Spokesperson: Alexis Cathalifaud, CEO
As build a matter to for synthetic intelligence compute continues to develop, the infrastructure supporting that build a matter to is becoming a strategic consideration in its believe loyal. Companies across the sector are racing to accurate secure entry to to increasingly more more highly effective GPUs, whereas questions around electricity, records-heart skill, cooling and connectivity are becoming more challenging to ruin free the compute itself.
Clichmont is taking a assorted manner. In residing of establishing its model essentially around rented GPU skill, the firm is centered on owning and controlling the physical infrastructure on which successive generations of AI hardware can purpose. On this interview, Clichmont CEO Alexis Cathalifaud discusses why the firm believes vitality and records-heart infrastructure may perchance well maybe turn out to be the more sturdy bottlenecks, the contrivance it approaches home need and the challenges of scaling physical infrastructure, as successfully as the position of its $CLAI token within the broader ecosystem.
1) Every firm in this category is fighting over GPU secure entry to loyal now. Clichmont’s resolution is to produce the records centers comparatively than renting the chips. Why does possession matter bigger than secure entry to?
On fable of GPU secure entry to offers you compute; infrastructure possession offers you take care of a watch on over the economics of compute.
For a firm delight in Clichmont, owning or controlling the records-heart layer can matter more strategically than simply securing rented GPUs. If you happen to rent GPU skill from a hyperscaler or GPU cloud, you inherit any individual else’s pricing, availability, vitality constraints, networking structure, deployment schedule, and margins. When build a matter to spikes, secure entry to can turn out to be expensive or constrained.
Owning the infrastructure changes the equation. Clichmont can doubtlessly resolve which GPUs to deploy, when to fortify them, how densely to set up them, how vitality and cooling are engineered, and how the skill is commercialized. The same facility can furthermore evolve from one GPU generation to the next in preference to tying the synthetic thesis to a particular chip.
There’s one other crucial distinction: GPUs depreciate like a flash; vitality-ready records-heart skill is a longer-lived strategic asset. A GPU generation also can turn out to be economically less competitive within a few years, whereas land, grid connections, substations, cooling infrastructure, fiber connectivity and licensed megawatts can remain precious across multiple generations of accelerators.
That makes the scarce helpful resource increasingly more more no longer loyal the GPU itself, nonetheless the flexibility to energise hundreds of GPUs at scale. A firm can possess chips and tranquil believe nowhere exact to deploy them. Securing 10,000 GPUs is one subject; securing the tens of megawatts of legitimate electricity, cooling and network infrastructure required to purpose them is one other.
2) You’re up in opposition to companies that are already public or heading there – CoreWeave, Crusoe, Lambda. What attain you suppose their model will get scandalous, if anything?
I don’t believe CoreWeave, Crusoe or Lambda purchased the model scandalous. They proved that AI compute is a wide market. Where we vary is in what we mediate will remain scarce. GPUs commerce every generation. The sturdy bottleneck is the infrastructure required to run them — vitality, land, cooling and connectivity. Clichmont’s thesis is that in preference to competing easiest to rent doubtlessly the most up-to-date GPU, we prefer to occupy up a watch on the infrastructure on which successive generations of GPUs will purpose. In a market the save each person is chasing chips, we’d comparatively believe the residing the save the chips prefer to dwell
3) There’s a increasing argument that vitality, no longer chips, is the exact bottleneck for AI infrastructure. How necessary does that shape the save and how Clichmont builds?
Power shapes almost every infrastructure resolution we create. A GPU without legitimate vitality is correct expensive hardware sitting in a rack. We mediate the exact competition over the next decade won’t simply be for GPUs—that is also for megawatts.
So when Clichmont evaluates a home, we don’t begin by asking the save we can web the least expensive building. We build a matter to: the save can we accurate legitimate vitality, at the loyal economics, with the flexibility to scale? What’s the time-to-vitality? What’s the grid subject? What cooling structure does the local weather enable? And can that home pork up the next generation of GPUs, no longer loyal the ones we’re installing this present day?
