Amazon drops data center NDAs as AI agent startups chase credit card access

Amazon said it will stop using non-disclosure agreements when negotiating data center deals with local governments, joining Microsoft in a transparency shift that the company's hosts framed as a response to organized community resistance to AI infrastructure.

The change surfaced on TechCrunch's Equity podcast, where Anthony Ha, Sean O'Kane, and Rebecca Bellan worked through the week's biggest AI and infrastructure stories. Microsoft made a similar move earlier in 2025, ending the same kind of secrecy in its own local-government negotiations, though both companies have continued to close multi-billion-dollar lease deals for new campuses.

Community opposition has produced what the Equity team called a striking policy footprint: hundreds of proposed and enacted moratoriums on data centers have moved through local governments from New York to San Francisco since the start of the AI build-out. That backlash drove the secrecy in the first place and now shapes the reversal.

What does dropping the NDA actually change?

Local officials negotiating tax abatements, water rights, and power contracts have historically signed NDAs before seeing hyperscaler expansion plans. Those agreements kept capacity figures, power draw estimates, and water consumption projections away from constituents until after council votes.

Amazon's new posture removes that constraint. The hosts argued on Equity that the change will make it harder for communities to claim surprise after approvals land, and easier for opposition groups to organize before the deal closes.

The shift does not address the underlying economics. Data center capex still flows toward whichever utility region can deliver the cheapest electrons. AI training demand still outpaces municipal grid planning in most US counties.

Can consumer AI agents earn wallet access?

A second thread on the episode: a wave of consumer AI startups is asking users to grant their agents access to inboxes, files, and credit cards. The pitch hinges on whether the agent can be trusted to act without supervision.

Three companies landed in the Equity crosshairs: Tab, Underdog, and Hark. Each markets privacy as the central feature. The hosts framed the underlying gamble as straightforward. If consumers hand over payment credentials, the agent must deliver enough value to justify the risk of a runaway purchase.

The business model discussion went further. If agents book flights, order groceries, and pay invoices on a user's behalf, they also become sales channels. Merchants will pay referral fees, which means the agent's recommendation engine starts to look like an Amazon search result page.

Anthony Ha argued on Equity that the real question for consumer AI agents is whether the first wave of products will arrive as tools or as sales infrastructure. Trust, he said, is the product itself.

Why are incumbents building walls against outside agents?

Three incumbents drew the Equity team's attention for the opposite strategy: blocking outside AI agents from their platforms. United Airlines, Apple, and Amazon each appear to be raising technical and contractual walls against agents that scrape, browse, or transact on a user's behalf.

The motivation follows the logic that drove App Store policies a decade ago. Companies that control the customer relationship do not want a third-party agent arbitrating between the user and the purchase. Apple has the most leverage given its hardware-software integration. Amazon faces a sharper tension: it sells to the same consumers whose agents might undercut its marketplace margins.

Sean O'Kane said on Equity that the airline industry is the most exposed of the three because fare comparison sites already commoditized ticket sales in the 2000s, and agents could repeat that collapse at the booking step.

What is Ghost's $3,499 always-on box for?

Ghost, an AI hardware startup, began shipping a $3,499 computer built to run autonomous agents around the clock. The product sits in a category the Equity hosts said they do not expect to reach mass market: dedicated silicon for personal agent workloads.

The pricing puts the device above most consumer laptops. The positioning targets power users who want their agents to execute tasks while the household sleeps, without burning a gaming PC's electricity budget.

The Equity crew said the commercial ceiling is narrow. Mainstream consumers will run agents on phones, browsers, and existing desktops. Ghost's bet is that a subset of professionals will pay a premium for an always-on machine that survives power management rules.

How concentrated is Lambda's $4 billion raise?

Lambda, the AI cloud provider, is reportedly raising $4 billion at a valuation that places it among the largest non-public infrastructure bets of 2025. The hosts highlighted a single line item: one customer accounts for $35 billion of Lambda's contracted backlog.

Rebecca Bellan called the concentration a structural risk worth flagging on the episode. A single hyperscaler customer represents more than half of any provider's near-term revenue when the backlog clears. Any slowdown in that customer's training plans cascades through Lambda's capex schedule.

The raise signals that investor appetite for AI infrastructure remains strong even as the top three hyperscalers slow their own capex growth. Lambda's pitch is differentiation: custom networking, neocloud pricing, and shorter contract cycles than the AWS-Microsoft-Google tier.

Why is Uber paying $2.3 billion for ezCater?

Uber agreed to acquire ezCater for $2.3 billion in cash, the largest pure-play deal in the company's push into B2B services. ezCater runs an office catering marketplace that Uber will fold into its Uber Eats business.

The transaction marks Uber's pivot toward predictable enterprise revenue. Consumer rideshare demand has matured in North America. Uber has spent two years extending its logistics network into freight, grocery, and now corporate food.

The hosts framed the price as modest relative to Uber's market capitalization but aggressive relative to ezCater's historical multiples. Catering marketplaces have thin margins and long customer onboarding cycles, which means Uber is buying both a sales pipeline and a foothold in HR and procurement budgets.

What does a White House 'Super Intelligence Force' actually do?

The Equity episode also surfaced reporting on a White House effort the team labeled a "Super Intelligence Force," an internal coordination body for federal AI policy. Details remained thin, but the framing suggested an attempt to align procurement, research funding, and export controls under a single operating group.

The hosts treated the announcement as more procedural than substantive. Past attempts to coordinate AI policy across agencies have produced overlapping working groups and few binding decisions. The White House's prior AI executive orders have set direction without funding.

A standing force with authority over interagency budget would be a different instrument. Whether the new body has that authority will determine whether it operates as a planning shop or as a federal AI buyer with real leverage.

How do these stories fit together?

Three pressures run through the episode: community resistance to physical AI infrastructure, the trust economics of consumer AI agents, and the consolidation of compute power among a small number of buyers and sellers. Each thread shapes who controls the next layer of the AI stack.

Amazon's NDA concession is the smallest of the three by dollar value. The agent commerce question is the largest by consumer reach. The Lambda concentration risk sits between the two as a financial exposure.

Rebecca Bellan told listeners the threads are not independent. Local moratoriums slow the data center build-out, which tightens GPU supply, which raises the cost of running an agent, which makes the Ghost box and Lambda's enterprise contracts more attractive. The loop runs in one direction.

What to watch next

The Equity team flagged TechCrunch Disrupt 2026 as the next major checkpoint, with a Builders Stage appearance by the Equity crew Tuesday at 9 a.m. Two follow-up items will move the story: whether Amazon's NDA policy extends to its AWS lease agreements rather than only retail and logistics real estate, and whether Tab, Underdog, or Hark disclose paid referral arrangements with merchants before the first consumer pilots ship. The AI infrastructure cycle is past the point where secrecy was sustainable, and whether transparency produces better local deals, healthier consumer agents, or more concentrated cloud revenue will define the next twelve months of coverage.