Everyone is selling you compute. Nobody is showing you the failure mode.
On a desk most crypto natives will never think about twice, an analyst at Morgan Stanley did something unusually quiet this quarter. They kept Oracle at market perform. They set a price target of $210. And then they offered three compressed points of reasoning — no revenue line, no capital expenditure breakdown, no backlog figure, no timestamp. Five pieces of information in total. Everything that would let a reader actually verify the claim was simply absent.
I have spent enough of my life reading disclosures to understand that what a document leaves out is frequently the document. The words that stayed with me were not the rating or the target. They were "capacity delivery" and "maintain margin." Six words describing a company under physical constraint. Six words that describe, with almost surgical precision, the exact trap that decentralized compute has spent four years insisting it could out-engineer.
Silence is the loudest audit.
The Company Everyone Forgot Was an Infrastructure Play
To understand why a database vendor should matter to anyone holding a hardware-dependent token, you have to rewind the narrative. For thirty years Oracle was the tax collector of enterprise software. You paid for a database license, you paid for support, and you paid forever, because migrating off a working Oracle deployment is one of the most expensive engineering decisions a bank or a telecom can make. That is not a moat built on love. It is a moat built on switching cost, and switching cost is the most durable moat in enterprise technology.
Then the story changed. Oracle stopped being described primarily as a database company and started being described as a cloud infrastructure provider. OCI — Oracle Cloud Infrastructure — became the vehicle. And the reason it became interesting to people far outside the enterprise IT world is that OCI does something the three hyperscalers were slow to commoditize: it runs large, dense, high-interconnect GPU clusters, and it runs them on bare metal with a network topology that model-training workloads actually appreciate. In plain terms, Oracle found a niche where it could be competitively differentiated rather than competitively crushed.
That is the version of Oracle that Wall Street has been pricing. The valuation stopped being a function of database renewals and started being a function of whether Oracle can build data centers fast enough to serve enormous AI contracts it has already signed.
If that sentence sounds familiar, it should. Strip out the corporate name and you have written the pitch deck of every decentralized physical infrastructure network raised in the last eighteen months.
The Word That Betrays the Model
"Capacity delivery" is not a phrase a company uses when demand is the problem. It is a phrase a company uses when supply is the problem. When an analyst writes that the key variable is whether capacity can be delivered, they are telling you, without saying it, that the demand side has already been underwritten by somebody with a very large balance sheet and that the bottleneck has moved to concrete, copper, cooling, and above all electricity.
I want to be honest about evidence here, because my entire professional instinct is to refuse the comfortable inference. The source material is five data points. There are no revenue figures, no capex breakdown, no order backlog, no timeline. Everything I am about to argue about Oracle itself is an inference drawn from industry structure, not a fact disclosed in the note. I am flagging that plainly, because the crypto industry has a chronic disease of treating inference as disclosure and then trading on it as if it were verified.
But the inference is not weak. When the two named variables are delivery and margin, the underlying physics of the business are being described: a company that must spend money before it can earn money, at a scale where the spending is enormous relative to the earning, and where the earning depends on a small number of counterparties who each represent a material fraction of the whole.
That is a capital-intensive business entering a phase where it must convert balance sheet into concrete faster than competitors can convert concrete into contracts. And the return on that conversion is not guaranteed. It is deferred, it is depreciated, and it is hostage to the cost of the money used to finance it.
Why This Is an Oracle Problem in the Other Sense
There is a temptation, when a blockchain writer reaches for a story like this, to force a connection. I want to resist that, because the connection here is not forced — it is structural, and it runs through the word itself.
In this industry, an oracle is a machine that takes a fact from the outside world and puts it inside a system that cannot see the outside world. A price feed. A weather reading. A settlement figure. The whole point of an oracle is that it bridges the unverifiable and the verifiable, and the whole unsolved problem of an oracle is that it requires you to trust something that lives outside the trust boundary.
The company is named for the same idea, for the same reason: a source of truth that other systems query. But the deeper resonance is not linguistic. It is architectural. Oracle the company is now selling a product — compute — whose defining property is that it is physically located, physically constrained, and physically verifiable only by the party that controls the location. And that is precisely the property that the decentralized compute movement claims to remove.
So the honest question is not whether Oracle's data centers will get built. The honest question is whether decentralized compute can deliver capacity faster, cheaper, and more reliably than a company doing exactly what decentralized compute promises to do, but with a centralized balance sheet.
Because if it cannot, then the decentralization is not the product. The compute is the product. And the compute is physical.
