Bain Puts a $6 Trillion Price Tag on the AI Data Centre Buildout

The consultancy's technology report series, covered by The National on 29 September, projects that annual AI infrastructure spending could reach $1.5 trillion by 2031. That covers new facilities plus upgrades to installed GPUs, memory and networking gear. If capex runs at about a quarter of industry revenue, a ratio Bain calls ambitious but reasonable based on cloud provider trends, the market has to approach $6 trillion a year to sustain it. Bain breaks the number down. New product development, including search, advertising, autonomy and physical AI, is expected to contribute $4.2 trillion, the largest slice. Enterprise productivity needs $1 trillion to $1.4 trillion. Consumer services, meaning subscriptions and advertising, adds $200 billion to $400 billion.

"The debate today is fixated on employee productivity. The economics of AI infrastructure demand trillions in new revenue beyond productivity gains," said David Crawford, chairman of Bain's global technology practice and lead author of the report.

The report also flags absorption speed, the pace at which companies actually put AI to work, as what Bain calls the new competitive variable. Leading AI labs are investing upwards of $9.75 billion in engineering models to help customers assimilate faster, the report says. Bain reckons drug discovery, mental health and energy generation are among the markets that abundant intelligence could open.

Meta's Ohio site: 600MW now, 9GW by 2030

The clearest illustration of the cost curve is Meta's Prometheus data centre in Ohio. According to research firm Epoch AI, cited in the same report, it has 600MW of capacity at an estimated $24 billion in 2025. That is projected to reach as much as 2GW and $80 billion by 2027, 5GW and up to $175 billion by 2029, and 9GW at $200 billion by 2030. Bain lists the obvious obstacles: grid capacity, GPU and component supply, a skilled workforce that has to be retained far above historical rates, and public opinion and regulation, including pushback on resource use and noise. On the other side of the ledger, governments in the UAE, Saudi Arabia, the EU, South Korea and the US are supporting the buildout. Bain notes data centres are now treated as central to economic growth and national sovereignty.

Canada's easternmost province is testing that proposition. CBC News reported on 29 September that Newfoundland and Labrador has been approached by companies about developing AI data centres. Energy and Mines Minister Lloyd Parrott said during a mid-September special session on the Churchill Falls agreement that the province could consider selling excess power to a data centre operator. "We said our door is open for business and we're looking forward to doing business with the world," Parrott said at the time. A department spokesperson, Brodie Thomas, told CBC the government cannot comment on specific proponents or projects.

Not everyone in the province is sold. Paris Marx, author of Hyperscale, questioned the trade-off: "Does it make sense to be giving that electricity over to data centres when there might be better ways that it can be used?" Andrea King, CEO of TechNL, said Labrador appears to meet the criteria proponents want, low-carbon hydro, cold weather and land, but that this does not settle the question. "It's not just 'does this make sense?' by itself. You have to look at economic costs and opportunity costs of what else you could do with that electricity," she said.

Batteries instead of peakers

Power supply is not only a question of building more generation. The BBC reported on 29 September that Green Mountain Power's virtual power plant, built from home batteries across Vermont, has become the state's largest single source of power. More than 5,500 customers have joined a scheme that installs two batteries per home for $55 a month on a 10-year lease. The US now has over 40GW of virtual power plant capacity, according to internal analysis by Wood Mackenzie, against roughly 600GW of natural gas and 200GW of coal. A 2025 US Department of Energy report cited by the BBC puts the total that could be added at 160GW by 2030, about 20% of expected peak demand.

Resilience is the pitch. GMP tells the BBC that seven of the 10 most damaging storms in its history occurred in the last decade, causing more than $225m in damage. Customer Pollyanna Bladyka, who previously lost power a dozen times a year, has not lost it since installation in 2024. "I'd like to say I did it for the environment, but I did it for myself," she told the BBC. "But I don't mind helping."

Cooling and power efficiency sit at the centre of all this. The Register reported on 29 September that researchers affiliated with Glow Security found more than 13,000 sensitive screenshots of corporate software projects from 343 companies posted to public GitHub repositories by AI coding agents, a finding they call PixelLeak. The agents were working around GitHub's lack of an API for uploading images to pull requests. The risk is not hypothetical for operators of large fleets: one case involved a manufacturer with more than 100,000 employees whose security team only learned of the posts after Glow reported them.

Meanwhile, the data itself keeps growing. BMLL Technologies, writing on the Polars blog, says it processes 1.5TB of market microstructure data across two weeks of Chinese equities in just over three minutes on a single 192-core, 1.5TB RAM machine, roughly 48x faster than pandas on a single day of US equity trades. The infrastructure savings, BMLL argues, make an entire layer of dedicated data infrastructure optional. That is the same argument Bain is making about AI at a much larger scale, and the same argument Newfoundland and Labrador is now weighing against the value of the power it already has.