Is Enterprise AI Ready to Deliver a Return on Investment?

There is little doubt that AI has tremendous utility. AI is already helping people conduct research faster, automate repetitive processes, write code, analyse large datasets and improve productivity across several functions.

At the same time, it is important to separate the usefulness of AI from the commercial expectations currently being built around it. The key question is not whether AI will be important. It almost certainly will be. The real question is how quickly it can become a reliable, commercially viable and deeply integrated part of enterprise operations.

AI Is Useful, but Is It Enterprise-Ready?

At present, a large part of AI usage is still happening at the individual level, that is on personal desktops, through productivity applications or in selected backend processes.

However, enterprise-level adoption requires a much higher standard.

Large businesses need systems that can operate accurately and consistently for years. Software used in banking, accounting, healthcare, manufacturing or other critical functions cannot frequently produce unpredictable outputs or make factual errors. It must be dependable, secure, auditable and capable of integrating with existing systems.

Traditional software and SaaS products were developed for specific purposes, tested extensively and deployed through structured implementation processes. AI systems, particularly generative AI, are more flexible but can also be less predictable.

Therefore, while AI can already assist employees, it may still have some distance to travel before companies can confidently allow it to independently run critical programmes and processes with minimal human supervision.

Large Investments Have Already Been Made

Another misconception is that the AI investment cycle is yet to begin. In reality, billions of dollars have already been invested in semiconductors, data centres, cloud infrastructure, AI models and related technologies.

Technology companies are spending heavily to build the infrastructure required to train and operate increasingly sophisticated AI systems. This has created strong demand for advanced chips, servers, power capacity, cooling equipment, networking systems and data-centre facilities.

The utility of this infrastructure is visible. However, the revenue-generating model across the wider AI ecosystem is still developing.

For AI to justify the scale of investment currently being undertaken, it will eventually need to generate meaningful and recurring economic value for enterprises. That visibility will improve when companies begin using AI commercially across major business functions and AI-related expenditure becomes a clearly identifiable line item in corporate financial statements.

More importantly, we will need to understand what that expenditure is replacing.

The Internet and ERP Revolution Offers a Useful Comparison

The adoption of the internet and enterprise resource planning systems around 25 years ago provides a useful example.

Before ERP systems became widely adopted, many accounting, reporting and administrative activities were performed manually. A company may have required 100 employees to handle certain processes. Once ERP systems were introduced, perhaps 30 of those roles were no longer required.

However, the savings were not entirely free.

A new cost structure emerged. Companies needed IT departments, software licences, system upgrades, cybersecurity, maintenance contracts and integration services. A portion of employee costs was effectively replaced by technology-related expenditure.

Despite these costs, ERP adoption created substantial value because businesses gained better control, efficiency, visibility and scalability.

A similar transition may happen with AI.

AI could reduce the need for manual work across research, coding, customer service, accounting and several other functions. At the same time, companies may need to spend considerably on AI subscriptions, computing infrastructure, data management, implementation, monitoring and specialised employees.

Therefore, the correct question is not simply, “How many jobs will AI replace?” The more relevant question is: “What will the complete economics of an AI-enabled enterprise look like?”

The Return on Investment Is Still Unclear

The biggest unanswered question is the return on investment that AI can generate for businesses.

Companies will eventually need clarity on several issues:

What can AI perform independently? Which existing functions can it replace or improve? How much human supervision will still be required? What will implementation and operating costs look like? And most importantly, will the resulting productivity gains justify those costs?

At present, AI clearly provides value in areas such as research, summarisation, coding assistance, automation and transaction processing. However, the journey from being a useful assistant to becoming an integral enterprise infrastructure layer is still evolving.

For widespread enterprise adoption, AI will need to demonstrate consistently high accuracy, data security, regulatory compliance and measurable financial benefits.

Until this becomes clearer, estimating the long-term economics of AI remains difficult.

Why This Matters for the Data-Centre Theme

This discussion is particularly important when evaluating the data-centre investment theme.

A significant portion of data-centre growth over the past 2-3 years has been driven by expectations surrounding AI. Forecasts for the next 3-4 years also assume that demand for AI computing will continue expanding rapidly.

If AI adoption accelerates across enterprises, demand for computing capacity, storage, electricity, networking infrastructure and cooling systems could remain strong.

However, if businesses struggle to generate adequate returns from AI investments, the pace of deployment could slow. Companies may become more selective, projects may be delayed and infrastructure demand may not grow as rapidly as currently expected.

This does not necessarily mean that the AI or data-centre theme is fundamentally flawed. It simply means that investors must understand how much of the expected growth depends on assumptions that are yet to be fully proven.

The Need to Track Enterprise Adoption

Going forward, investors should closely track how AI adoption progresses at the enterprise level.

The most important signals may not come from announcements regarding new AI models or higher infrastructure spending. They may come from corporate financial statements and management commentary.

Are companies reporting measurable productivity improvements? Are AI costs becoming visible in operating expenses? Are these costs replacing employee expenses, outsourcing costs or traditional software spending? Are companies achieving higher margins or stronger revenue growth because of AI adoption?

These developments will provide a clearer picture of whether AI is creating genuine economic value or merely driving a period of aggressive investment and experimentation.

Invert the Argument

Charlie Munger often emphasised the importance of inversion. Whenever we form a strong opinion about an investment theme, we should ask what could go wrong.

The bullish case for AI is compelling. The technology has enormous potential, the infrastructure build-out is substantial and its usefulness is already visible.

But inversion requires us to ask the opposite questions.

What if AI takes longer than expected to reach the required level of accuracy? What if enterprises cannot generate sufficient returns on their AI spending? What if operating AI systems becomes more expensive than anticipated? What if some infrastructure is built ahead of commercially sustainable demand?

These questions should not be interpreted as a prediction that the AI boom will fail. They are simply necessary for maintaining a balanced view.

AI may ultimately become as transformative as the internet, cloud computing or ERP systems. However, transformations of this scale are rarely linear. They generally involve periods of excitement, overinvestment, experimentation, disappointment and eventual commercial maturity.

Therefore, the right approach is neither to dismiss AI nor to assume that every expectation surrounding it will automatically materialise.

The opportunity is real, but so are the uncertainties. Investors should continue tracking how AI moves from experimentation to enterprise adoption, how its economics appear in financial statements and whether the returns generated by the technology can justify the massive investments being made today.