Amazon drops data center NDAs, and AI agents want your credit card

Amazon says it will stop using nondisclosure agreements when negotiating data center deals with local governments, according to TechCrunch. The shift follows a similar move by Microsoft earlier this year. The two companies are among the largest builders of the massive computing facilities that train and run AI models, and their land, power and tax negotiations with local officials have drawn growing scrutiny.

The secrecy around those deals has fed community backlash against AI infrastructure. Opposition has produced hundreds of proposed and enacted moratoriums on data centers, spanning from New York to San Francisco, according to the report.

On the latest episode of TechCrunch's Equity podcast, Anthony Ha, Sean O'Kane and Rebecca Bellan discussed whether ditching NDAs will actually change how data center deals get done. They also examined why trust may be the real product for personal AI startups, along with a couple of deals of the week.

Among the topics covered on the episode: the White House's plans with its new Super Intelligence Force; Ghost's $3,499 computer built to run AI agents around the clock, which the Equity crew does not see going mass market; Tab, Underdog and Hark pitching privacy as their product, plus a business model that could turn a user's agent into a salesperson; United, Apple and Amazon putting up walls against outside agents; Lambda's reported $4 billion raise and the single customer behind $35 billion of its backlog; and Uber's $2.3 billion all-cash deal to enter the office catering space with ezCater.

The second thread in the episode concerns a wave of startups betting that consumers will hand AI agents access to their inboxes, files and credit cards, provided those startups can get websites to let the agents in. That is the central tension for what the industry often calls personal AI agents: to be useful, an agent needs credentials and permissions that reach deeply into a person's digital life, but the sites and services it wants to act on have their own incentives to block it.

This suggests trust is the scarce resource in consumer agents. A user handing over inbox and payment access is taking on real risk — an agent that misreads a message or misclicks a checkout can cause financial or privacy harm, and the user may not be watching when it happens. Startups that position privacy as their core product are effectively arguing that they can be the trusted broker between a person and the many services that person uses. Whether that becomes a durable business or a feature absorbed by larger platforms is an open question the episode raises rather than answers.

The walled-garden dynamic is already visible. Companies named in the discussion — United, Apple and Amazon — are described as putting up barriers against outside agents. For developers building agents, that means the hard part is often not the model but the integrations: logging in, parsing data, completing purchases and staying within each platform's terms. A likely trade-off is that agents will proliferate first in categories where companies choose to cooperate, such as through official APIs or partner programs, and stall in categories where incumbents see agents as a threat to their own checkout, loyalty or advertising revenue.

The business-model angle raised is worth noting for product teams: if an agent can act on a consumer's behalf at checkout, it could also steer that purchase. The episode points to a model that could turn a user's agent into a salesperson, which raises the question of whose interests the agent is optimizing for when it recommends a product.

On the infrastructure side, Lambda's reported $4 billion raise and the single customer behind $35 billion of its backlog illustrate how concentrated AI compute demand has become. In practice, that concentration is one reason communities and regulators are paying closer attention to data center siting and the deals behind it — the same deals Amazon now says it will negotiate without NDAs.

Amazon's change is a disclosure policy, not a commitment about what it will build or where. The report does not say whether other large data center operators will follow, or whether dropping NDAs will alter the terms communities ultimately accept. But it does put more of the negotiation into public view, which is where moratorium fights and local approvals are decided.

Why it matters: Developers and product teams working on data center siting, AI infrastructure or local permitting should expect more public visibility into deals, and more organized opposition regardless. For teams building consumer AI agents, the practical constraint is permissioning: getting inbox, file and payment access from users is only half the problem if major platforms keep their doors closed. The trust and business-model questions the episode raises are likely to shape which agent products reach scale.