AI - Is the hype justified?

I should begin by stressing that I embark on this attempted analysis with an absence of detailed prior knowledge and am therefore reliant on a wide range of informed views — some of which derive from not entirely balanced perspectives or agendas, but they nevertheless ask the important questions.

A trillion dollar commitment 

Although returns are, as yet, a small fraction of the sums already invested, the world appears already committed to something that it might be unable to stop.

A trillion dollar commitment is already shaping the global economy, with Open AI, Nvidia, big tech and corporate CEOs burning through hundreds of billions, chasing an AI future that might not be profitable. Goldman Sachs analyses the vast cost of data centres, likening the market's enthusiasm for AI to the 'dot-com' bubble — an opportunity that proved absolutely real, but where investors suffered in companies that never figured out how to profit from the new technology.

A strong bull case is expressed by Joe Davis, Global Chief Economist at Vanguard. Despite headwinds in 2025 such as demographic slowdowns and rising tariffs, US corporate earnings growth and fundamentals stayed strong, powered by artificial intelligence investment. If AI investment can counteract negative shocks, it could continue to shape the economic outlook. Vanguard sees an 80% chance that economic growth exceeds consensus expectations over the next five years.

So where is the revenue?

In 2024, analysts at Sequoia asked industry specialists: if companies continue to spend at current levels, how much income must AI throw off each year to generate the necessary returns? The answer given was $600 billion. Generative AI is already throwing off significant income, but reported revenues are not always offset to reflect the enormous cash burn of new capital being simultaneously injected. Traditional software can be highly profitable, because once written, users can be added at almost zero cost. For AI, every question costs money to answer, including electricity and costly infrastructure, so scaling can incur diseconomies of scale. If profitability is as yet uncertain, why do so many CEOs stake so much on AI's future? 90% believed its introduction will prove transformative by 2028, but to date, less than a quarter have delivered the expected returns on investment. So is the boom partly down to peer group pressure?

The evidence from the ground

In February 2024, Forbes reported Klarna's CEO's claim that an AI chatbot was doing similar work to 700 full time customer service agents. Fifteen months later he admitted to Bloomberg that this first push had destroyed customer service quality. He reversed the AI push. Duolingo's CEO distributed a memo declaring that company AI policy meant phasing out contractors; following an immediate backlash, he backtracked. Salesforce shifted away from raw autonomous AI and returned to 'humans-in-the-loop' hybrid workflows, hiring 1,000 new graduates after initially cutting thousands. Starbucks discontinued an AI inventory system across 11,000 stores — it could not reliably differentiate between milk types — and returned to counting by hand. Although nobody is seriously suggesting that the majority of AI applications miss their targets, the evidence points towards early and ill-thought-out AI adoption causing serious problems and needing adjustments requiring more human input.

Can the infrastructure be introduced quickly enough?

The AI investment case is built on two assumptions: that demand will grow fast enough to absorb the infrastructure costs and that productivity gains will arrive quickly enough to justify the price. If either slip, the entire premise is threatened. There is a fear that claims for the commissioning of fully operational data centres are being overstated. Microsoft's claim of a one gigawatt data centre might be an aspiration; at the time of the statement, only one phase had started and none were yet completed. Amazon publicly stated that their data centres were fully operational at a time when only seven of thirty were actually built. A data centre project started today will not be operational before 2029. These issues alone suggest that reality might not meet with current valuations.

In January 2025, Microsoft pledged to spend $80 billion on data centres in a single year. A month later, TD Cowen analysts found that major hyperscalers had cancelled and walked away from data centre projects in the UK, US, Australia and Indonesia totalling over 2GW capacity. A slower pace of data centre construction begs the question of how Nvidia is managing to sell so many of its GPUs. Fears of future shortages appear to be leading to warehouses being filled with inventory under Nvidia's 'bill and hold' facilities before data centres are built and ready — a practice drawing scrutiny from investors and analysts. There is apparent overcapacity, circular funding and limited market awareness of data centre completions. 'Circular financing’ refers to the web of interconnected deals and investments in the AI sector where tech giants - primarily Nvidia - invest billions of dollars into AI startups and cloud providers. These recipients then use that capital to purchase Nvidia's chips and cloud services, effectively routing the money back to Nvidia as revenue.

Another risk relates to the availability and cost of energy and water. It is estimated that a single AI enquiry might use up to ten times the electricity of a standard web search. A single large new data centre can consume 100 MW, more than enough to power 15,000 homes. That is one building of the 8,000 expected worldwide, one-third of which in the US, and many in climates too hot for efficient operation. 

Finally there is safety and security. Approximately one-third of all data systems contain sensitive information and most companies have little or no controls in place.

The verdict

Some critics liken the AI structure to the telecom bubble of the late 1990s, where suppliers financed their customers' purchases. Time will tell, and the timelines and supply logistics related to data centre build outs would seem critical to the question of whether from these levels we are still looking at a compelling investment case, or whether delayed profitability will lead to a late 1990s ‘dot com’ investment risk. Mark Zuckerberg acknowledges the real possibility that AI will follow the same pattern as past infrastructure bubbles. Jamie Dimon of JP Morgan believes AI companies to be massively overvalued. Ray Dalio, founder of the world’s largest hedge fund, compares the current situation to the years before the dot-com crash. None of them dispute the potential power of AI, but fear that current valuations are some distance ahead of reality.

About the author 

Jo Welman had a career in the City spanning 45 years and worked in a wide variety of financial sectors. After graduating from Exeter University in 1979 with a degree in economics, Jo spent ten years at Baring Asset Management where he managed a range of UK and US pension funds and unit trusts, investing across multiple sectors including bonds, international equities, commercial and residential property and private equity.

In 1989 Jo became Managing Director of merchant bank Rea Brothers’ institutional and private wealth investment management division. Over the following decade Jo launched a series of specialist investment trusts and funds in a variety of industry and property sectors, before forming a joint venture with reinsurance broker Benfields (now Aon Benfield) and raising one of the first limited liability corporate capital vehicles for the Lloyds insurance market in 1993. As part of his long-standing involvement in the insurance industry, Jo co-founded the Benfield Re-Insurance Investment Trust plc (Brit) in 1995. Following the sale of Rea to Close Brothers in 1999 Jo became Chairman of Brit Insurance Holdings Plc and in 2001, in partnership with Brit and Benfields, he co-founded specialist asset management firm, EPIC Investment Partners (EPIC).

Jo continues to provide corporate finance and investment advice to entrepreneurs and private investors. He sits on the board as a non-executive director of ARK Syndicate Underwriting.

“Feet up by the pool”

Jo does not receive any remuneration for his EPIC commentary. Instead, EPIC is pleased to promote the latest edition of his book “Feet up by the pool”.

Profits from sales of the book go to The Money Charity, a charity that shares Jo’s objective to help fill in some of the worrying gaps in the school curriculum. These omissions leave many young adults lacking in the financial awareness that they need to survive in a world where they will rely on their own savings if they are ever to stop working. Even if they earn the right to a full State Pension, today this amount hardly covers council tax and utility bills, and so they need to save and build up a sum of capital amounting to around twenty times their desired retirement income. A frightening number.

As Jo eloquently says, “If we can do our bit to raise awareness of the impending UK saving and pensions crisis, the exercise will have been worthwhile.”