I want to begin with a clause rather than a headline.
Last quarter I sat on a diligence call for a bandwidth DePIN that had just closed a nine-figure round. The deck was genuinely good: forty thousand nodes across ninety countries, a permissionless last-mile thesis, a token model that survived my second pass. Then we reached the service terms. Page nine, section four, one sentence, and I am paraphrasing but not by much: the operator reserved the right to adjust quality of service at its discretion to satisfy legal, contractual, or operational requirements.
Nobody else in the room flinched. I flinched, because that sentence is the entire Starlink dispute compressed into a footnote, and I had just watched an industry raise a hundred million dollars without noticing it had imported the exact governance flaw it claims to be solving.
Read the docs. Question the whisper.
The dispute itself is thin on facts and thick on meaning, which is precisely why it is useful. Ukrainian forces depend on SpaceX's Starlink constellation for battlefield communications. President Zelensky said publicly that speed restrictions were being applied to military use. Elon Musk said any restrictions were decided by the U.S. military and ultimately by the President. Trump, in comments reported at the time, said he was unaware of any restrictions and that they had been lifted.
That is the record. Six information points. No documents, no timestamps, no service annex, no sequence of events. The reporting's own temporal anchor is ambiguous, which matters more than it sounds, because "unaware" and "already lifted" cannot both be literally true at a single moment. Three public statements, mutually inconsistent, from three actors who each had a structural reason not to be the one holding the dial.
Now here is why this belongs in a crypto publication and not only a defense journal. The topology being described — a payer who does not own the asset, an owner who does not hold the final decision, and a decider who does not operate the network — is the same governance topology that a large share of DePIN and infrastructure tokens are currently marketing as an innovation. We did not invent that topology. We tokenized it, put a governance forum on top, and started calling it decentralization.
Consider the three roles in their crypto form, because they map with almost embarrassing precision. The token holders pay, through emissions and dilution, in the same way Kyiv pays or has a third party pay on its behalf. The foundation owns the intellectual property, the entity, and the contracts. The core development team decides. And at no point does any one of those three hold what a sovereign would recognize as full control, because the payment, the ownership, and the decision are deliberately separated across three parties who cannot coordinate without public disclosure. That separation is presented to investors as a security feature. In the Starlink case it is precisely what produced the crisis.
The asset that was weaponized was not oil
The first thing to notice is what got used. Not grain, not rare earths, not a shipping lane. Connectivity: sustained, stable, high-quality communication. And connectivity has a property that makes it a uniquely efficient coercive instrument — it cannot be stored, it cannot be resold, and it cannot be substituted at operational speed. An oil embargo is blunted by strategic reserves and by rerouting. A bandwidth throttle is blunted by nothing, because there is no buffer layer anywhere in the stack. The degradation is instant, it is total for the affected workload, and it is completely unhedgeable by the party on the receiving end.
I have been telling founders for two years that "we are the bandwidth layer" is not a marketing line, it is a liability disclosure. Most of them hear it as flattery. It is not.
The deeper shift is one that defense analysts have been circling for a while and that crypto people should recognize instantly: chokepoints are migrating. The old chokepoints were geographic — straits, canals, pipelines. They were visible, they had sovereign owners, and they could in principle be seized. The new chokepoints are institutional — payment rails, satellite constellations, cloud regions, chip design tools, app store review queues. They are invisible, they frequently have no sovereign owner at all, and no marine battalion can take one. Which means the entire traditional toolkit of deterrence by denial has no purchase on them. You cannot defend a chokepoint that has no territory.
Blockchains are institutional chokepoints by construction. Validator sets. Sequencer keys. RPC providers. Stablecoin blacklists. Bridge multisigs. Oracle committees. The migration that defense analysts are describing is the migration the crypto industry has been living inside for a decade without naming it out loud. We call these things "infrastructure," we call the aggregate "decentralized," and then we stop asking questions at exactly the point where the questions get expensive.
