Arqis
Solutions ▾
Neoclouds & GPU cloudsGPU supply and customer offtake Data-centre & AI infrastructure operatorsLeases, power and tenant concentration OEM & hardware suppliersDelivery, allocation and warranty exposure Banks, funds & private creditGPU-backed financing and covenants Infrastructure funds & investorsCross-border AI infrastructure capital Insurers & reinsurersStructured exposure and accumulation AI labs & compute buyersFrontier training and inference transactions
How Arqis is used Risk coverage Pricing Extensions Security
Client login
Risk coverage

Eight risks traditional insurance was not built for.

Eight risks sit inside every offtake transaction, and eight modules answer them on the same graph. This page is each one in full: what goes wrong, what we compute, the signals we watch and the levers we hand you.

Four covered todayOfftaker / Customer Risk, Compute Utilization / Demand Risk, GPU Residual Value / Obsolescence and Regulatory Deployment / Stranding Risk. Every tier above Pilot includes all four.
Four to comeBuilt on the same record, so nothing you enter now is re-entered when they arrive. Order follows what customers put under it first.
One record underneathEvery module writes to the same transaction record, with one methodology and one audit trail across all eight.
01Offtaker / Customer Risk 02Compute Utilization / Demand Risk 03GPU Residual Value / Obsolescence 04Compute Price / Revenue Risk 05Regulatory Deployment / Stranding Risk 06Contract Bankability Risk 07Technology Substitution Risk 08AI Efficiency / Compute Compression Risk
01

Offtaker / Customer Risk

Counterparty and demand
Covered today
What goes wrong

A customer weakens, defaults, downsizes or fails to renew after infrastructure has been committed around them.

How it reads in your world
Data centre
Anchor tenantThe tenant you built 300MW around downsizes or does not renew
GPU cloud
Customer credit and concentrationTwo customers carry most of the revenue, and one is venture-funded
What we compute
Exposure per customer, liveCommitted revenue and receivables against each customer, aggregated across every contract they sit under.
Deterioration, scoped to your transactionsA credit, ownership, funding or renewal change mapped only to the transactions that depend on it.
What protection you actually holdTake-or-pay, guarantees, collateral and caps already in your documents, and the gap where none exists.
Signals we watch
  • Credit files and rating actions
  • Ownership and funding changes
  • Revenue concentration by customer
  • Renewal and downsizing notices
  • Payment behaviour against terms
LeversTake-or-pay · parent guarantee · collateral · exposure cap · credit insurance
02

Compute Utilization / Demand Risk

Utilization and stranded capacity
Covered today
What goes wrong

You build the GPU or MW capacity, but actual demand and utilization fall below the level required to support the investment case.

How it reads in your world
Data centre
Tenant and stranded capacityYou build 300MW and the anchor tenant weakens, or never takes it
GPU cloud
Demand and idle capacityYou commit to 20,000 GPUs against demand contracted for a fraction of the term
What we compute
Contracted against availableCommitted capacity, take-or-pay floors and ramp schedules read against what is actually deliverable, by date.
Revenue at risk if demand is lateThe gap between contracted revenue and the capacity behind it, priced, not described.
Where the floor stops protecting youThe utilisation point below which the floor no longer covers fixed cost.
Signals we watch
  • Contracted vs available capacity
  • Take-or-pay and minimum-spend floors
  • Ramp and commencement schedules
  • Tenant or customer pipeline
  • Utilisation against plan
LeversCapacity reservation · utilization floor · anchor offtake · minimum revenue commitment · phased deployment
03

GPU Residual Value / Obsolescence

Asset value
Covered today
What goes wrong

The hardware still works, but newer chips or architectures reduce its earning power, resale value or useful economic life faster than expected.

How it reads in your world
Lender
Collateral and borrowing baseThe GPUs securing the facility lose value faster than the amortisation schedule assumes
GPU cloud
Earning power and refreshYour fleet still runs, but customers reprice it against the next generation
What we compute
A value curve per asset classExpected resale value and earning power by GPU generation, re-dated as new architectures are announced and shipped.
Collateral cover against the curveLoan-to-value and borrowing-base headroom recalculated as the curve moves, not at the next appraisal.
Where protection actually sitsBuyback, residual-value guarantee and refresh terms read from the contracts, with the point at which each one applies.
Signals we watch
  • OEM roadmaps and launch announcements
  • Secondary-market GPU pricing
  • Rental rates by GPU generation
  • Buyback and residual-value terms
  • Depreciation and amortisation schedules
LeversOEM buyback · residual-value guarantee · leasing · refresh rights · asset-value protection
04

Compute Price / Revenue Risk

Market and revenue
In sequence
What goes wrong

GPU-hour, token or compute pricing declines faster than expected, compressing revenue even when utilization remains high.

