A single line in a trade brief caught my attention this week. AWS, according to the summary, terminated non-disclosure agreements with government agencies tied to its data center buildout. That was it. No figures. No agency names. No scope โ federal or municipal, all of them or some. Four data points, and two of them were the author's own speculation. I have spent eleven years reading disclosures this thin, and they resolve into one of two categories: a planted narrative, or a real story the person reporting it does not yet understand. Either way, the correct move is not to repeat the headline. It is to price the thing the NDA was hiding. Because a non-disclosure agreement is not a document. It is a balance sheet line item that nobody has to mark.
Let me establish the machine before I dissect it. A hyperscale data center is a physical asset with a twenty-to-thirty-year depreciation schedule and a brutal operating cost curve. Power is the dominant line item โ in my own modeling of these facilities, electricity alone runs 30 to 50 percent of operating expense. Land, water rights, cooling, and grid interconnect sit behind it. The margin on a data center is not a function of the rack price the customer sees. It is a function of the price the operator negotiated before the customer ever arrived.
That negotiation is where the NDA lives. A cloud provider does not sign a confidentiality agreement with a county government because it is shy. It signs one because the terms of the deal โ the tax abatement, the discounted power purchase agreement, the water allocation, the land grant โ are the actual product. The public sees a ribbon-cutting and a jobs number. The ledger sees a multi-decade subsidy that never appears on an income statement as a subsidy. It appears as a lower cost basis.
Now widen the frame. The same architecture of opacity built centralized crypto. For years, exchanges told depositors their balances were safe while the actual reserves lived in commingled wallets that no counterparty could inspect. The 2022 collapse I worked on โ the one where I traced 1.2 billion USDC from Alameda-controlled wallets into FTX operating accounts across fourteen days โ was not a failure of cryptography. It was a failure of disclosure. The exchange ran the same playbook as a data center deal: sign the customer, hide the terms, let the marketing carry the narrative. So when AWS moves on transparency, I do not read it as an ESG gesture. I read it as a ledger event. Something previously off-book is being dragged toward the light, and the only real question is who absorbs the repricing.
Start with what the NDA actually conceals, because the brief never says. A government-facing NDA in a data center deal typically bundles four things. First, the property tax abatement schedule โ often a decade or more of reduced assessment in exchange for job commitments. Second, the power agreement, frequently structured so the operator pays a rate that retail customers cannot access. Third, water and utility rights, which in drought-affected jurisdictions are the genuinely scarce asset. Fourth, the jobs and infrastructure commitments, which are the public-facing part and usually the smallest part of the value.
None of these are fraud. All of them are hidden. That is the entire point. The NDA is not protecting trade secrets; it is protecting a price. And a protected price is indistinguishable, from the outside, from a subsidy โ because it is one.
I learned to read this pattern in DeFi, not in cloud. In 2020, during the first DeFi summer, I audited a yield protocol using nothing but Etherscan and a Hardhat fork. The advertised APY was a number. The emission schedule was the truth. I modeled the reward distribution and found the algorithm would dilute existing holders by roughly 40 percent inside six months. The project did not fail because the code was wrong. The code did exactly what it was written to do. It failed because the yield was a subsidy โ paid to early entrants out of the pockets of later ones โ dressed up as a return. When the subsidy ran out, the number collapsed. Greed optimizes for yield, not for survival.
A data center tax abatement is the same instrument with a different wrapper. It is a yield paid by a municipality to an operator, funded by the tax base everyone else carries. For as long as the abatement holds, the operator's unit economics look superior. When it expires โ or when it is disclosed and challenged โ the economics revert to the real cost of power. The NDA was the mechanism that kept the number from being questioned.
This is why the AWS move matters more than the brief implies. If the company is genuinely terminating these agreements, it is choosing to mark its own subsidy to market. That is rare. It is also rational โ but not for the reason the headline suggests.
Here is the accounting reality. A hidden subsidy is worth more when it stays hidden, because visibility invites competition for the same terms. Every other hyperscaler, every AI compute startup, every colocation operator wants the same abatement, the same power rate, the same water allocation. An NDA keeps the queue short. Transparency lengthens it. So a pure profit-maximizing operator would never volunteer to disclose.
Which means either the disclosure is coerced, or it is priced differently than I am modeling. The brief uses the phrase "transparency push," and that word โ push โ is doing a lot of work. It implies an external force: investigative reporting, a community organization, a state legislature, an environmental review. If a push exists, the NDA was already failing as a shield. The company is not giving something up. It is retreating from a position it can no longer hold, and taking credit for the retreat.
I have seen this exact choreography on-chain. When a project faces a credible exploit, the team does not announce the vulnerability. It announces a "proactive security review." The breach was the teacher. The announcement is the student's essay. Code does not lie, but developers do โ and so do press releases.
Now the part the brief entirely omits: the physical constraint underneath the transparency push. Data centers are not contested because they are secretive. They are contested because they consume. A hyperscale campus draws power at the scale of a small city and water at the scale of a large farm. In jurisdictions with stressed grids and depleted aquifers, the community does not care about disclosure philosophy. It cares about the reservoir level. The transparency push is a downstream symptom of a hard physical limit โ the same way a DeFi exploit is a downstream symptom of a bad external call, not a bad intention.