The Mirror Image: DePIN and the Same Six Words
Let me describe the decentralized version of this business, because I have audited enough of it to know its shape.
A DePIN compute network raises capital — often through a token rather than a loan — and uses it to incentivize hardware providers to join a network. The providers contribute GPUs, storage, bandwidth, or inference capacity. The network aggregates that capacity and resells it to buyers who need compute. The token pays the providers, aligns them with the network, and, in theory, captures the value of the aggregate.
The pitch is beautiful. The pitch is always beautiful. And the pitch has a specific claim buried in it: that because the hardware is owned by many participants instead of one, the network can grow faster and cheaper than a hyperscaler, and can therefore deliver capacity at a lower price.
I want to test that claim against the two words from the Oracle note, because they are the same two words that will decide whether DePIN compute survives the next eighteen months.
First, capacity delivery. A DePIN network does not order GPUs. It convinces strangers to buy GPUs and plug them in. That sounds like a growth advantage until you remember what enterprise compute buyers actually require. They require a service level agreement with teeth. They require deterministic latency, contractual uptime, physical security of the workload, and a counterparty they can sue. A decentralized network of heterogeneous hardware, contributed by anonymous operators across residential and commercial internet connections, can satisfy exactly none of those requirements at the top of the market. It can satisfy some of them at the bottom, where the workloads are embarrassingly parallel, interruption-tolerant, and price-sensitive.
That is the trap. The decentralized network wins the workloads that pay least, and loses the workloads that pay most. It is structurally a spot market for compute, not a contracted capacity provider. And spot markets, by nature, do not have capacity delivery problems — they have capacity quality problems, which is a quieter and more persistent version of the same disease.
Second, maintain margin. Here the DePIN story is even more exposed than Oracle's, because a token-funded network has a cost structure that a balance-sheet-funded company does not: it must continuously pay its suppliers in an asset whose price the network does not control. When the token rises, the network overpays for hardware that a competitor procures at fixed prices. When the token falls, the suppliers leave, and capacity evaporates at exactly the moment the network needs to prove reliability. Oracle's margin is threatened by depreciation and electricity. A DePIN network's margin is threatened by its own incentive curve.
The subsidy is the product. That is the sentence I keep returning to, because it is the one that most blockchain founders never write down. In liquidity mining, the APY is the project paying you to pretend to be a user. In DePIN compute, the emissions are the project paying a stranger to pretend to be a data center. Stop the emissions and watch what remains — because that residual is the only part of the business that was ever real.
Code Doesn't Make Electrons
The romance of decentralized compute is the romance of coordination. The idea that thousands of independent actors, bound by a smart contract and a shared token, can outperform a centralized firm is not absurd. It is the founding intuition of this entire industry, and I share it in principle. Coordination is real. Incentive design is real. Removing middlemen from markets that never needed them is real value.
But coordination does not generate power. Code does not make electrons. A smart contract cannot cool a rack. A DAO cannot negotiate an interconnection agreement with a regional grid operator, and it cannot put a substation where the substation needs to be, and it cannot sign a twenty-year power purchase agreement with an investment-grade counterparty, because it has no legal personality with which to sign anything.
This is the boundary that the last several years of DePIN enthusiasm has been quietly bumping into. The decentralization is genuine at the layer of coordination and settlement. It is fiction at the layer of physical siting and utility contracting. And value in the compute business accrues to whoever controls the physical layer, because that is where scarcity actually lives.
I watched this pattern once before. In 2020, at the height of DeFi Summer, I spent weeks inside the contracts of a high-yield farming system, and I found a reentrancy vulnerability that could have drained roughly five million dollars. The community around it was celebrating an annualized yield that made no sense to anyone who looked at where the yield came from. The yield came from new deposits. The new deposits came from the yield. It was a beautifully engineered machine for producing a number, and the number was not a return — it was a subsidy wearing the costume of a return.
I wrote about it then, and I was not popular for it. I am not trying to be popular now. I am trying to be accurate. The mechanism is identical. A token-funded compute network that cannot articulate where its margin comes from after emissions end is a yield farm with GPU fans.
The Financing Framework Is the Whole Story
The most revealing line in the entire note — and the one a casual reader will scroll past — is the claim that the financing framework has become clearer. Read that sentence again and notice what it is actually saying. It is not saying the company found more money. It is saying that the question of where the money comes from has finally been answered. Which means the question of where the money comes from was, until recently, the thing the market was most afraid to ask.
That is a profound admission, and it generalizes far beyond Oracle. In any capital-intensive expansion, the financing path is the story, because the financing path determines who bears the risk and who captures the upside. A company funded by retained earnings dilutes no one. A company funded by debt dilutes no one today and dilutes everyone tomorrow if the return on the borrowed capital underwhelms. A company funded by equity issuance dilutes immediately and transfers the future to the new holders.
A company funded by a token does something stranger than all three. It creates an instrument that behaves like equity in a bull market and like nothing at all in a bear market, because it has no claim on cash flow, no lien on assets, and no claim in insolvency. Holders did not buy ownership. They bought a claim on the network's continued willingness to emit. And the network's willingness to emit is a governance decision, not a contractual right.
I sat on the other side of this in 2024, quietly advising a family office in Abu Dhabi through their first allocation into this asset class. Ten million dollars, structured carefully, with an insistence that the portfolio carry privacy-focused and infrastructure positions alongside the obvious assets. What struck me during those months was how the traditional allocators thought about risk. They did not ask about yield. They asked who signs the contract, who holds the collateral, who has standing in a court, and what happens to their money if the counterparty simply stops answering. Every one of those questions is either unanswerable or answered only by trust when the instrument is a token.
Decentralization removes the counterparty. It does not remove the question. It merely makes the question invisible, which is convenient in a bull market and catastrophic in anything else.
The Margin Signal Nobody Wants to Read
The note lists maintaining margin as a key variable. That is a confession in plain language, and it deserves to be treated as one.
A company that is scaling into a genuinely good business rarely has to reassure the market that margins will hold. The reassurance is a signal that the market already suspects the margin will not hold, and that management knows the market suspects it. When a firm moves from licensing high-margin software to selling capital-intensive infrastructure, the gross margin arithmetic moves against it in a way that no amount of narrative can reverse. Software is nearly free at the margin. A megawatt is not. A GPU is not. A building full of them is emphatically not.
This is not a reason the transition is wrong. It may be entirely the correct strategy. But it means the company is trading present margin for future scale, and the trade only pays if the future scale arrives at a price that compensates for the margin given up. Every lens I have for evaluating this — from a software auditor's eye to a token economist's eye — lands on the same conclusion. The transition is a bet, the bet is levered, and the bet's payoff is deferred beyond the horizon of most people doing the pricing.
A margin that must be defended is a margin that is under attack. Oracle is defending it. Every token-funded compute network is defending it in a different way, by pointing at growth instead of profit. Both are telling you the same thing without meaning to.
Customer Concentration: The Whale Problem in Enterprise Clothing
The one dimension the note does not name but the structure implies is concentration. When the growth engine is a handful of enormous AI contracts, the revenue is visible but fragile. The customer that can make your quarter can also break it, and it can do so by adjusting a procurement schedule rather than by walking away.
Blockchain natives should recognize this instantly. It is the whale problem. A protocol whose total value locked is dominated by three addresses is not a protocol with three hundred users — it is a protocol with three users and a very confident presentation. Enterprise software has the same disease with better tailoring. A backlog figure that looks magnificent is a promise that the largest counterparties will keep buying at the pace the model assumes. Concentrate that backlog and you have not de-risked the business. You have renamed the risk.
The DePIN parallel is exact. A compute network whose inference demand comes from one model lab is not a market. It is a single bilateral contract with a token and a block explorer attached. When that lab builds its own capacity — and they all eventually do — the network discovers that its demand was never a market at all. It was a customer. And customers leave.
The Verification Gap
Here is where I think the crypto industry can actually learn something from the Oracle situation rather than merely comment on it, and where my own work has moved.
Last year I worked with a small team on a standard we called Proof of Human Intent — cryptographic signatures that let a piece of digital work carry verifiable evidence of human authorship. The motivation was not sentimentality about art. It was the recognition that as generative systems scale, the scarce resource stops being the generation and becomes the verification. Anyone can produce output. Almost no one can produce output with a verifiable chain of provenance that survives adversarial scrutiny.
The same shift is happening in compute. The scarce resource is not going to be raw FLOPs for very long. The world is building FLOPs at a rate that will eventually make them a commodity, and the moment they are a commodity, the scarce resource becomes the verifiable claim about where the compute happened, under what conditions, with what guarantees, and at what cost. Oracle is selling that claim implicitly in the form of an SLA. DePIN networks are trying to sell that claim cryptographically. Both are selling the same product, and only one of them has a legal counterparty.
My prediction is that the winners in decentralized compute will not be the networks that promise the most hardware. They will be the networks that solve verification — that can prove to a skeptical buyer, without trusting the operator, that a specific workload ran on specific hardware under specific conditions. That is a cryptographic problem, and it is genuinely solvable. Throughput and capacity coordination are not the unsolvable parts — the unsolvable parts are power, land, and interconnection, and those do not care about your consensus mechanism.
The Contrarian Test
So let me apply the pragmatism test that I owe every reader, and state the thing that is least popular to state.
The most common belief in decentralized compute is that it will eventually undercut centralized hyperscalers on price because it avoids the capital burden of owning data centers. I think that belief is backwards.
Avoiding the capital burden does not avoid the cost. It relocates the cost to the token holder, who pays for the hardware indirectly through emissions and dilution, and who has no claim on the resulting asset. When the emissions end, the hardware providers leave, because they never had equity in the network — they had a yield. And what remains is a market where the winners are whoever locked in the cheapest power and the densest interconnection, which is to say: exactly the incumbents the network claimed to disrupt.
Decentralization relocates scarcity. It does not dissolve it. The scarcity in computing does not live in the coordination layer, where blockchains genuinely excel. It lives in the interconnect queue at a regional grid operator and in a twenty-year power contract signed by a legal entity, and no amount of clever incentive design changes that. The networks that survive the next cycle will be the ones honest enough to admit this, and to design for a world where they are the flexible spot layer beneath the contracted capacity layer, rather than the fantasy of replacing it.
This is a corrective, not an obituary. The correction is that we should stop pricing decentralized infrastructure as if it were a substitute for capital, and start pricing it as what it is — a coordination technology sitting on top of a physical business that obeys physical rules. The blindness worth naming is the industry's refusal to distinguish between a mechanism and a moat. A token is a mechanism. A grid connection is a moat.
What the Note Actually Tells Us
The most useful thing about reading a thin disclosure is how much it forces you to hold open. There is no revenue figure. There is no capex figure. There is no backlog figure. There is no timeline. Because of that, every claim about the company — including the confident ones in this essay — carries a lower confidence than the confident tone implies, and I want that noted rather than hidden. This is diligence under scarcity, not judgment under disclosure. If the full financials arrive and they show capex converting into revenue ahead of schedule and operating margin stabilizing, the whole picture moves from unproven to proven in a single quarter. If they show the opposite, the story moves from unproven to cautionary just as fast.
The same is true of every DePIN network in your portfolio, with one important difference. Oracle's financials will eventually be disclosed to the public and audited by parties with real liability. The networks will report what the networks choose to report, verified by parties whose downside is a damaged reputation inside a community that rewards enthusiasm. That asymmetry is the thing I would price most heavily of all.
The Two Oracles Converging
Step back and the picture is stranger and more interesting than either story alone.
A fifty-year-old software company is being valued on whether it can pour concrete and secure megawatts. A generation of blockchain networks is being valued on whether it can convince strangers to plug in hardware it does not own. Both are betting on the identical demand: unquenchable, insatiable, model-scale machine computation. Both are constrained by the identical physics: power, land, and interconnection, in that order, none of which respond to incentives.
And both are, at bottom, oracles in the original sense — systems that must pull a truth from the physical world into a system that cannot verify it. The company does it through a service agreement. The network does it through a signature. Neither is lying. Both are asking the same question of the future: when compute becomes the substrate under everything, who gets to certify that the computation happened the way it was claimed?
The company that answers that question legally and the network that answers it cryptographically will, I suspect, end up less as competitors than as a two-layer architecture — a contracted base and a verifiable spot layer above it. The trouble is that both currently believe they are the whole building.
Trust the Protocol, Not the Pitch
Trust the protocol, not the pitch. The protocol here is not a blockchain. The protocol is the physical and financial reality that governs every compute business ever built: capacity must be delivered before it can be billed, margins must be defended before they can be reported, and capital must be raised before it can be returned. A $210 target and a market-perform rating are not opinions about the future. They are the market's way of saying that the future has been described but not yet verified — and the market, whatever else it is, is a machine that has learned to wait.
So the real question is not whether Oracle builds it, or whether the decentralized networks build it. The real question is who builds the infrastructure that lets a skeptical buyer verify that what was promised was actually delivered, without having to trust the seller, the operator, or the analyst who wrote the note. Compute has become the most demanded input in the world. Trust is still the scarcest. Whoever engineers that — legally or cryptographically — will own the next decade, and it will not be the person with the loudest pitch.