There’s increasing pressure for organisations to exploit their data and realise the value of AI. But their success, scalability and cost will depend significantly on the cloud environment beneath them. We look at why your cloud environment matters and discuss the key factors for success.
The AI-cloud disconnect
Organisations now understand the need to make use of AI, but very often their AI ambitions exceed their cloud readiness. There’s an understandable temptation to get caught up in the exciting stuff – models, use cases and business outcomes. However, the success of AI initiatives is heavily dependent on the more mundane foundations: data + compute + networking + security + governance + operations + cost control. That list could go on.
AI projects don’t operate independently of your cloud environment – they inherit its strengths and weaknesses. Poorly organised resources, fragmented data, weak governance or uncontrolled cloud costs won’t disappear when you introduce AI, but they may be magnified.
Data and AI initiatives are raising the stakes
AI is changing what we demand from our cloud infrastructure. Whereas traditional workloads are largely predictable, with AI we’re encountering more variable characteristics, such as large and changing data volumes, variable demand, and unpredictable consumption.
So, the question expands from ‘can our cloud run this?’ to ‘can our cloud run this securely, reliably, efficiently and economically at scale?’ This raises five further fundamental questions:
- Do we understand what we have, how it’s being used, and the value it supports?
- Can AI use the right data effectively and safely?
- Can our cloud adapt efficiently to the demands of data and AI?
- Can people innovate at speed without us losing control?
- Is increasing cloud and AI consumption translating into sufficient business value?
The key foundations for data and AI initiatives
Each of these questions relates to a prerequisite for an effective cloud infrastructure for data and AI: visibility, data, architecture, governance, and economics.
Foundation 1: Visibility – understand how cloud is being used
Before adding increasingly demanding AI workloads, you need a clear view of how your existing cloud environment is being used.
This isn’t simply an exercise of resource inventory – it’s about enabling better decisions on future investment. It means understanding how resources are consumed, what purpose they serve, who is accountable, and how efficiently they’re being utilised.
Understand:
- Which workloads and services account for most of our cloud consumption?
- What business purpose or outcome does each significant workload support?
- Who is accountable for its performance, security and cost?
- Where are resources under-utilised or unnecessarily duplicated?
Without this visibility, it’s difficult to know whether you’re building AI on an efficient foundation or simply adding consumption to an already flawed environment – one that contains waste, duplication and technical debt.
Foundation 2: Data – make the right data accessible
You already know that AI is only as good as the data it works with. But having large volumes of data doesn’t necessarily mean you can exploit it.
Ask yourself: Can we make the right data available to the right AI workloads, in the right form, with the necessary quality, performance and controls?
Then, probe deeper by asking:
- Do we know which data is most valuable to our priority AI use cases?
- Can applications and AI workloads find and access that data efficiently?
- Is important data fragmented or unnecessarily duplicated across platforms?
- Can we trust its quality, accuracy and currency?
- Do we understand its sensitivity and who should be permitted to use it?
- Is data location, storage and movement creating avoidable cost or performance constraints?
An optimised cloud environment makes data easier to discover, access, govern and move efficiently, while avoiding unnecessary duplication and storage or data-transfer costs.
Foundation 3: Architecture – ensure your cloud can adapt to the workload
Your cloud environment probably evolved over several years around conventional application workloads. Data and AI initiatives introduce very different demands.
The issue isn’t whether your cloud can run an AI workload. It’s whether your architecture can provide the right resources, in the right place, at the right time – and scale them efficiently as requirements change.
Consider:
- Is compute matched to the workload? Poor matching can cause either inadequate performance or unnecessary expenditure.
- Can capacity scale with demand? AI workloads are variable, so elasticity is both a performance and an economic consideration.
- Are compute and data appropriately located relative to one another? Otherwise, you risk avoidable latency, complexity and cost.
- Are you designing for experimentation as well as production? Teams need the freedom to experiment, while the business needs a route into well-governed and resilient production.
- Can the architecture accommodate change? The latest model will only be the preferred one for so long, so avoid unnecessary architectural lock-in.
- Are resilience and availability appropriate to the business importance of the workload? As you move from experimentation into production, resilience requirements change.
You can’t predict exactly what you’ll need in three or five years… the right cloud foundation is one flexible enough to accommodate requirements you can’t yet predict.
Foundation 4: Governance – guardrails without gridlock
Traditionally, governance has often meant restriction. With AI, it’s especially important that governance doesn’t block experimentation and improvement. The goal is to create an environment in which teams can innovate within clearly understood boundaries.
Ask:
- Who should be allowed to create and consume AI services?
- What data can AI workloads access?
- Are policies embedded into your environment, so they can be applied consistently?
- Do you have visibility of anything that operates outside these guardrails?
- Does accountability remain clear as AI moves into production – both now and into the future?
Foundation 5: Economics – optimise for value, not simply cost
AI is changing the economics of IT, with experimentation, specialised compute, model usage and growing data volumes all creating new and sometimes unpredictable costs. But don’t confuse cloud optimisation with cloud cost reduction.
An optimised environment is one that delivers the required performance, scalability, resilience, security and business outcomes at an appropriate cost.
We’ve previously explored how a FinOps approach can help control AI costs without limiting innovation. Here, we’ll focus on questions you’ll want to be able to answer.
- Which activities have increased cloud and AI spend, and was that increase justified?
- Are expensive resources being used productively?
- Can consumption and costs scale down when demand falls?
- Are experimentation and ultimate production costs considered from the outset of AI projects?
- Can you connect consumption with business value?
Make optimisation continuous
It’s true that an environment optimised for today’s needs won’t necessarily be optimised for tomorrow’s – and that’s especially true for data and AI.
AI workloads will change as they mature
An experimental workload may have few users and modest requirements, but in production it could support hundreds of users or a business-critical process. Changing its performance, resilience, security, governance and cost requirements. Optimisation needs to follow that journey from experiment to production, scale and eventual retirement.
What was optimisation can quickly become inefficiency
Optimisation decisions have a shelf life. Correctly sized resources become under- or over-provisioned, data ages, application usage changes and new services create better architectural options.
Review the five foundations together, not separately
Optimisation shouldn’t be several seperate activities conducted by different teams – it’s about balancing potentially competing requirements, rather than maximising individual metrics. An optimised cloud isn’t necessarily the cheapest cloud. It’s one that delivers the right balance of performance, scalability, resilience, security and business value at an appropriate cost.
Optimise frequently, if not continuously
Traditionally, optimisation has occurred periodically. But the fast-changing world of data and AI requires a new mindset. Cloud optimisation should be an operating discipline: continually observing consumption and performance, identifying change, acting on it and measuring the outcome.
Where does this leave you?
An optimised cloud provides the visibility, accessible and trusted data, adaptable architecture, and governance needed to enable innovation, while ensuring that consumption delivers business value.
Optimisation isn’t a one-off exercise and as initiatives move from experimentation to production their requirements will change. So, the big challenge isn’t simply to create an AI-ready cloud, but to keep it optimised as technology, workloads and business priorities evolve.
Before asking how quickly you can scale data and AI, ask a more fundamental question: are the cloud foundations beneath them ready too?
Are your cloud foundations ready for data and AI?
Why not speak to one of our Azure experts about your current environment, your data and AI ambitions, and possible optimisation needs. It’s an opportunity to compare notes, identify priorities and explore the practical steps towards an AI-ready cloud environment.
