Much of today’s AI infrastructure conversation is focused on models and compute. Enterprises and technology providers are investing heavily in data centres, GPUs, cloud platforms, AI tooling and governance frameworks, with the expectation that stronger infrastructure will naturally lead to better AI outcomes.
So when Alphabet announced plans in early June to invest $80 billion into AI infrastructure, the industry response was predictable. The discussion focused on compute capacity, model development and the race between hyperscalers to build the foundations of the AI economy. This highlighted a growing disconnect in the market: hyperscalers are solving the supply-side challenge of AI; meanwhile, enterprises are grappling with the demand-side challenge of whether their data is ready to support AI in production.
For many enterprises, the biggest barrier to AI adoption is no longer access to models or compute. It’s what we are classifying as the missing layer – the operational layer required to make enterprise data usable by AI.
The growing gap between AI capability and AI deployment
Over the past two years, AI capabilities have advanced at an extraordinary pace. Large language models can reason, summarise, analyse and generate content at levels that would have seemed impossible only a few years ago; agentic systems are beginning to automate increasingly complex workflows; and multimodal AI is expanding the range of tasks machines can perform.
Despite this progress, many AI initiatives across enterprises continue to stall. And that comes down to one major challenge: data AI-readiness. Enterprise data was never designed for AI. Most enterprise information exists inside environments built for transactional systems, regulatory compliance and human decision-making. Data is often fragmented across departments, restricted by policy, inconsistent in quality or missing the context required for effective AI execution.
In many organisations, the challenge is therefore not a lack of data – it’s an abundance of data that cannot be operationalised for AI. And this, in turn, creates a growing disconnect between what AI systems are capable of and what enterprises can actually deploy.
The infrastructure layer no one talks about
When people discuss current AI infrastructure, they typically think about data centres, cloud platforms, GPUs, and foundation models. These are all critical components. But what they don’t consider is whether the underlying data being provided is fit for AI consumption.
Storage platforms can store data. Governance tools can secure it. AI platforms can execute against it. But none of them have been designed to transform fragmented, restricted, and operationally complex enterprise data into a state in which AI can reliably use it.
As a result, a critical infrastructure gap has emerged between enterprise data and AI execution. This gap is giving rise to what I believe will become one of the defining categories of infrastructure in the AI era: AI-ready data infrastructure. Just as cloud infrastructure made compute scalable, AI-ready data infrastructure makes enterprise data usable, traceable and operationally reliable for AI systems.
Why AI needs an operational data layer
Every major technology shift has created a new operational and foundational layer that solved a problem the previous stack wasn’t designed to solve. Before enterprise networking, systems were largely isolated. The network layers enabled cross-system communication, distributed applications, shared access to information, and connected enterprises. Traditional IT required organisations to buy, own and manage infrastructure, but cloud introduced on-demand computing, infrastructure as a service, and consumption-based economics. And as data volumes exploded, organisations needed more than databases. Modern lakehouses introduced centralised storage, data integration, analytics at scale, and data sharing across teams.
But none of these layers were designed to answer the operational questions that enterprise AI now depends on: is this data usable by AI, is it complete enough, can we use it safely, and can we trace which data state produced a specific AI output?
AI is now exposing the demand for this new operational layer focused on data readiness. This layer sits between enterprise data and AI execution, and its purpose is to continuously prepare, validate, transform, and maintain enterprise data to a state that AI systems can use safely and reliably.
The next challenge after compute
Right now, compute remains a strategic constraint and demand continues to outpace supply. But as infrastructure investments by companies like Alphabet continue to expand the availability of AI resources, data readiness will become more prominent.
Many organisations can identify when an AI system behaves differently. Far fewer can explain why. A model that performs reliably today may produce different outcomes tomorrow because the underlying data has changed, drifted or lost critical context. In most enterprise environments, there is no operational mechanism to trace exactly which data state informed a particular AI decision.
This creates a blind spot that becomes increasingly dangerous as AI moves into customer-facing, operationally complex and mission-critical workflows.
Beyond the AI arms race
Alphabet’s $80 billion investment reflects confidence in the future of AI. But building the future of AI requires more than larger models and larger data centres. It requires infrastructure that bridges the gap between enterprise data and AI execution.
For years, the industry has focused on making AI more capable. The next phase of innovation will focus on making enterprise data more usable. This is the layer CUBIG is building: the missing operational data layer for enterprise AI. While compute and models continue to scale, the next decade of competitive advantage will belong to enterprises that solve the data readiness problem first.
The organisations that solve this challenge first will not simply deploy more AI. They will deploy AI more reliably, more safely and at greater scale. The next decade of enterprise AI will not be defined by who builds the biggest models. It will be defined by who makes enterprise data finally operable.
