OTWopentechwire
Tech Intelligence, Openly Wired
Policy

AWS Drops NDAs as Data Centre Permits Face Triple-Digit Moratorium Wave

Amazon's cloud chief pledges transparency on facility approvals while disputing water, power and emissions critiques - yet trust remains the harder infrastructure to build.

LT
Linh T. Pham
Southeast Asia Reporter · Hanoi
Oct 7, 2026
5 min read
AWS Drops NDAs as Data Centre Permits Face Triple-Digit Moratorium Wave
Credit: Ethan Cairns / Getty Images

The Transparency Pledge

Amazon Web Services has stopped requiring nondisclosure agreements when working with government agencies on data centre approvals, according to Matt Garman, the cloud division's chief executive. The announcement arrived buried in a lengthy defence of the industry, published as local opposition to hyperscale facilities intensifies across the United States.

Garman's brief statement on NDAs - a single sentence in a broader essay - marks a tactical shift for a sector that has long preferred to secure permits quietly, then face neighbours after construction begins. Environmental campaigner Erin Brockovich captured the pattern in recent remarks: projects materialise after permits are signed, developers ignore calls, and local officials bound by confidentiality clauses leave residents in the dark until bulldozers arrive.

The timing is deliberate. More than 100 data centre moratoriums are under consideration across US jurisdictions, Garman disclosed, a figure that underscores how quickly sentiment has soured. New York State has already imposed a one-year freeze on permits for large facilities. At Opentechwire, we've tracked similar resistance emerging in jurisdictions from Northern Virginia to rural Oregon, where electricity demand projections and water withdrawal volumes have triggered organised pushback.

The Myths Garman Wants Dispelled

Garman's essay attempts to rebut four claims he frames as misconceptions: that data centres consume excessive water, drive up electricity prices, emit vast pollution, and deliver negligible community benefit.

On water, he cited an Amazon-commissioned report asserting that direct data centre consumption represents half of one per cent of US industrial water use - less than golf courses or almond orchards. Nvidia recently unveiled cooling systems that the chipmaker claims eliminate nearly all in-facility water use. Yet both arguments address only on-site consumption. They sidestep the water embedded in electricity generation and semiconductor fabrication, two upstream dependencies that dominate the true footprint. Independent scientists have called for mandatory, audited disclosure of data centre resource use, noting the absence of federal or state reporting standards that would allow verification.

Electricity pricing, Garman argued, has risen in some states with concentrated data centre construction but fallen or grown more slowly in others. Where rates have climbed, he attributed the cause to aging grid infrastructure that failed to expand ahead of demand. An independent grid watchdog recently reported a 76 per cent year-on-year price increase on the largest US electrical grid, identifying data centres as the primary driver - a finding that complicates Garman's narrative.

On emissions, Garman objected to critics citing maximum pollution levels authorised in permits. A planned Amazon facility in Texas holds a permit for 33 million tonnes of carbon dioxide annually, exceeding any other power plant in the country. Garman countered that backup generators sit idle 99.9 per cent of the time, running roughly ten hours per year for maintenance testing. The distinction matters: permitted capacity reflects worst-case scenarios, but whether operators will rely on diesel generation during grid stress events remains an open question as AI inference workloads push baseload demand higher.

Community Contributions and the Trust Deficit

Garman highlighted Amazon's contribution of more than one billion US dollars over three years to communities hosting significant data centre footprints. The figure is substantial in absolute terms, yet communities increasingly weigh these payments against opportunity costs: industrial land rezoned from other uses, transmission infrastructure prioritised for private loads, and water rights reallocated during drought.

The deeper challenge is not financial but epistemic. Anthropic chief executive Dario Amodei recently described the AI backlash as fundamentally a crisis of trust, where citizens assume technology firms and regulators are collaborating against public interest. Writer Jasmine Sun observed that when data centre operators present rebuttals, the common response is disbelief. Dropping NDAs addresses one procedural irritant, but it does not rebuild the underlying confidence required for social licence.

The Moratorium Mathematics

Garman warned that if the wave of moratoriums becomes law, the United States risks ceding competitive advantage in AI infrastructure for generations. The framing positions data centres as a zero-sum global race, with permits as the bottleneck. Yet the opposition is not uniformly anti-technology; many jurisdictions are demanding that approval processes include earlier public consultation, binding commitments on renewable energy procurement, and transparent reporting on resource consumption.

At Opentechwire, we've followed similar dynamics in Singapore, where the government lifted a three-year data centre moratorium in 2022 only after tightening energy efficiency standards and requiring operators to demonstrate green energy sourcing. South Korea's Jeju Island rejected a hyperscale proposal in 2025 after residents organised around water scarcity concerns. The pattern is global: communities are not rejecting computing infrastructure outright, but they are rejecting the expectation that they should accept it on faith.

What Transparency Requires Beyond NDAs

Eliminating nondisclosure agreements is a necessary but insufficient step. Transparency in this context means auditable, third-party-verified data on water withdrawal and return flows, grid load profiles, and actual - not permitted - emissions. It means publishing power purchase agreements and renewable energy additionality calculations, so communities can assess whether a facility genuinely accelerates decarbonisation or merely shuffles existing clean electrons.

It also means acknowledging trade-offs. Hyperscale AI training and inference demand extraordinary energy density, and no amount of messaging will make that demand disappear. The question is whether operators will treat communities as stakeholders whose consent must be earned, or as obstacles whose objections must be managed.

The Broader Infrastructure Reckoning

The data centre moratorium wave is part of a wider reckoning over who bears the costs and captures the benefits of AI infrastructure. Chip fabrication plants face similar scrutiny over water use in Arizona and Taiwan. Subsea cable landings encounter resistance from coastal communities. Even renewable energy projects - essential to any credible decarbonisation pathway - trigger opposition when developers bypass local input.

Garman's essay reflects an industry still calibrating its approach. The shift away from NDAs signals recognition that the old playbook no longer works. Whether operators will go further - embracing binding commitments, independent audits, and genuine co-design with host communities - will determine whether the moratorium wave crests or becomes permanent policy in dozens of jurisdictions.

For now, the trust deficit remains the harder infrastructure challenge. Dropping a legal instrument is straightforward. Rebuilding confidence requires a longer build cycle, and no amount of capital expenditure can accelerate it.

Read next
Policy

Anthropic Shuts Down VPN Workarounds for Claude in Hong Kong

Arjun S. Mehta · 5 min
Policy

Why One Former Safety Leader Wants AI Regulated Like a Nuclear Reactor

Daniel R. Whitfield · 6 min
Policy

Washington Rebrands AI as "Super Intelligence" in Executive Order

Daniel R. Whitfield · 6 min
Spot something wrong? Email corrections@opentechwire.com. We log every correction publicly.