There is a second dimension to the asset question that deserves more attention than it gets. What was contested was not orbital position or radio spectrum in the abstract, but the specific combination of low-earth-orbit slot allocation, Ku/Ka-band coordination, and the on-orbit service capability to keep a constellation functioning. That combination has first-mover occupancy characteristics that money cannot reverse. A later entrant with unlimited capital still has to wait for orbital shells and spectrum coordination that were claimed years earlier. Which means the operator of a mature constellation holds something closer to an infrastructure monopoly than to a competitive market position, and infrastructure monopolies inevitably acquire political functions whether they want them or not.
Crypto should recognize that dynamic because it is the same dynamic that governs liquidity. The venue that gets there first with enough depth does not merely compete; it becomes the place where price is discovered, and price discovery is a public function performed by a private actor. We spent years arguing about whether that was acceptable and then stopped arguing once the fees arrived.
The ladder, and why the middle rungs are the dangerous ones
The Starlink case describes something more precise than a switch. It describes a ladder: normal service, then throttle, then geofence, then account suspension, then cutoff. The throttle is the lowest rung. It is reversible, it is deniable, and it can be escalated without ever being announced.
The insight I keep returning to is that the throttle's political efficiency far exceeds its military efficiency. It does not need to degrade combat capability by much. Its actual function is to remind the counterparty that the ceiling of their capability is defined by someone else's decision. That is a governance signal wearing the costume of a technical parameter. And because it is reversible, it costs almost nothing to send.
What makes the throttle a scalpel rather than a hammer is the asymmetry of what it degrades. Voice traffic survives a bandwidth restriction almost intact. High-bitrate video, imagery, mapping, and the real-time telemetry that connects a drone to its operator do not. Which means a throttle is not a general punishment; it is a precision strike on one specific mode of warfare — the drone-enabled reconnaissance-strike complex — while leaving command voice channels untouched. That selectivity is the whole point. It applies pressure with a fraction of the political cost of a shutdown, and it can be dialed back the moment the pressure achieves its purpose.
The crypto translation is uncomfortably direct. Selective degradation in a token network rarely looks like a freeze. It looks like a rate limit applied to specific contract interactions while general transfers flow normally. It looks like an RPC provider quietly deprioritizing requests from certain geographies. It looks like a sequencer that includes every transaction type except one. Nobody can point to a shutdown, the network is technically live, the block explorer is green, and the specific workload that mattered has been functionally disabled. If you want to know whether a network is really permissionless, do not ask whether it is running. Ask which workload it would refuse to carry under pressure, and then ask who would have to sign off on that refusal.
Every DePIN has this ladder whether or not it is documented. Normal emissions. Then "reward recalibration," which is rung two. Then geographic restriction, which arrives disguised as compliance. Then slashing. Then whitelist removal. Then exclusion.
Everyone audits the bottom rung. Every community AMA contains the question "can they freeze me?" and the answer is always no, it's permissionless, read the contract. But nobody asks about rung two. Rung two is where the coercion actually lives, because rung two is cheap, reversible, and does not trigger a governance crisis. It is a parameter change inside a range that the documentation permits. It happens in an operations runbook, not in a proposal.
Alpha hides in the silence of the audit. The alpha here is specifically in the middle rungs — the ones the docs don't describe, because the docs describe the protocol and the middle rungs live in the operations layer, which is where nobody looks and where everything is decided.
Conditional supply risk, the fourth class
Standard supply-chain taxonomy has three risks: capacity, inputs, logistics. This case reveals a fourth that is genuinely distinct, and that most crypto treasuries have no framework for at all. Conditional supply risk. The supply is not interrupted. The supply continues, indefinitely, but with political conditions attached that can change without notice.
The reason this class is structurally different is that it cannot be buffered. Inventory hedges capacity risk. Supplier diversification hedges logistics risk. Nothing hedges conditional supply risk except architectural diversification — meaning a second, functionally independent pathway that is already live before you need it. Not "we could build one." Already live. The gap between those two states is where wars are lost and where token ecosystems get captured.
Crypto's version of architectural diversification is the fork. Every community believes that if the foundation misbehaves, they can fork. This belief does enormous emotional work and very little operational work. A fork requires that the code is genuinely open and buildable, that the social layer is coordinated enough to execute under pressure, and that the value of the network is not primarily in the brand, the liquidity, and the exchange listings — none of which fork.
