The infrastructure bill has become the infrastructure battlefield. Every major tech player is simultaneously building compute capacity and fighting the communities where it lands, while those same communities are discovering that regulatory capture works both ways. Google pays households to use less electricity so its data centers can use more, then pivots to damage control by promising transparency on water use. Amazon announces a Data Center Commitment that includes asking government officials to sign non-disclosure agreements about planning applications, which is transparency only if you believe secrecy agreements count as disclosure. The pattern is unmistakable: hyperscalers are not negotiating with communities over whether data centers should exist. They are negotiating over how much visibility to grant while the decision is already made.
The token economy is starting to reveal its true cost structure, and it is not pretty. An OpenAI employee spent ten million dollars on AI tokens in a single month, a figure casually mentioned at a conference and treated as noteworthy rather than alarming. Simultaneously, forty-five percent of AI decision-makers say their investments must show positive return within one year, yet only twenty-two percent have seen positive impact on operating profit. The gap between token burn and business value is widening in real time. Google launches Agent Compute Engine to suspend inactive agents and preserve their state, a direct admission that the idle time cost of running agents will otherwise kill adoption. This is not optimization. This is triage for a business model that does not yet work at scale.
The regulatory response has split into two incompatible directions. Meta faces a twenty-five billion dollar fine for data sharing between 2010 and 2021, a judgment that arrives years after the harm and does nothing to change current behavior. Simultaneously, European governments are quietly moving toward Matrix-based messaging tools to escape US technology dependence, and the UK is setting up an AI lab to monitor biosecurity threats. One jurisdiction punishes past violations while companies continue present ones. Another builds sovereign alternatives. The NSO Group founder is now pitching governments on sovereign AI as national security. Nobody is actually preventing the harm. They are just arguing over who gets to do it.
Sloane Duvall