That’s one motive locations with solid vitality fundamentals are strategically attention-grabbing to us. Chips also can furthermore be shipped across the realm. You would’t ship 100 megawatts. The compute one contrivance or the opposite has to plod the save the vitality is.
So I wouldn’t advise chips discontinuance being a bottleneck. They proceed to be serious. But increasingly more more, owning GPUs isn’t ample. The competitive advantage is being in a save to vitality, frigid and performance them economically at scale. That’s what we’re building Clichmont around.
4) Clichmont’s sites differ from a photograph voltaic-powered facility in Alicante to a brand fresh produce in Bodo, Norway. What basically decides the save a records heart will get built – is it about vitality, land, local weather, one thing else?
We don’t resolve a location because one variable appears to be stunning. We resolve it because your total infrastructure equation works.
Power is the first filter: how many megawatts can we accurate, at what tag, how legitimate is that offer, and—seriously—how like a flash can it basically be delivered? Then we look for at cooling, local weather, fiber connectivity, land, allowing, safety and the flexibility to create bigger.
Bodø and Alicante are attention-grabbing precisely because they symbolize assorted strengths. Northern Norway offers us a neighborhood weather that can pork up ambiance pleasant cooling and a solid vitality ambiance. Alicante offers us a assorted vitality profile and the different to combine photo voltaic into the infrastructure approach. We don’t mediate every Clichmont records heart needs to plod making an attempt to web same—the structure must tranquil respond to the sources of the home.
And land by itself isn’t significantly precious to us. An economical parcel and not utilizing a scalable vitality or fiber is rarely any longer a records-heart home. What matters is whether we can turn that location into legitimate, economically competitive compute skill.
In the extinguish, we’re no longer basically taking a look for for land. We’re taking a look for for locations the save vitality, connectivity, cooling and scalability converge. That’s the save we produce.
5) Right here is an infrastructure firm with a token linked to it. For a reader who’s skeptical of that aggregate, what’s the beautiful case for why $CLAI exists at all?
The skeptical rate is fully stunning. A token shouldn’t exist loyal because a firm operates in AI. If $CLAI were simply a financing wrapper around our records centers, I wouldn’t take notice of that a compelling motive to produce it.
Clichmont is the infrastructure substitute. It builds and operates compute skill. $CLAI is intended to be a digital financial layer across the broader ecosystem — one thing that can at last pork up on-chain participation, treasury assignment and neighborhood governance in ways in which faded fairness isn’t designed to attain.
And we prefer to construct the loyal to create that distinction. The physical infrastructure has to exist independently of the token, and the token has to display exact utility independently of hypothesis. If we can’t level to each and every, then the skepticism is justified.
So I wouldn’t build a matter to anybody to mediate in $CLAI simply because Clichmont owns GPUs or builds records centers. The check is necessary simpler: does the token at last attain one thing priceless, transparent and measurable that couldn’t be performed as successfully with a same outdated database or faded corporate structure? That’s the same outdated we believe to be held to.
6) What’s the hardest part of scaling physical infrastructure that other folks who’ve easiest built tool are inclined to underestimate?
The hardest part is that physical infrastructure doesn’t scale at tool rush. In tool, if build a matter to doubles, you furthermore mght can on the total provision more skill like a flash. In a records heart, every extra megawatt has a physical dependency at the assist of it — grid skill, transformers, switchgear, cooling, fiber, permits, building and one contrivance or the opposite hardware.
And these dependencies don’t plod in parallel as neatly as other folks imagine. You would believe the land and no longer believe the vitality. You would believe the vitality allocation and wait months for electrical tools. You would believe the building ready and tranquil be making an attempt ahead to a grid connection. One missing side can lengthen a total deployment.