You cannot fork the satellites. You cannot fork a spectrum license. You cannot fork forty thousand units of hardware already bolted to rooftops. What you can fork is a governance token with no physical backing, which is precisely the thing that was never the constraint.
I want to state the central claim of this essay plainly, because I think it is the sentence the sector most needs to hear: decentralization of hardware is not the binding constraint; decentralization of the degradation right is. You can distribute every node on earth and still have one entity that decides which nodes get paid this epoch. That entity is the chokepoint. You will have built a distributed antenna and a centralized switchboard, and you will have called it a network.
The three-layer deniability stack, and the DAO shell game
The most transferable finding in the Starlink analysis is the structure of the deniability. Three layers, stacked.
Technically, the action is a quality-of-service adjustment — a parameter inside a permitted range, requiring no intent and admitting no malice. Legally, it is a commercial decision under a private contract between private parties. Politically, it falls within executive authority over military matters. Each layer is individually plausible. Stacked, they make the action essentially unappealable, because there is no forum where an injured party can bring a claim that survives all three simultaneously.
I have watched this exact stack operate in token governance, and I have watched it operate on purpose.
The technical layer in crypto is the timelock and the parameter range. The change passed the timelock, the guardian did not veto, the value was inside the bounds. Neutral. Nobody has to have decided anything.
The legal layer is the entity maze. The foundation in Zug. The Labs corporation in Delaware. The DAO that governs but has no legal personality and therefore cannot be sued or held to anything. The decision always lives in whichever entity has the most distance from the token holders and the least exposure to them.
The political layer is the community vote — and the community vote is the terminal deniability layer in this industry. It launders any decision into legitimacy regardless of who actually made it, because the outcome gets ratified by a process whose parameters were set by the deciding party. A vote with a quorum threshold you chose, on a question you framed, in a window you scheduled, is not a constraint. It is a receipt.
When all three layers are present, the token holder's remedy set is identical to Ukraine's: public complaint. That is the whole toolkit. Say it loudly, hope the reputational cost exceeds the operational benefit, and try again next quarter.
And notice the symmetry that I think is the most quietly damning detail in the entire episode. Nobody claimed the decision. Trump said he didn't know. Musk said it was the military and the President. The military said nothing at all. A decision that no one claims is a decision that no one can be held to. I have seen foundations build that pattern deliberately and I have seen them arrive at it by accident through organizational entropy, and the functional result is identical: an action with no author, and therefore no accountability surface.
There is a second-order effect worth naming. Musk's framing assigned the decision to two entities — the U.S. military and the President — that do not hold equivalent legal authority. The military executes a contract; the President exercises political discretion. Collapsing them into a single unnamed decider is either deliberate ambiguity or a genuinely imprecise description of a decision chain, and the available record cannot distinguish between those two possibilities. That ambiguity is itself the finding. When the description of who decided is unfalsifiable, the accountability is unfalsifiable too.
Governance sentiment: watching who resolves disputes where
I allocate roughly thirty percent of any infrastructure thesis to governance sentiment — not price action, not TVL, but who mobilizes, who abstains, and who actually holds keys. That allocation traces back to 2020, when I coordinated two hundred small holders in a MakerDAO vote against a collateral expansion I believed was underpriced for systemic risk. We ran weekly calls, we got to about fifteen percent of the vote, and we lost the substance while winning the process. What I learned was not that voting works. What I learned was that you can read a community's real power structure by watching where its disputes get resolved.
Apply that lens here. The interesting governance signal in the Starlink dispute is not the throttle. It is that all three parties chose public signaling over private coordination. Either no trusted back channel exists, or all three believe public pressure beats private negotiation. Both readings are bad. An alliance in which the members prefer to litigate in the press is an alliance that has already lost its private dispute-resolution layer.
