The AI Price War Is Breaking the Business Model

A Ban, a Blowup, and a Business Model Under Pressure

The same week that the leaders of Anthropic, OpenAI, and Google DeepMind flew to France for the G7, the US government gave Anthropic 90 minutes to shut down access to its most widely available model. That is not a coincidence. It is a signal of how completely geopolitics, corporate rivalry, and an unresolved regulatory vacuum have fused into a single pressure point for the AI industry.

The episode crystallizes something Neeta has been flagging on Good Revenue for months: the frontier AI race is not just a technology competition anymore. It is a political and financial fight, and the bill is coming due on multiple fronts simultaneously.

How Fable 5 Got Banned

The backstory here is genuinely strange. According to reporting from The Verge, Amazon CEO Andy Jassy flagged to Treasury Secretary Scott Bessent that researchers at Amazon had identified jailbreaking vulnerabilities in Fable 5, Anthropic's most capable publicly available model. By the end of Friday, June 12th, the White House had banned Anthropic from exporting the model and barred foreign nationals from using it.

The practical problem with that order was immediate. The AI research and development workforce is deeply international. A blanket restriction on foreign national access is, as the episode notes, simply impossible to enforce without shutting the product down entirely. So Anthropic did exactly that, pulling both Fable 5 and its more powerful underlying model, Mythos, from general access.

A week later, a White House meeting had produced no resolution. The stalemate continues.

What makes this particularly thorny is that Anthropic's defense is not that their model is perfectly safe. It is that a 100% jailbreak-proof model does not exist, that their safeguards are best in class, and that the same standard applied to Fable 5 would apply equally to every other frontier lab, including OpenAI. A coalition of cybersecurity leaders signed a letter backing that position, calling for regulation grounded in scientific evaluation, developed transparently, and applied consistently across the industry.

The political dimension matters here too. The Defense Department labeled Anthropic a security risk earlier this year, and Anthropic has been fighting that designation in court. The 90-minute shutdown order lands in that context, and it is hard to read it as purely a technical dispute.

The Financial Floor Has Dropped Out

Whatever happens with the regulatory fight, the numbers released this week make the financial stakes visceral. OpenAI's audited financials, published through a partnership between journalist Ed Zitron and the Financial Times, show the company lost more than $5 billion in 2024. In 2025, that figure exploded to $38.5 billion, against $13 billion in revenue and $34 billion in costs.

These are not rounding errors. These are structural deficits at a company actively pursuing an IPO.

The underlying dynamic driving those losses is one that is now hitting demand as well as supply. As AI labs have moved from flat subscription pricing toward token-based billing, the true cost of using powerful models has become visible to customers. And customers are responding rationally. Large firms and startups alike are routing simpler tasks to lighter, cheaper models and reserving the expensive frontier models for genuinely complex work. The AI subsidy era, where labs absorbed costs to build usage habits, is ending. What replaces it is not yet clear.

The Cost Side Wildcard

There is one piece of potential relief on the horizon, though it is early. Amazon reported a new data center architecture that delivers 40% lower power consumption, 69% fewer routers and switches, and 33% better throughput at its AWS facilities. If those numbers hold at scale, the cost structure for running AI infrastructure could shift meaningfully, and every competitor in the data center space will be working to match or beat it.

That kind of efficiency gain is exactly what the industry needs. But it is a supply-side development in a market where the demand-side question, whether customers will pay what frontier AI actually costs, remains wide open.

What Comes Next

The Anthropic situation is not going to resolve cleanly. The model will eventually come back in some form, the political fight will grind on, and the regulatory framework for AI export controls remains incoherent. Meanwhile, the economics of the business are forcing every player toward a reckoning that cannot be postponed indefinitely.

The surprising insight here is not that AI is expensive or that governments are nervous about powerful models. It is that both crises are converging at exactly the moment the industry was supposed to be proving out its commercial model. The companies building the most capable AI on earth are simultaneously fighting regulators, burning extraordinary amounts of capital, and watching their customers flinch at the bill. That is not a temporary turbulence. It is the shape of the business right now.