And the demand curve pushing that limit is AI. The current buildout is not driven by ordinary cloud growth; it is driven by training clusters and inference fleets that consume power in blocks no previous generation of infrastructure required. That is the engine behind the land rush, the power contracts, and the water permits. It is also why the NDAs existed in the first place: the terms of an AI-era data center are more valuable, and more politically combustible, than the terms of a web-hosting barn. The transparency fight is an AI infrastructure fight wearing a compliance costume.
I will trace this back to first principles, because this is the part that transfers directly to crypto. In 2021, I dissected a major NFT collection and found that roughly 90 percent of its "unique" traits were hardcoded values stored off-chain, with no IPFS redundancy. I ran a link-rot check across ten thousand assets and found most of the images were either unrenderable or dependent on a single cloud bucket. The collection had a floor price and no content. Metadata is not ownership; it is merely a pointer. When the pointer's host goes dark, the asset does not degrade โ it vanishes. The owner held a receipt for a thing that lived on someone else's server.
A data center NDA is the same structure at institutional scale. The community holds a promise โ jobs, tax revenue, "partnership" โ that lives on the operator's private server. The public ledger, the one that would show the actual terms, is encrypted by contract. The residents are NFT holders who never read the metadata. They bought the picture. The picture was a pointer.
This is the genuine information gain in the AWS story, and it is the opposite of what the ESG framing wants you to conclude. Transparency is not a values position. It is a cost disclosure. And cost disclosures are only made when the cost has become unpayable in its hidden form.
Look at the incentive from the operator's side. If I am AWS, and I can no longer keep the abatement quiet because a local reporter has filed a records request, then my best move is to disclose first, on my terms, and reframe it. I trade a leaked liability for a controlled narrative. I convert an accusation into a commitment. That is not altruism. That is hedging โ the same move a protocol makes when it announces a "decentralization roadmap" the week before a regulator opens a file.
There is a parallel I ran into this year that sharpens the point. I audited an "AI trading agent" protocol that promised autonomous profitability. By reverse-engineering the oracle inputs, I found the AI was merely predicting market trends from centralized news APIs, not from on-chain data. The intelligence was a wrapper around a feed nobody could verify. The protocol was delisted by three aggregators within a week of the writeup. The lesson was not that AI is dangerous. It was that an unverifiable input produces an unverifiable output, no matter how sophisticated the label. A data center NDA is the same opaque input. You cannot audit the unit economics of a facility when the terms of the facility are encrypted.
And here is the tell the brief cannot answer, because it never asked. What is the scope? The brief does not say whether the terminated agreements are federal or municipal, whether the termination is total or partial, or what they covered. This is not a minor gap. It is the entire event. A partial termination โ say, retiring the property tax NDAs while keeping the ones touching critical infrastructure security โ is a different story than a blanket disclosure. The first is a strategic reframe. The second would be structurally reckless, because some data center agreements genuinely touch national security, and no serious operator discloses those.
So the honest read is this: the event, as reported, is too thin to confirm. What is not thin is the pattern it belongs to. Centralized infrastructure โ cloud, exchanges, stablecoin issuers โ has spent a decade running on disclosed marketing and undisclosed terms. The push is arriving from multiple directions at once: community opposition, grid stress, environmental review, and the slower, heavier hand of financial regulation. The NDA is just the first document to fall.
The bulls, for once, are not entirely wrong โ and I want to give them their due before I take it back. The conventional skeptic says transparency is a cost, and a cost is a weakness. That is too simple. A large operator can absorb a disclosure that a small one cannot. If AWS genuinely opens its data center terms, the immediate effect is that AWS alone bears the repricing. But the durable effect is that transparency becomes an expectation. And an expectation becomes a requirement. And a requirement is a moat.
This is the part the bear case misses. Transparency, once it becomes an industry standard, is a barrier to entry. A hyperscaler with the balance sheet to survive full disclosure can afford to make disclosure mandatory โ because the marginal operator cannot. The AI compute startups, the colocation challengers, the second-tier clouds: none can absorb the compliance cost of a full records regime. AWS can. So the "ethical" move and the "anti-competitive" move are the same move. A mirror reflects the face, not the value โ and this mirror happens to flatter the largest face in the room.
The blind spot is subtler, and it is the one I care about most as someone who audits both worlds. Everyone is treating "transparency" as a binary: disclosed or hidden. It is not. A disclosure that arrives through a press release, a controlled blog post, or a third-party attestation is not a ledger. It is a pointer. The ledger remembers what the marketing forgets โ but only if the ledger is the actual record, not a summary of it. A proof-of-reserves attestation from an accounting firm is a pointer. A tax abatement disclosed in a corporate sustainability report is a pointer. Both can be accurate, and both can be edited, delayed, or framed. The only disclosure that cannot be reframed is the one written to an immutable public record at the moment of the transaction. That is the difference between a chain and a memo. The AWS story is interesting precisely because it is a memo pretending to be a chain.
So here is the question I would put to anyone treating this as an ESG win. If the terms of a data center deal are worth disclosing now, why were they worth concealing then โ and who is still concealing theirs? The brief gives me four facts and no ledger. The ledger is where the answer lives. Until someone produces the actual agreements, with dates, counterparties, and figures, the transparency push is itself a pointer โ and a pointer is not ownership of the truth. Risk is a number until it becomes a breach. Right now, nobody has published the number.