The AI craze is fueling massive growth in infrastructure, but the industry will need to hit $2 trillion in revenue by 2030 to keep funding this habit. Consultants at Bain & Company think it is going to come up short.

A slew of AI-focused investments have been announced recently, with OpenAI planning five new massive server farms across the US and Microsoft claiming it is building the “world’s largest datacenter” in Wisconsin, for example.

Management consultant biz Bain calculates the industry is heading toward creating an extra 100 gigawatts of capacity in the US by 2030 to meet demand, but it also estimates spending of $500 billion per annum on building datacenters will be needed to get there.

In its Technology Report 2025, Bain & Co asks how the industry plans to fund this infrastructure. Looking at the AI’s topline, it says that a sustainable level of capex to revenue suggests the sector as a whole will have to be making $2 trillion in annual sales to be able to “profitably” afford this.

The problem is Bain estimates that even if companies shift all of their on-premises IT budget to the cloud, and reinvest any projected savings from AI productivity gains into capital spending on new datacenters, the total amount would still come out $800 billion short.

However, all of this assumes the AI bubble continues unabated, whereas there are warning signs that companies are beginning to question the value. A report out last month found that US firms had invested between $35 billion and $40 billion in generative AI projects, but 95 percent of organizations had seen zero return from these efforts.

Some experts also suspect the tech industry is unlikely to inject quite the level of investment indicated by Bain’s figures.

“Is $500 billion per year purely for AI infrastructure a reasonable number? For 2025, I’d say a categorical no. That is far too aggressive a number for this year. It is possible the numbers will reach something like that in the fullness of time – and for fullness of time read probably 2-3 years out,” said John Dinsdale, chief analyst at Synergy Research.

Recent figures from Synergy show that total hyperscale operator capex hit $127 billion in the second quarter of 2025, up 72 percent in a year, with investments in AI infrastructure as the main driver.

But Dinsdale warns that it is difficult to draw a hard demarcation line between what is AI and what is non-AI datacenter investment, and a lot of funding will continue to be spent on non-AI infrastructure.

“There is a lot of hyperbole and silly numbers bouncing around the industry and media at the moment. You have to filter those numbers to get back to some form of reality,” he told The Register.

Sid Nag, president and chief research officer at Tekonyx, estimates global capex on AI datacenter infrastructure in the range of $300 billion on an annualized basis, but the figure could fall if demand tails off.

“Keep in mind these numbers may go higher if demand meets or exceeds supply. In my opinion, I do not see the demand meeting supply, especially in light of the recent MIT study that indicated that 95 percent of AI projects fail. If that is the case, we will witness a drop in capex spend on an annualized basis in the years out to 2030,” Nag said.

Bain’s report also claims that it will be difficult to build datacenters fast enough to meet demand due to constraints in four areas: energy supply, construction services, compute enablers (GPUs), and limits on the supply of ancillary equipment such as electrical switchgear and cooling systems.

“Of these, increasing the supply of electricity may be the most challenging as bringing new power generation, transmission, and distribution online in a highly regulated industry can take four years or longer,” the report authors state.

Bain previously noted this last year, warning that US utility companies need to revamp the way they operate to meet the rapidly changing demand for power.

Unlimited growth dream a mirage

Meanwhile, an analysis by London Economics International (LEI) found inn July that many estimates of future datacenter growth are unrealistic, since they imply an increase in hardware that would be beyond the capacity of global chipmakers to provide.

Advances in technology could provide an answer, according to Bain. Without such innovations or breakthroughs, general progress could slow, and the field could be left to only those players in markets with adequate public funding – presumably from governments.

However, the latter ignores the fact that some of the big players in the market are already spending more on new infrastructure than the GDP of some countries – for now, at least. The annual datacenter capex of cloud giant Amazon exceeds $100 billion, for example, making it roughly comparable to the entire GDP of Costa Rica, and greater than that of Luxembourg or Lithuania.

Regardless, the AI hype train rolls on for now. OpenAI chief Sam Altman said this week: “If AI stays on the trajectory that we think it will, then amazing things will be possible. Maybe with 10 gigawatts of compute, AI can figure out how to cure cancer. Or with 10 gigawatts of compute, AI can figure out how to provide customized tutoring to every student on earth.”

OpenAI aims to “create a factory that can produce a gigawatt of new AI infrastructure every week,” he added. ®