Crypto's version of that tell is very specific and very reliable: when a protocol's core actors start resolving technical disputes on social media instead of in the forum or the RFC, the forum has lost legitimacy. It usually precedes a governance failure by about two quarters. The failure is never the argument itself. The failure is that the venue where arguments used to be settled stopped being authoritative, and nobody noticed the handoff.
The corollary for the party on the receiving end is that disclosure is a weapon and it has a cost structure. Zelensky going public converted a private option into a public commitment cost. Before disclosure, "we might throttle" is free. After disclosure, the next use has to survive "they throttled us and were forced to admit it." That does not prevent the action. It raises its price, and raising the price is the only thing a weaker party in an asymmetric dependency can actually do.
The freeze button nobody protests
I want to make this concrete with the case crypto already accepted without argument, because it proves the pattern generalizes.
Stablecoin blacklisting is the most widely deployed degradation right in the industry. It is a rung-three action on the ladder — not a throttle, a full freeze — and it is applied to addresses rather than to people, which gives it the same technical neutrality that the QoS adjustment has. The legal layer is a terms-of-service clause. The political layer is sanctions compliance, which is unarguable.
The market accepted this almost instantly, and I think the reason is worth stating plainly. The users most dependent on dollar-denominated rails are concentrated in economies where the local currency is failing, and those users cannot litigate a freeze, cannot diversify into a second rail at the same cost, and cannot wait out a compliance review. The people who most need the infrastructure have the least leverage over it. That asymmetry is not a bug in the adoption story. It is the adoption story.
I spent three months in 2022 counseling a hundred and fifty retail investors in Rome through the FTX aftermath — taxes, claims, the paperwork of being someone's unsecured creditor. What that period taught me is that trust is the scarcest asset in this industry, scarcer than liquidity and scarcer than talent. What the stablecoin example teaches is that trust is also the asset most easily borrowed against and least easily rebuilt. Every freeze that a user cannot contest spends a little of it.
So when I see a DePIN operator write a discretionary QoS clause into page nine, I do not score it as a legal formality. I score it as a withdrawal against the same account.
Why being used politically is a moat, not a wound
Here is the conclusion that makes people uncomfortable, so I will sit in it rather than resolve it quickly.
SpaceX's government business did not suffer from this episode. It probably strengthened. The dispute proved that the asset has policy-tool value, and policy-tool value increases the government's incentive to remain deeply bound to the supplier. Being used politically is a form of official endorsement. The thing that looks like a reputational wound from the outside is, from the inside, a procurement argument.
I watched the same logic play out in 2024 around the spot Bitcoin ETFs. The prevailing crypto-native read was that institutional wrappers were a dilution — that the asset was being domesticated. The read I published at the time, in a series I called From Speculation to Sovereign Reserve, argued the opposite: the wrapper was an educational instrument, and normalization through regulated vehicles is how an asset becomes too structurally important to kill. The same legislators who campaigned against this industry ended up writing the framework that entrenched it. That was not a reversal. That was the mechanism working.
The uncomfortable implication for token holders is that systemically load-bearing assets get protected and purely speculative assets get cut. The regulatory apparatus everyone fears is also the mechanism by which an asset becomes load-bearing. The projects that spent the last cycle maximizing distance from the state are disproportionately the ones that did not survive the drawdown.
This contradicts the industry's self-image, which treats proximity to power as contamination. The evidence treats it as a moat. I don't love that conclusion either. But I have stopped writing analysis that flatters the reader's politics at the expense of the data, and I would rather say it here than have someone discover it the expensive way.
Dual-use by architecture: the same stack, a different role
The structural fact that ties all of this to DePIN is the dual-use design.
Same constellation. Same terminal hardware. Same software stack. What separates civilian broadband from military communications is a service tier and an account permission. That's the whole difference. The military capability is not manufactured by a defense industrial base; it is invoked from commercial infrastructure that was built for someone else's use case.
That is a precise description of what a DePIN is. Same node, same client, same contract. What separates a retail participant from a strategic consumer is a role in an access-control list.
Which means the underwriting question for these networks is not "how many nodes." It is "who controls the role list, and what is the documented process for changing it, and does the party that benefits from the token sit on both sides of that process." That is the valuation. Everything else is a node-count chart with a narrative stapled to it.
