Comparing neocloud business models

Treat the directory as a discovery tool

American Compute’s October 2026 directory identifies 536 GPU-capacity providers across a broad set of business models. Its coverage includes dedicated GPU operators and organisations adding compute to an existing business. The report explicitly says it is not a census. It also retains 66 entries for which its survey found no public GPU offering. These are useful research leads, rather than 536 independently confirmed active neoclouds.

That distinction matters when describing market breadth. Operator count, installed capacity, revenue and financeable collateral measure different things. A small host and a large contracted platform each add one company to a directory; they do not represent equal market shares. Changes in directory coverage can also resemble industry growth without any new capacity being built.

Classify the economic activity

For credit analysis, begin with what the entity sells. A GPU operator earns compute revenue and carries hardware-refresh exposure. A colocation landlord earns contracted site revenue and carries development, tenant and infrastructure risk. A marketplace may match buyers and sellers without owning the financed equipment. A diversified telecom group can combine several models. Preserve these activities as attributes instead of forcing every participant into one mutually exclusive label.

Next identify the borrower and its perimeter. A recognizable parent name may sit above a limited-recourse project vehicle. Map guarantees, cash movement and competing creditor claims to the actual legal entities. Then identify which customers pay the financed vehicle and how much revenue depends on each one. Group borrowers by these exposures when comparing terms.

Show the missing denominators

The American Compute report has headcount information for 358 of 536 entries and disclosed funding totals for 63. Its funding-concentration findings describe that disclosed subset. They should not be relabelled as the concentration of all capital invested in the sector. Headquarters location also says little about where a company’s equipment, customers or security are located.

InfraCompute’s market view should therefore keep company, financing, project and customer relationships separate. Compare similar instruments and operating activities, label unknown fields and distinguish an absence of disclosure from zero borrowing. Use public transactions to study the visible financing frontier; use the wider operator universe to locate unanswered questions. Neither dataset provides an exhaustive market total on its own.

Turn segmentation into a research agenda

The practical next step is to examine operators between the largest public borrowers and small equipment-finance customers. Ask how they fund deposits, secure customers and bridge capacity additions. A well-supported case study of one financing structure can reveal more about that segment than an expanded directory with unresolved identities and no contract evidence.