Major US consulting firm Bain has estimated that the global AI industry needs to generate $6 trillion (roughly ¥944 trillion) in annual revenue by 2031 to justify the capital being poured into data center construction around the world. On 5ch, many commenters reacted to the figure by saying it was ‘less than I expected,’ while others raised skepticism about the sheer scale of electricity AGI would require, pointed to a ‘game of chicken’ structure where companies keep racing to invest despite bleeding money, and voiced concern that data centers are straining local water and power supplies.
AI Industry Needs $6 Trillion in Annual Revenue to Justify Data Center Investment — