"One hundred megawatts cannot be shipped."
That sentence arrived in my inbox at 2:14 a.m. Chengdu time, forwarded by three different people, each with a different subject line. The speaker was Alexis Cathalifaud, chief executive of a company called Clichmont, and the document was framed as a founder's reasoning about a strategic bet — not a product launch, not a funding round, not the token announcement I had braced for by page three.
I read it twice. Then I read it a third time with a different question in my head. Not "is this good?" but "where is the blockchain?"
There is a token. It is called $CLAI. It appears as the third of three discussion topics, inside a document that spends generous space on land, interconnection queues, and Nordic hydroelectric capacity. The token gets roughly one sentence of oxygen.
That asymmetry is the story. Not the megawatts. The silence around the token.
For eighteen months, the AI trade has been migrating. It moved from software to silicon, and then from silicon to something far less glamorous: electricity. The neoclouds — CoreWeave, Lambda, Crusoe — were the first to price that shift, building businesses not on model architecture but on the unglamorous work of finding GPUs, racks, and enough amperage to keep them alive. Their rise proved something the market had been slow to accept: the binding constraint on artificial intelligence stopped being cleverness some time ago and became capacity.
Clichmont's argument extends that logic one layer down. The company describes itself as focused on owning and controlling the physical infrastructure that AI hardware requires — land, power, cooling, shells — rather than renting it. Two sites have been confirmed publicly: Alicante, in Spain, where the sun is abundant and grid interconnection is a political process; and Bodø, in Norway, where renewable capacity and cold air do much of the engineering work for free.
The framing is deliberate. Alicante for solar. Bodø for hydro and latitude. Both fall inside the European Union, which means both sit under MiCA, the bloc's crypto asset rulebook, whether the company likes it or not.
Beneath the concrete sits $CLAI, described in the deck's own words as one of three topics — alongside the CEO's case for why power and data center infrastructure may prove the more durable bottleneck. In a market where AI-adjacent tokens have re-rated faster than the assets they claim to represent, that is either humility or a hole. In my experience it is rarely both.
The physical thesis is stronger than the token thesis
Start with what is genuinely good, because there is a real insight buried here and it deserves to be separated from the packaging.
The single most useful idea in the document is the depreciation asymmetry. A GPU has an economic life measured in a few years before a new generation resets the price-performance frontier underneath it. A substation, an interconnection agreement, a cooling plant, and a slab of land have lives measured in decades. Financing a thirty-year asset with four-year tenants is an uncomfortable thing to do, but owning the thirty-year asset is a fundamentally different business from owning the four-year one — and the market has spent two years conflating the two.
GPU rental is a spread business. Infrastructure ownership is a duration business. The first lives and dies on spot pricing and utilization rates; the second lives and dies on the cost of capital and the length of its contracts.
The second useful idea is that interconnection rights have quietly become a scarce asset class. In several European markets, the queue to connect a new large load runs years, not months. That means the moat is not the silicon. It is the permitting. You cannot download a grid connection. You cannot fork a substation. When Cathalifaud says chips can be shipped anywhere but a hundred megawatts cannot, he is making a statement about geography as an asset class, and it is correct.
I have watched this pattern before. In 2020, when I led the volunteer audit team that reviewed OpenYield's flash loan module, the vulnerability we found was not exotic. It was a reentrancy bug — a classic, decades-old, entirely preventable flaw. What made it dangerous was not its sophistication but the fact that thousands of users had already assumed somebody else had checked. Physical infrastructure works the same way. The risk in a data center is rarely the clever part. It is the part everyone assumed was handled.
Where the blockchain goes missing
Here is where my third read became uncomfortable.
I went looking for the on-chain layer and did not find one. No smart contract addresses. No audit report. No disclosed mechanism for how a token interacts with a megawatt. No vault, no staking module, no settlement rail, no oracle design, no custody arrangement.
To be fair, the document never claims otherwise. It is a strategy note, not a technical paper. But that is exactly the point: Clichmont's "technology" is operational, not cryptographic. It is heavy-asset management with a token attached, and the attachment is described in a sentence rather than a section.
I have a rule I built the hard way, and I repeat it often: code is law, but humans are the protocol. That rule cuts in both directions. When a project has no code to examine, the humans carry the entire burden — every promise, every allocation, every unlock. There is nothing to verify and therefore everything to trust. And trust is earned in drops, lost in buckets.
I am not accusing anyone of bad faith. I am describing the state of available evidence, which is close to zero.
The token economics are a black box
This is where the analysis has to be honest about its own limits, because the data simply is not there.
