Anthropic responded promptly with its own price cuts. A race to the bottom in which neither side is making a profit anyway.
The pattern is intensified by Chinese competition. In mid-2026, the Chinese company Moonshot released an AI model called Kimi K3, which outperformed the best models from Anthropic and OpenAI in independent benchmarks for code generation — and made it available online for free. Companies only have to pay for their own computing costs; the model itself is free. The share of American models in total AI traffic on OpenRouter fell from 70 percent to 30 percent within a year. Anthropic's share among customers using both providers fell from 75 percent at the beginning of the year to 50 percent by September 2026.
A survey of 400 companies in the spring of 2026 found: Only 11 percent were able to predict their AI spending with 10 percent accuracy. Uber is said to have exhausted its entire AI budget for the current year in the first three months. The reaction: cost audits, switching providers, and searching for cheaper models.
Jay Martin draws a historical comparison. In the 1930s, American railroad companies lost customers to trucks — cheaper, more flexible, door-to-door. Instead of competing, they pushed Washington for regulation. The official justification: order on the roads, an end to ruinous competition. The result: for decades, new trucking companies needed a government permit, which was routinely denied.
The principle is called Regulatory Capture — when companies that are supposed to be regulated end up writing the rules themselves. Jay Martin sees the pattern in the current AI debate: Dario Amodei demands that every AI model — whether American or Chinese — must pass government safety tests. Anthropic has lawyers, offices, and employees in the country to undergo these tests. A free model from China usually does not. OpenAI is also pushing for a national security ban on Chinese models. Political groups supported by Anthropic and OpenAI spent more than $50 million on the midterm elections by the end of September 2026.
The message to the public: Our AI is dangerous, the government must intervene. The message to investors: The same product is a historic buying opportunity. A product that can allegedly destroy the world is conveniently also worth $2 trillion.
Around 40 percent of the total market value of the S&P 500 is accounted for by just 10 companies — 9 of which are directly linked to AI. In the first three months of 2026, about three out of every four dollars of new American economic growth came from the AI industry. The AI capital expenditures of big tech corporations are approaching one trillion dollars annually, with around 80 percent of these funds flowing into raw materials — copper, nickel, silver, lithium — that make the operation of data centers possible.
If the promised revenues from Anthropic and OpenAI fail to materialize, the payment promises to Microsoft, Amazon, Google, and Oracle cannot be fulfilled. Capital expenditures slow down. The companies that depend on them feel it: chip manufacturers, energy suppliers, and construction companies. And retirement plans, which are significantly invested in these same ten companies, feel it as well. Jay Martin summarizes it this way: If you can't beat a cheaper competitor in the market, you ask the state to take care of them for you — and in the end, the customers pay.
An economic moat is anything that prevents customers from switching to a competitor and stops competitors from permanently undercutting prices. Classic examples: network effects, high switching costs, and cost advantages through access to resources. AI chatbots are in pure price competition because switching providers usually requires no effort.
Regulatory Capture describes the state in which companies that should actually be under government control de facto help shape the regulatory rules themselves — in such a way that competitors are excluded without an official ban having to be issued.
Jay Martin is a Canadian economic analyst and host of the annual Vancouver Resource Investment Conference. His YouTube channel analyzes macroeconomics, commodities, and market dynamics. Further articles on similar economic analyses can be found at Spargold: Markus Krall on the looming Euro crisis.
Legal Notice: This article is for informational purposes only and does not constitute investment advice. All information is provided without guarantee.
