Bain says AI needs $6 trillion a year in revenue by 2031 to justify its data centres

Started by Darren, Oct 01, 2026, 07:08 AM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

Topic: Bain says AI needs $6 trillion a year in revenue by 2031 to justify its data centres   Views(Read 112 times)

Darren

Bain & Company has put a number on the question everyone keeps asking about the AI boom. According to its latest report, the industry will need around $6 trillion in annual revenue by 2031 to justify all the data centres being built. Existing AI products are only expected to bring in $1.2 to $1.8 trillion a year by then. That leaves a gap of roughly $4.2 trillion that nobody has yet worked out how to fill

The spending side is enormous. Bain expects annual infrastructure spending could hit $1.5 trillion by 2031, covering new data centres, chips, memory and networking. The five biggest spenders, Microsoft, Google, Amazon, Meta and Oracle, are projected to put $780 billion into capital expenditure this year alone, almost five times what they spent three years ago. The largest data centres are doubling in size and cost every 12 to 16 months

Bain's model has $5 to $6.5 trillion being spent through 2030 to build almost 150 gigawatts of new capacity, which would nearly triple global AI capacity in five years. Power supply, chip availability, skilled labour and planning permission are all listed as bottlenecks. Anyone who followed our threads on underwater data centres and India's data centre boom will recognise those problems

So where could the missing money come from? Bain suggests ad supported chatbots could bring in $100 to $200 billion, autonomous vehicles and industrial automation around $400 billion and physical AI, such as robots, digital twins and simulations, about $900 billion. The rest would have to come from applications that do not exist yet. David Crawford, who chairs Bain's technology practice, said the economics demand trillions in new revenue beyond productivity gains

This fits neatly with the $10 trillion buildout piece we discussed last week and the huge compute commitments in Anthropic's prospectus. Either AI becomes one of the biggest industries in history very quickly, or a lot of investors are going to be disappointed. Which way do you think it goes?


BretHart_99

Skilled labour is an overlooked bottleneck. Electricians and engineers who can build these sites are in short supply everywhere. You cannot conjure them up overnight

Saka19

Physical AI at $900 billion seems the most plausible chunk. Robots, factory automation and simulation have clear paying customers. Chatbot ads feel much shakier. People already hate ads in search results. Putting them inside conversations could backfire
Normal is overrated

Cyclops46

Every infrastructure boom looks like this. Railways, fibre optic cables and the early internet all overbuilt before demand caught up. Plenty of companies went bust, but the infrastructure stayed and got used eventually. AI data centres might follow the same path. Great for society long term, painful for investors short term
Making the internet slightly better one post at a time

Freya_27

A $4.2 trillion gap with the plan being applications that do not exist yet is a bold business case. That is not a forecast, it is a hope

Investors are being asked to take a lot on faith. Some of them will

Wireless Poet

Seven hundred and eighty billion in one year from five companies is staggering. That is more than the entire defence budget of most countries combined. It is hard to picture that much concrete and silicon. Yet it is being built right now

Gary

Bain is a consultancy, so some scepticism is healthy. Big scary numbers make for great headlines and sell consulting work

Still, the basic maths is hard to argue with. Spending is running miles ahead of revenue right now

HeartbreakKid_Royale

Productivity gains are the part Bain seems to discount. If AI makes millions of office workers even ten percent more productive, that value is huge but hard to capture as AI revenue

It shows up as company profits elsewhere. The accounting does not quite capture it

That might mean the gap is smaller than it looks. Or it might mean the AI firms themselves never see the money

QubitZero

Doubling in size every 12 to 16 months cannot continue for long. Power grids simply cannot keep up with that. Something has to slow down, and it will probably be the build rate rather than demand

Grid connection queues are already years long in some places. Expect delays to become the story next year

Violet_49

Six trillion a year?! At this rate my local Greggs is going to need an enterprise GPU cluster just to forecast sausage roll demand by Tuesday. Seriously though, the ROI math here feels like someone trying to pay off a mortgage by selling imaginary clouds. :P
Tap out or train harder, no in between

Layla81

Remember when crypto promised to revolutionise everything while melting small towns' electrical substations? AI is just that with better PR and actual working software. Call me when the autonomous agents start paying my electricity bill directly.

Pilgrim11

Disagree slightly with the doom and gloom crowd here because markets have a funny way of finding cheap energy sources when desperation kicks in. Once private nuclear fusion or space-based solar gets rolling, these data centre power bills are going to look quaint. :D