How it reads in your world
GPU cloud
Rate compressionUtilization holds, but the renewal price per GPU-hour falls below the cost of the capacity behind it
Investor
Revenue caseThe model assumed a price path the market has already moved past
What we compute
Revenue at the current market rateContracted and uncontracted revenue repriced against observed GPU-hour and token pricing, by term remaining.
Exposure by repricing dateWhen each contract resets, rolls or renews, and what it is worth if it reprices at today's rate.
Break-even price per unitThe price below which revenue stops covering power, financing and depreciation.
Signals we watch
  • GPU-hour pricing by generation and region
  • Token and inference pricing
  • Renewal and reset dates
  • Price floors, collars and indexation
  • Cost per unit of capacity
LeversFixed-price offtake · price floor · collar · long-term reservation · revenue hedge
05

Regulatory Deployment / Stranding Risk

Regulatory and geopolitical
Covered today
What goes wrong

Export controls, sanctions or AI rules change after deployment and restrict which chips, customers, workloads or jurisdictions can be served.

How it reads in your world
Data centre
Restricted tenants and workloadsA rule change limits which tenants or workloads the site may serve
GPU cloud
Export control and jurisdictionWhere the GPUs sit, and who may use them, moves under the rules
What we compute
Which deployments a rule change touchesA designation or rule update mapped to the chips, customers, workloads and sites that depend on it.
Capacity at risk of strandingRevenue and assets that can no longer be served where they sit, priced by jurisdiction.
The contractual answer to itChange-in-law, relocation, representation and termination-compensation terms that carry the risk, or the fact that none does.
Signals we watch
  • Export-control rules and classifications
  • Sanctions and restricted-party designations
  • Foreign and outbound investment regimes
  • Data, bulk-data and AI rules
  • Effective dates and rule versions
LeversChange-in-law · relocation rights · customer representations · termination compensation · regulatory cover
06

Contract Bankability Risk

Contract and financing
In sequence
What goes wrong

A large headline contract provides less real protection than expected because termination, volume, pricing or guarantee provisions weaken the revenue.

How it reads in your world
Data centre
Lease protectionA fifteen-year lease the tenant can exit, reduce or reprice well before term
GPU cloud
Contract mismatchA five-year obligation sits under an eighteen-month customer commitment
What we compute
Protected revenue, not headline valueContracted value adjusted for termination rights, volume flex, pricing resets and guarantee strength.
Where protection falls awayThe clauses that let revenue fall below what the financing assumes, ranked by value at risk.
The lever still availableTake-or-pay, termination payment, guarantee or step-in terms to seek, with the deadline that governs each.
Signals we watch
  • Termination and exit rights
  • Volume flex and reduction rights
  • Pricing resets and MFN terms
  • Parent guarantees and credit support
  • Covenants and step-in rights
LeversTake-or-pay · termination payment · parent guarantee · step-in rights · tighter covenants
07

Technology Substitution Risk

Technology
In sequence
What goes wrong

A new GPU, accelerator architecture or software stack makes existing infrastructure economically uncompetitive before the end of its expected life.

How it reads in your world
Data centre
Facility fitA new architecture needs power density or cooling the site was not built for
GPU cloud
Stack shiftCustomers move to an accelerator or software stack your fleet does not serve
What we compute
Remaining competitive lifeHow long existing infrastructure stays economically competitive against announced alternatives, by asset.
Gap against the next architecturePower density, cooling, networking and software compatibility compared with what new deployments require.
Exposure beyond competitive lifeCommitted tenor, debt and contracted revenue that runs past the point the asset stops competing.
Signals we watch
  • Accelerator and architecture roadmaps
  • Rack power and cooling requirements
  • Software framework and stack adoption
  • Upgrade and refresh rights
  • Asset tenor against commitments
LeversUpgrade rights · OEM refresh commitment · buyback · shorter asset tenor · technology diversification
08

AI Efficiency / Compute Compression Risk

Demand and technology
In sequence
What goes wrong

Models and inference software become much more efficient, so the same AI workload requires significantly less compute than originally assumed.

How it reads in your world
AI lab
Over-commitmentCommitted capacity was sized for workloads that now need a fraction of the compute
GPU cloud
Demand compressionCustomers serve the same traffic on fewer GPUs and cut reservations at renewal
What we compute
Compute required per workloadThe demand assumption behind each commitment, re-run as model and inference efficiency improves.
Committed capacity against compressed demandThe gap between what is contracted and what the same workload now needs.
Where flexibility existsVolume floors, flex bands and repricing rights read from the contracts, with what each recovers.
Signals we watch
  • Model and inference efficiency benchmarks
  • Compute per token and per request
  • Customer reservation changes
  • Volume floors and flex bands
  • Workload mix by customer
LeversShorter commitments · volume floors · workload diversification · flexible capacity · repricing rights
One counterparty across nine transactions is one exposure, not nine
Before a module is live

The four not yet covered still have an answer.

One of them already reaches you through the extensions, because the packaged output is the part customers need first. Contract Bankability Risk (06) is what an insurer-ready or lender-ready risk package evidences. The record exists today; the module is what makes it continuous.

Insurer-ready and lender-ready packages → What each tier includes → Ask what is covered for your book →