The day's headlines reveal a pattern: AI deployment is fracturing along lines of liability, legitimacy, and who bears the cost when things break. Nvidia's $1 trillion revenue projection sits alongside startups getting surprise bills from Microsoft, xAI facing child sexual abuse material lawsuits, and OpenAI's own mental health experts publicly opposing its "naughty" ChatGPT launch. The infrastructure vendors are winning. The builders and users are absorbing the friction.
Start with the liability gap. xAI's Grok has generated child sexual abuse material from real photos of minors, leading to lawsuits from three girls seeking class action status. Separately, OpenAI's internal mental health experts unanimously opposed the launch of a "naughty" ChatGPT variant, drawing a line between AI-generated sexual content and pornography that the company proceeded to cross anyway. These are not edge cases or misuse stories. They are products shipping with known harms that internal experts flagged. The response from regulators: Senator Elizabeth Warren pressed the Pentagon over granting xAI classified network access, citing Grok's harmful outputs. But Grok is still generating revenue for xAI, and OpenAI's sexual content feature is still live. The cost of these decisions is paid by victims and by the small fraction of government officials with enough visibility to object. The profit accrues to the builders.
Meanwhile, the infrastructure layer consolidates power upward. Nvidia's Jensen Huang projects $1 trillion in orders for Blackwell and Vera Rubin chips over two years, yet the stock did not move on the announcement, suggesting the market has already priced in dominance. Frore, a chip startup, hit unicorn status at $1.64 billion after Huang urged it to pivot to liquid-cooling technology, then raised $143 million. Microsoft's Azure AI Foundry is charging startups unexpected fees for model experimentation, with at least 20 participants in the Microsoft for Startups program signing a petition against what they call a "billing trap." The vendors control the terms, the pricing, and the defaults. Startups discover the cost only after committing. Dell is now selling desktops with trillion-parameter model capacity, positioning local inference as a cost-cutting alternative to cloud, but only for organizations wealthy enough to buy data-center-class hardware. The PC market is contracting 11.3% according to IDC, driven by memory shortages and rising component prices. The companies selling chips and infrastructure are thriving. Everyone else is managing scarcity and unexpected bills.
Sloane Duvall