AI infrastructure is being built at an extraordinary scale.
New data centres need land, power, cooling, network connections, processors, memory and storage. Building every facility directly is not always the fastest option, so major technology companies also lease data-centre capacity from specialist operators.
This allows them to secure computing resources years before the facilities are ready.
It also creates a large financial commitment.
Reuters reported on 4 August 2026 that Microsoft, Meta, Oracle, Amazon and Alphabet have committed approximately $1.09 trillion in future lease payments, mainly for data centres required to support artificial intelligence growth. These leases have been signed but have not yet started because the facilities are not available for use.
The figure is enormous.
But the practical lesson is relevant even to companies operating on a much smaller scale.
Infrastructure should support business growth without trapping the company in a contract that no longer matches its needs.
Why Are Technology Companies Leasing So Much Capacity?
Demand for AI computing has increased quickly.
Cloud providers and technology platforms need capacity for:
- AI model training
- AI inference
- Cloud applications
- Data storage
- Networking
- Cybersecurity services
- Customer workloads
- Internal AI products
A large data centre can take years to design, obtain approval for, build and connect to the electrical grid.
By signing leases early, technology companies can reserve capacity before it becomes available.
This helps them avoid a situation where customer demand arrives but there are not enough servers or data-centre facilities to support it.
If AI demand continues to grow, these commitments may support the next stage of cloud expansion.
If growth slows, the companies may still need to pay for expensive capacity that is difficult to reduce.
Why the Full Amount Is Not Yet Shown as a Lease Liability
The $1.09 trillion is not hidden, and it should not simply be added directly to normal company debt.
These commitments are generally disclosed in the notes to financial statements.
Under the accounting treatment described by Reuters, signed leases are usually recorded as liabilities only when the facility becomes available for use. Until then, the future payments remain classified as uncommenced lease commitments.
The total is still important because it shows how much future spending has already been committed.
Reuters found that the five companies had approximately $285 billion in recognised lease liabilities. Their uncommenced commitments were nearly four times that amount.
In simple terms, a large part of the infrastructure bill is still waiting to enter the active contract period.
Microsoft Has the Largest Disclosed Pipeline
Microsoft had the largest disclosed pipeline, with approximately $329.1 billion in uncommenced lease commitments compared with $88.52 billion of recognised lease liabilities.
Meta disclosed approximately $278.99 billion and later signed another $68 billion of data-centre leases in July.
Oracle disclosed approximately $260 billion, while Alphabet reported $85.2 billion and Amazon reported $137.21 billion. Amazon’s number also includes other leased assets such as warehouses, offices, aircraft and vehicles, so it is not directly comparable with the others.
These commitments show how strongly the largest cloud and technology companies expect AI demand to continue.
They also show why future cloud capacity is not unlimited or free.
Someone must pay to build and operate the physical infrastructure behind it.
Long Contracts Create Long-Term Risk
Oracle’s disclosed commitments provide a useful example.
Its data-centre leases are generally expected to begin between fiscal years 2027 and 2029 and may continue for 15 to 19 years.
Oracle has warned that the duration, renewal terms and pricing of its leases may not always match its customer contracts. That could create risk if customers do not renew or are unable to fulfil their own commitments.
The same basic risk can affect any business.
Imagine signing a five-year infrastructure contract based on an expected project.
The project is delayed.
Customer demand is lower than forecast.
The application changes.
A more efficient technology becomes available.
But the company still has to pay for the original capacity.
The contract may have been reasonable when it was signed.
The problem appears when business requirements change faster than the agreement.
Businesses Should Not Avoid Long-Term Contracts Completely
A longer contract can provide useful benefits.
These may include:
- Lower monthly pricing
- Reserved computing capacity
- Predictable budgeting
- Protection from short-term price changes
- Dedicated infrastructure
- Better commercial terms
- Guaranteed resource availability
For stable workloads, a multi-year agreement may be sensible.
Examples may include an ERP system, established customer platform, production database or long-running business application.
The issue is not contract length by itself.
The issue is committing before the workload is understood.
Five Questions to Ask Before Committing
1. Is the Workload Proven?
Do not size infrastructure based only on optimistic projections.
Start with actual usage where possible.
Measure:
- CPU utilisation
- Memory consumption
- Storage growth
- Network traffic
- Number of users
- Peak demand
- Required availability
For a new AI project, begin with a controlled trial before reserving large-scale infrastructure.
2. Can the Capacity Be Adjusted?
Ask whether resources can be increased or reduced during the contract.
A flexible agreement may cost slightly more per month but carry less long-term risk.
Check whether the contract allows:
- CPU or RAM changes
- Storage expansion
- Server-plan changes
- Migration to another platform
- Early renewal
- Partial termination
3. Does the Contract Match the Customer Commitment?
A company should be careful about signing a five-year infrastructure agreement to support a customer who has committed for only one year.
The contract periods do not always need to be identical.
But the financial exposure should be understood.
If the customer leaves, can the infrastructure be used for another workload?
4. What Is the Exit Cost?
Review more than the early-termination charge.
Leaving a platform may also involve:
- Data-transfer fees
- Application migration
- Engineering work
- Software changes
- Downtime
- New licensing
- Backup migration
- User testing
A low monthly price may come with a high exit cost.
5. What Happens If the Project Grows Faster Than Expected?
Underestimating demand can also create problems.
The provider may not have additional capacity available when it is needed.
Ask whether the environment can scale and how long an upgrade normally takes.
Flexibility should work in both directions.
Cloud Can Reduce Commitment Risk
Cloud infrastructure can help businesses avoid purchasing large amounts of hardware before demand is clear.
A company can begin with a smaller environment and increase resources as usage grows.
However, cloud services can still create long-term dependency when businesses use reserved capacity, proprietary services or multi-year spending commitments.
Cloud does not remove infrastructure risk.
It changes the type of risk.
Companies still need to manage:
- Consumption
- Contract duration
- Provider dependency
- Data location
- Security
- Backup
- Recovery
- Portability
The best cloud design is not necessarily the largest or cheapest one.
It is the one that matches the workload and can adapt when requirements change.
Closing Thoughts
The AI data-centre boom has created more than $1 trillion in future lease commitments for Microsoft, Meta, Oracle, Amazon and Alphabet.
These agreements may help the companies meet strong future demand for AI and cloud computing. They also create long-term financial exposure if that demand does not develop as expected.
Most businesses will never commit to infrastructure at this scale.
The planning principle remains the same.
Understand the workload before signing.
Avoid buying capacity only because growth is expected.
Match infrastructure commitments with customer and project timelines.
Review scalability and exit terms.
Keep backup and migration options available.
Long-term infrastructure can support long-term growth.
It should not become a long-term burden.
At Net Onboard, we help businesses design cloud, dedicated-server, backup and business-continuity environments according to real operational requirements, expected growth and acceptable risk.