Why data center deals are finally losing their secrecy

Amazon and Microsoft are abandoning NDAs on local infrastructure deals just as a new wave of AI agent startups bets that trust, not secrecy, will decide who wins.

For years, data center deals between tech giants and local governments got sealed behind airtight nondisclosure agreements. Residents found out their town was getting a sprawling new server campus only after the trucks showed up and the zoning variance was already signed. That era looks like it’s ending. Amazon says it will stop requiring NDAs when it negotiates with cities and counties, following a similar pledge from Microsoft earlier this year — and on a recent episode of TechCrunch’s Equity podcast, hosts Anthony Ha, Sean O’Kane and Rebecca Bellan dug into whether dropping the secrecy will actually change how these deals get done.

Why Is Amazon Dropping Its NDAs?

The backlash has been building for a while. Secrecy around data center siting, water use, tax breaks and power draw has fueled community pushback against AI infrastructure projects across the country, and that opposition has turned into hundreds of proposed and enacted moratoriums, from New York to San Francisco. When neighbors can’t see the terms of a deal until after it’s signed, every surprise — a new substation, a diverted water line, a tax abatement nobody voted on — reads as a betrayal rather than a negotiation.

Dropping NDAs doesn’t make these projects less contentious. It moves the fight earlier, into public view, where residents and local officials can weigh in before the concrete gets poured. That’s a meaningfully different posture for an industry that has mostly treated local pushback as a PR problem to manage rather than a signal to respond to. Local businesses and civic groups watching these fights closely are often the same communities experimenting with building an AI-powered local niche agency of their own, which makes the optics of a hyperscaler’s secrecy even more consequential.

AI’s Communication Problem Runs Deeper Than NDAs

The Equity hosts pointed out that the industry’s trust gap isn’t only about data center deals — it’s about how AI companies talk about what they’re building at all. Grand pronouncements about superintelligence and sweeping corporate rebrands land very differently in a town hall than they do on a keynote stage. If a company can’t explain in plain terms why a facility needs as much power as a small city, dropping the NDA won’t fix the relationship with the people living next door to it.

For founders outside the hyperscaler world, the lesson travels well: how you explain your product’s resource use, pricing, or data practices to a skeptical customer matters as much as the product itself. Teams already rethinking how AI fits into daily operations might find useful groundwork in how AI productivity gains get put to real work once the hype settles into routine use.

Trust Is the Real Product in Personal AI

Agents Want Into Your Inbox

The second half of the episode turns to a different kind of trust problem: a wave of startups betting that consumers will hand AI agents access to their inboxes, files and credit cards — if those startups can convince websites and users to let them in. The pitch is straightforward. An agent that can read your inbox can find refunds you forgot to claim, subscriptions you forgot to cancel, or a flight credit about to expire. The money is real. The access required to get at it is the hard part.

A Trust-First Pitch, Literally

That access problem is why a cluster of agent startups are now pitching themselves on privacy and trust first, features second. The hosts also flagged Ghost, a hardware startup building a $3,500 computer purpose-built for running personal AI assistants — a sign some founders think the trust problem is partly a hardware problem, not just a software one. Whether that’s the winning bet or an expensive detour is still an open question, but it’s a familiar one for anyone who has watched small firms borrow from the YC AI playbook and discover that the hardest part of shipping an agent isn’t the model — it’s getting someone to trust it with their data.

It’s also why evaluation matters more than most pitch decks admit. An agent that mishandles one email, one payment, or one file is an agent nobody recommends to a friend, and building an eval setup that catches failures before users do is turning into its own competitive layer in this market.

Quick Hits From the Deal Desk

Two other deals rounded out the episode. Lambda, the GPU cloud provider, raised $4 billion, a reminder that neocloud providers are still racing to meet demand even as some of them face a backlog of orders they can’t yet fill. Uber announced a $2.3 billion move into catering, a reminder that not every growth bet in this economy is about AI at all — sometimes it’s just logistics, scaled up.

The Takeaway for Builders

Put the two stories side by side and a pattern emerges: whether it’s a hyperscaler negotiating with a city council or a startup asking for inbox access, the companies winning the next round are treating transparency as the opening move, not the concession they make after getting caught. The reason data center deals are getting more public is simple — secrecy stopped working. Personal AI agents are leading with trust for the same reason: asking forgiveness after a privacy mistake is a much harder sell than asking permission up front.

For founders building anything that touches sensitive data — financial, personal, or local-government-sensitive — the takeaway isn’t complicated: say what you’re doing, why, and what happens if it goes wrong, before someone has to ask. That’s cheaper than rebuilding trust after losing it, and it’s a lot cheaper than a moratorium.

The reason data center deals are getting more public is simple — secrecy stopped working.

That’s cheaper than rebuilding trust after losing it, and it’s a lot cheaper than a moratorium.

▶

Sources

• Amazon and others are done keeping their data center deals secret | Equity Podcast — TechCrunch — Episode with Anthony Ha, Sean O'Kane and Rebecca Bellan, published 2026-10-09