What is a neocloud?

A neocloud is an operator whose business is buying accelerated computing hardware at scale and selling access to it. The fleet is inventory rather than infrastructure supporting something else, and the revenue comes from third parties.

The term is most usefully defined by what it is adjacent to.

Not a hyperscaler. The large cloud platforms operate compute alongside diversified businesses that generate cash independently of it. A downturn in accelerated computing is absorbed by the rest of the enterprise. A neocloud has no rest of the enterprise.

Not a colocation provider. A colocation business sells space, power and cooling, and its customers own the equipment inside. Its revenue is contracted against real estate with a long life and a broad resale market. A neocloud owns the equipment and sells its output, which is a different asset and a much shorter one.

Not an enterprise cluster owner. An organisation building compute for its own workloads has no external contract to underwrite and is assessed on its own balance sheet — a distinction that propagates through every subsequent question and is worked through in [financing a neocloud against an enterprise cluster](/compute/financing-a-neocloud-vs-an-enterprise-cluster).

What remains is an infrastructure operating business with an unusual profile: capital-intensive like a utility, contracted like a services company, holding an asset that depreciates like consumer electronics. It is financeable — as an operating business, which is more demanding than most sponsors expect.

Whether you are lending to a neocloud, leasing equipment to one, or letting it a building, the questions are the same and the order is stable.

The contract book is the credit. With no parent balance sheet behind it, what a neocloud has is its contracts. Firm commitments from counterparties that can pay are the structure; a book of consumption arrangements with no minimums is a much weaker proposition wearing similar language. The distinction between a firm obligation and a right to consume is the substance of [compute offtake as credit](/compute/compute-offtake-as-credit).

Concentration is usually severe, and often undisclosed in its true form. A small number of customers is normal at this stage of the market. What matters is whether the largest is related to the sponsor, whether it can pay through a downturn, and whether the second-largest would still be there without the first.

Tenor against equipment life. Contracts that run for a fraction of the hardware's financed life leave a recontracting assumption sitting in the middle of the structure. It may be a reasonable assumption; it is not a contract, and it should not be modelled as one.

Its position in the layers below it. A neocloud's power and space arrangements must run at least as long as its customer commitments. An operator that has sold three years of capacity from a facility it holds for two has an exposure that will surface — and it is the landlord's problem as much as the lender's.

Equity behind the fleet, and where the last dollar came from. Compute businesses grow by buying more equipment, continuously. The terms of the most recent capital, and whether it was equipment-level or corporate, tell you more about the position than the growth rate does.

Every counterparty has risks. What distinguishes a neocloud is that its risks are not independent of one another.

A diversified business fails one part at a time — a customer is lost while the balance sheet holds, or margins compress while the asset base retains value. A neocloud's revenue, its collateral, its refinancing capacity and its exit are all functions of a single variable: the market price of accelerated compute.

When that price falls, four things happen at once.

• Contracted revenue does not fall, and recontracted revenue does. The book reprices as it rolls, and the rate at which it rolls is set by the tenor of the original contracts.
• Utilisation falls. Merchant capacity that was clearing sits idle, which is where the operating leverage bites first.
• The fleet's resale value falls. The collateral behind the facility is worth less at precisely the moment the borrower needs to realise it, because the buyers for a used fleet are other operators facing the same market.
• New capital becomes harder. The lenders and lessors who would refinance are marking the same asset class down.

Diversifying customers does not diversify this. Ten customers all buying compute are ten exposures to the same variable, and correlation among a neocloud's customers is typically higher than a customer list suggests. It is also why an asset that looks well secured on paper can strand while it is still running perfectly — the mechanism set out in [stranded assets](stranded-asset).

None of this makes the counterparty uncreditworthy. It makes the structure around it load-bearing: security taken at the level that owns the equipment, contract quality tested rather than counted, and a residual assumption that has been priced rather than inherited from the invoice.

The view from the site side is less discussed and, for anyone letting a powered shell, more immediately useful. A specialist compute operator is an attractive tenant — it takes capacity quickly, at scale, and it wants exactly what a data-center landlord has to sell. It is also a different credit from the diversified corporates the lease templates were written for.

The tests a landlord or facility lender should run:

• Lease term against the tenant's own contract book. A ten-year lease supported by a two-year contract book means the landlord holds the recontracting risk from year three onward, whatever the lease says. That is a position worth pricing rather than discovering.
• The security package. Deposits, letters of credit, parent guarantees where there is a parent worth guaranteeing. The instrument matters less than whether it is sized against the cost of re-letting rather than against a few months of rent.
• Power cost, and who bears its movement. Compute economics are sensitive to power price. A lease that fixes the tenant's power cost puts that exposure on the landlord; one that passes it through puts pressure on a tenant already exposed to a correlated market.
• Specification, and what happens on default. A shell built tightly around one operator's configuration has, in practice, one plausible occupier — so the landlord holds that operator's credit for the life of the building whatever the lease says. Building something re-lettable is generally worth more than the rent premium a bespoke fit-out earns, for the reasons set out in [powered shell](powered-shell).