No supply schedule. No maximum supply. No team allocation. No early investor allocation. No community or liquidity allocation. No treasury policy. No unlock calendar. No disclosed emission rate, no staking design, no stated yield, no revenue sharing, no buyback, no burn.
Four allocation categories — team, early investors, community, treasury — and every cell reads the same: not disclosed. That is not a gap in the deck. That is the entire financial picture of the token, missing.
The question that matters most is the one nobody asked in the write-up: does the token have a compulsory use? Not an encouraged use, not a governance vote, not a points program with a token wrapper — a use that a rational operator cannot avoid if they want to participate in the network. If the answer is no, then $CLAI is a claim on narrative rather than a claim on cash flow, and its price will be set entirely by expectation.
The value capture question compounds it. Physical infrastructure produces revenue the old-fashioned way: kilowatt-hours sold, rack space leased, contracts signed. Whether any of that revenue touches a token depends on a legal instrument — a claim, a share, a contractual right. RWA is not a technology. It is a legal wrapper with a technical delivery mechanism. If the wrapper is absent, what remains is a ticker.
A team of one name
Full disclosure of my own priors: I check the human layer first now, and I check it hardest when the technical layer is thin.
The document mentions exactly one person. Alexis Cathalifaud, chief executive. No co-founders, no chief technology officer, no commercial leadership, no board, no advisors, no code history, no prior ventures, no biographies. The French surname suggests a European base, possibly France, which would sit sensibly alongside the Spanish and Norwegian sites. That is an inference, not a fact.
There is no governance model disclosed either. No multisig, no on-chain voting, no treasury committee, no decision framework. No investors, no round sizes, no valuations, no lockups. For a Web3 project, that combination is a warning light — not yet a red flag, but the difference between an early-stage company and an unverifiable one.
The securities question, for what it is worth, remains genuinely open. Money invested, likely. A common enterprise, plausible. Reliance on the efforts of others, highly likely given the asset-heavy model. Expectation of profit, undisclosed — and that undisclosed element is doing a great deal of work. In a MiCA jurisdiction, classification is not academic. It determines what a team may legally say, where it may list, and who may buy.
The competition will not wait
Clichmont is entering a race that already has leaders, and the chief executive essentially concedes it: CoreWeave, Crusoe, and Lambda "proved AI compute is a huge market." That is a gracious sentence and also an admission of who arrived first.
The differentiation is real but narrow. CoreWeave built a leasing model at scale. Crusoe built a distinctive energy story out of stranded gas. Clichmont's angle is ownership, ESG-aligned geography, and duration. That is a coherent thesis. It is also a thesis that requires an enormous amount of capital expenditure before the first euro of revenue, on a timeline that rewards patience and punishes leverage.
Construction delays are not a tail risk in this business. They are the base case.
The contrarian read: this is a maturity mismatch wearing a token
Everyone who writes about this will say the same thing: no token mechanics equals red flag. I think that is the shallow read.
The deeper problem is structural, and it is not about honesty. It is a duration mismatch between a fifteen-to-thirty-year physical asset and a speculative instrument whose holders measure time in weeks. The people who fund a substation and the people who buy a liquid token do not share a time horizon, a risk appetite, or a definition of "working." One group wants a power purchase agreement. The other wants a candle.
That mismatch does not resolve through better tokenomics. It resolves — or fails to resolve — through legal engineering that most infrastructure companies are not built to do.
And here is the genuinely counterintuitive part: the omission may be the most responsible thing in the document. Everyone else mints first and pours concrete later. This team poured concrete first. Hold through the noise, build through the silence — the silence here is literal, and there is at least a case that a token with no published mechanics is better than a token with invented ones.
But a token that exists without a job is not neutral. It is a liability with a marketing budget.
What to watch, and what it would mean
I am not going to pretend I can price this. What I can do is name the evidence that would change my read.
A published token paper with a real allocation table and a stated value-capture mechanism would move this from narrative to something analyzable. An announced first site going live with signed offtake would validate the heavy-asset thesis independent of any token. A named, complete founding team with verifiable history would change the risk profile more than any technical disclosure. And a MiCA compliance posture, stated plainly, would tell us whether the team intends to operate in daylight or at the edge of it.
Until then, what exists is a strong physical argument, two confirmed sites on a map, one executive name, and a ticker.
Education is the antidote to exploitation — and in a cycle where AI tokens trade on adjectives, the willingness to say "I do not know yet" is itself a form of discipline.
So here is the question I keep turning over, and I will leave it with you: if the power is real, the land is real, and the interconnection queue is real, does the token need to exist at all — or is it there because this market refuses to fund anything until a ticker exists?
From winter's cold, spring's structure emerges. The structure here is concrete. The spring, if it comes, will be measured in megawatts, not market capitalization.
