For the better part of a decade, Amazon Web Services perfected a particular art form: arriving in small towns with promises of jobs and tax revenue, extracting favorable deals, and binding local officials to silence with non-disclosure agreements that prevented communities from understanding what, exactly, was being built in their backyards. That era appears to be ending—not because Amazon discovered transparency as a virtue, but because the backlash finally became more expensive than the secrecy was worth.
The company announced this week that it will no longer require NDAs from government officials involved in data center negotiations, a practice that had drawn lawsuits, investigative journalism, and increasingly organized opposition from Virginia to Ohio to Oregon. The policy change arrives as AWS races to expand capacity for its AI workloads, a buildout that requires not just land and power but something harder to purchase: community consent.
The NDA machine
Amazon's use of confidentiality agreements in infrastructure deals was never technically illegal, but it created a democratic deficit that proved politically toxic. Elected officials found themselves unable to discuss projects with constituents. Environmental reviews proceeded with key details redacted. In some cases, residents learned a facility was coming only when construction equipment arrived. The company justified the practice as protecting competitive intelligence, but critics argued it was really about preventing the kind of organized opposition that can delay or kill major projects.
The tipping point came as data center energy demands collided with grid constraints and climate commitments. A single large AI training facility can consume as much electricity as a small city, and communities began asking why they should bear the infrastructure burden for services they might never use. When those questions couldn't be answered—because the answers were under NDA—frustration curdled into outright hostility.
The AI imperative
Amazon's retreat from secrecy is inseparable from its AI ambitions. The company is locked in an infrastructure arms race with Microsoft, Google, and a resurgent Oracle, all of whom are scrambling to build the compute capacity that enterprise AI customers will demand over the next five years. Delays measured in months can mean billions in lost contracts. Against that calculus, the marginal benefit of keeping a megawatt figure confidential looks increasingly foolish if it means adding a year of community opposition to your permitting timeline.
The announcement also reflects a broader recalibration in how tech giants approach physical infrastructure. After years of treating real estate and energy as problems to be optimized away, companies are discovering that atoms are harder than bits—and that the people who live near those atoms have votes.
Our take
Amazon deserves no particular credit for abandoning a practice it should never have adopted. But the policy shift is still meaningful, because it acknowledges something the industry has been slow to accept: the AI boom requires a social license, not just a building permit. Other hyperscalers would be wise to follow suit before their own NDA strategies become litigation targets. The communities hosting these facilities are not obstacles to be managed—they are stakeholders whose cooperation will determine whether the next generation of AI infrastructure gets built on schedule or gets mired in the kind of political resistance that no amount of lobbying can overcome.
