Banking on the Cloud

Public cloud offers to swap upfront capital investments with consumption based pricing for infrastructure spending. This has 2 obvious consequences:

1. Reduce infrastructure costs by optimising consumption and right sizing the infrastructure estate.

2. Make technology cashflows more granular and on-demand enabling start-ups and scale-ups to compete with enterprises.

The 3rd less obvious consequence is the impact on the perceived financial performance of a firm. Most established businesses use EBITDA (Earnings Before Interest, Taxes, Depreciation and Amortisation) to report their financial performance. Its appeal is its simple calculation despite it not being an acceptable metric for financial performance under GAAP (Generally Acceptable Accounting Principles). 

EBITDA ignores the capital expenditure (CapEx) but accounts for operating expenses  (OpEx). Spending on on-premises infrastructure is considered CapEx while cloud costs are considered OpEx. As the infrastructure CapEx transitions to cloud OpEx, it decreases the firm’s EBITDA. This creates the perception of poor financial performance.

Increased cloud spending with AI is further compressing operating margins and stressing EBITDA. This creates tension within a firm due to increased OpEx, especially before the commercial benefits from AI are realised, just like in cloud adoption. Prematurely and speculatively cutting costs tends to stifle innovation which may increase EBITDA in the short term. But stalled innovation and lower than expected commercial benefits show only marginal improvements in EBITDA subsequently. 

The purpose of this blog is to introduce this key financial metric to technology leaders and how their technical decisions impact the perceived financial performance of their firms as a set of FAQs.

• So, what is EBITDA?
• Why does it matter to the public cloud and AI?
• What reduces the impact on EBITDA when migrating to public cloud and adopting AI?
• How to practically avoid impacting EBITDA?
• The middle ground – how cloud and AI costs can be capitalised?

So, what is EBITDA?

EBITDA represents a company’s cash profits generated by its operations. It is a widely used measure of corporate profitability to net income despite not being recognised by GAAP (Generally Accepted Accounting Principles).

EBITDA is calculated as follows:

EBITDA = Net income + Interest + Tax + Depreciation + Amortisation

Hence it excludes capital expenditure but includes operating expenses. 

EBITDA is largely used in the analysis of asset intensive companies that own a lot of equipment and thus have non-cash depreciation costs, but the costs it excludes may obscure the underlying profitability.

Why does it matter to the public cloud and to AI?

Public cloud’s usage based pricing is regarded as an operating expense, hence it reduces the net income which is not discounted as CapEx is, thereby reducing the EBITDA. On premises infrastructure incurs a capital expenditure which depreciates in value thereby not affecting the EBITDA.

Most firms use AI models hosted on public cloud, which means an increase in their operating expenses for executing tasks that rely on AI. AI promises to provide operational efficiencies that may reduce the operating expenses in the medium to long term. But the increased cloud consumption because of AI may lower EBITDA in the short term and may also eat into the savings due to operational efficiencies eventually gained from AI. 

Therefore, firms reporting their financial performance in EBITDA, may report lower profitability when migrating to and operating on public cloud and using AI on public cloud. 

This may lead to an overbearing and regressive focus on cloud and AI cost control.

What reduces the impact on EBITDA when migrating to public cloud and adopting AI?

Public cloud has higher unit cost as compared to on-premises infrastructure. A like-for-like setup on public cloud will invariably be more expensive with costs effectively reducing EBITDA. Cloud provides opportunities for efficiency and economy as well as innovation and growth that can reverse the impact on EBITDA, let alone lower it. 

This requires addressing OpEx both on cloud and on-premises. As cloud usage increases and the on-premises footprint reduces, there are opportunities to also progressively lower operating expenses on-premises.

Frequent cloud usage optimisations help avoid unnecessary cloud OpEx. This is not a substitute for negotiating and contracting favourable cloud usage rates. While a team may focus on both, usage optimisations can yield continuous efficiency and economy on public cloud.

It is important to remember that cloud costs will increase as business volumes scale. Cloud usage optimisations should also consider unit economics as an indicator to the impact on EBITDA with increasing costs. Unit economics refer to costs per unit of business activity and it should not be confused with unit costs. Examples of unit economics include cost per million transactions or cost per million active users. While  costs increase with more transactions or customers, so does the revenue. But if the unit economics remain constant, EBITDA may not reduce because of increasing cloud costs.

A balanced and disciplined approach to adopting AI can help reduce the impact on EBITDA. AI without hygiene in technology and data will invariably increase costs without providing the expected returns. Furthermore AI should be reserved for tasks most suited for it in a hybrid ecosystem of AI and conventional services that are orchestrated together to deliver the desired results.

How to practically avoid impacting EBITDA?

Avoid a big bang lift and shift involving long testing cycles and suboptimal cloud usage. Instead incrementally migrate capabilities to the public cloud and decommission them on-premises when they are in production in the cloud. Decommission capabilities that provide diminishing returns on-premises to reduce cloud usage. Optimise for base load and horizontally scale for increased load.

Rightsize resources allocated in the cloud. This also means only provisioning enough resources when incrementally migrating instead of the entire infrastructure footprint at par with on-premises infrastructure. Understand and exploit the duty cycle of non-production environments to release them when they are not needed.

Improve the quality of data which will lead to improved context with fewer tokens and more efficient processing by the LLMs. Further, agents that use conventional processing tools (business domain services, e.g., pricing calculators etc.) with LLMs also optimise processing. All of these help optimise AI usage.

Last but not least, consider opportunities to increase revenues on cloud, e.g., monetise data and value added services.

The middle ground – how cloud and AI costs can be capitalised?

Some firms are looking to account for cloud and AI operating costs as CapEx. This may (superficially) reduce the impact on EBITDA but may also reverse the cost savings that a firm wishes to achieve when using public cloud.

For costs to be regarded as CapEx, they need to be assessed and budgeted up front in annual budgeting cycles and not realised on the go. This will drive the same on-premises behaviour which leads to over provisioning and waste on cloud.

This may also lead to regressive controls on cloud provisioning and AI use, even when that usage clearly demonstrates a substantial upside for the business, just because that usage exceeds the annual budget.

Having said that, there may be opportunities to capitalise long term reservation and committed usage contracts. Care must be exercised to not restrict or constrain the on-demand usage exceeding these contracts to take advantage of emerging business opportunities.

Here, amortised long term cloud and AI reservations and committed usage costs are not reflected in the EBITDA but those exceeding these contracts are.

The FinOps function here needs to be firm that such contracts are made to minimise overall cloud and AI costs instead of optimising the EBITDA. This will ensure a focus on optimising costs and unit economics instead of one on just the perceived financial performance of the firm.