There is a corollary that the sector has not priced. Dual-use architecture lowers the cost of building strategic capability and simultaneously raises its political fragility. You get the capability cheaply and you hold it conditionally, because you never built the industrial base that would let you hold it otherwise. Crypto is currently supplying that architecture to a growing list of institutional and sovereign buyers who have not modeled the conditionality, and who will discover it at the worst possible moment.
I ran into the same shape in 2026, working on a human-in-the-loop consensus framework for an AI-crypto hybrid protocol. We spent weeks on a question that sounds philosophical and is actually an engineering spec: when an autonomous agent's behavior conflicts with a stated human norm, who holds the interrupt, and what is the latency of that interrupt? The answer determined whether the system was safe. It also determined whether the system was ours. An agent whose connectivity can be throttled is an agent whose economic agency is conditional, no matter how sophisticated its decision loop. The throttle does not need to fire often to govern everything upstream of it.
The other thing I learned from auditing privacy claims
In 2017, during the ICO mania, I led a small team auditing the privacy architecture of Zcash. We found three gaps between what the marketing claimed and what the mechanism actually delivered, and we published a paper that reached about five thousand new users. The finding that stayed with me was not any of the three gaps. It was that the gaps existed because the team had been describing their intentions rather than their implementation, and the community had accepted the description because it was flattering.
I have carried that lesson into every infrastructure thesis since, and it applies with full force here. The DePIN sector describes its intentions — permissionless, censorship-resistant, community-governed — and the mechanism it ships is a discretionary operator with a role list. The description is not a lie. It is an intention. The mechanism is the thing that will actually be executed under pressure, and the two diverge exactly when the stakes are highest.
So my diligence process now starts from the mechanism, not the description. Show me the role list. Show me the operations runbook. Show me the parameter ranges in the deployment script, because the ranges tell me what can be changed without a vote, and what can be changed without a vote is what will be changed without a vote.
Contrarian: the lesson everyone is drawing is the wrong one
The consensus takeaway from this episode is "decentralize the hardware." I think that conclusion is going to cost a great deal of capital.
Decentralizing hardware does not help if the degradation right stays centralized, and in most networks the degradation right is centralized by construction because it lives in the operations layer rather than the protocol layer. Forty thousand independent node operators and one entity that decides which of them get paid this epoch is not a decentralized network. It is a centralized network with a distributed cost base. The capital expense got socialized and the control did not.
The second contrarian point is about narrative, and it is less comfortable. The DePIN sector is currently using this episode as a demand story: sovereign connectivity is a national security requirement, therefore our network. But the buyers who respond to a national security story are governments and defense primes, and those are precisely the customers least likely to accept a genuinely permissionless network. They will buy the network and then demand the throttle. The demand narrative and the decentralization thesis are in direct conflict, and the sector is selling both in the same deck, sometimes on the same slide.
I have watched this exact pattern in Layer 2. The real difference between the major rollup stacks was never technical; it was who could convince more projects to deploy chains first. Distribution beat architecture, every time, and the winner was the stack most willing to accommodate the deployer's institutional requirements. DePIN is about to run the same experiment at higher stakes. The first network to sign meaningful sovereign and enterprise deployments will win, and it will win by reintroducing the throttle under a compliance heading, and the community will ratify it because the alternative is losing the deal.
Takeaway: the next narrative is degradation rights
The next narrative in this sector is not "decentralized infrastructure." That phrase has been spent. The next narrative is degradation rights — who can turn what down, under what documented process, with what notice period, at which rung of the ladder, and what the counterparty can actually do about it before the next block.
The first team that publishes a contractually binding service floor, submits it to an auditor with the authority to check it quarterly, and accepts a real penalty for breaching it will absorb the institutional capital that everyone else is currently faking its way toward. Not because it is the most decentralized network. Because it is the only one whose ceiling has a number on it.
Read the docs. Question the whisper. And this time, read page nine.
The question was never whether your network can be turned down. It is whether anyone would tell you before they did.