The different distinction is that mistakes are expensive and bright to reverse. Machine also can furthermore be patched overnight. You would’t patch a badly designed 50-megawatt electrical machine overnight. You’re making capital choices this present day in accordance to what GPUs, vitality densities and cooling requirements also can look for delight in every years from now.
So the exact skill isn’t simply building records centers. It’s sequencing capital, vitality, building and customer build a matter to so that they arrive at roughly the same 2d. Assemble too early and it’s most likely you’ll even believe expensive slothful infrastructure. Assemble too slack and the shopper goes somewhere else.
That execution self-discipline is perchance what other folks coming purely from tool underestimate most. In physical AI infrastructure, rush matters — nonetheless timing matters even more.
7) If you happen to had to call doubtlessly the most attention-grabbing chance in having a bet on a produce-it-your self model comparatively than a capital-light condo model, what would it be?
The most attention-grabbing chance is capital intensity mixed with timing. If you happen to produce infrastructure your self, you’re committing well-known capital this present day in opposition to assumptions about build a matter to, vitality economics and skills several years into the future.
A condo model offers you flexibility. If the market changes, you furthermore mght can nick skill, plod providers or undertake the next generation of hardware. If you happen to believe the infrastructure, you don’t believe that luxury. A substation, cooling machine or records-heart building is a long-duration resolution.
For us, doubtlessly the most attention-grabbing hazard due to this truth isn’t simply spending too necessary — it’s building the scandalous skill, in the scandalous residing, at the scandalous time. If you happen to produce ahead of build a matter to, capital sits slothful. If you happen to produce too slowly, you miss the market.
That’s why we don’t rate possession as ‘produce everything ourselves.’ The aim is to occupy up a watch on the strategic infrastructure whereas ideal versatile around skills. The building, vitality, cooling and connectivity must tranquil continue to exist multiple generations of GPUs in preference to becoming dependent on one hardware cycle.
So yes, the capital-light model has a exact advantage: optionality. Our bet is that if we offer out accurately, giving up some quick-term optionality creates one thing more precious over the future — take care of a watch on over skill, vitality economics and the physical infrastructure that AI increasingly more more relies on.
8) Three years from now, the save attain you need Clichmont to take a seat relative to the CoreWeaves and Nebiuses of the realm?
Three years from now, I don’t demand Clichmont to be doubtlessly the most attention-grabbing firm in the category, and that’s no longer the aim. CoreWeave and Nebius believe enormous scale and secure entry to to capital. Looking out to replicate them may perchance well maybe be the scandalous approach for us.
I need Clichmont to be identified as one of doubtlessly the most ambiance pleasant honest AI infrastructure operators in Europe — with exact working sources, secured vitality, excessive-density GPU skill and a notice file of bringing fresh compute online like a flash.
Our advantage has to arrive from being disciplined about the save we produce and what we believe. We need locations the save the vitality economics create sense, infrastructure designed around successive generations of accelerated computing, and the flexibility to serve intelligent in AI, HPC and deepest compute in preference to simply competing for GPU condo quantity.”
If CoreWeave and Nebius are building hyperscale AI clouds, Clichmont can beget a assorted residing: a centered owner and operator of compute-ready infrastructure in strategically selected markets.
Conclusion
Clichmont’s approach one contrivance or the opposite comes all of the manner down to a long-term infrastructure bet: that secure entry to to GPUs will remain crucial, nonetheless the flexibility to vitality, frigid, join and performance these GPUs successfully at scale will turn out to be an increasingly more more precious advantage.
That manner comes with meaningful substitute-offs. Constructing physical infrastructure requires immense capital, long planning horizons and careful coordination between vitality, building, hardware and build a matter to. Clichmont’s thesis is that accepting these constraints can present increased take care of a watch on over the infrastructure required for successive generations of AI compute. Whether or no longer that thesis proves out will depend less on the ambition of the model than on the firm’s skill to occupy up out it successfully and at the loyal time.
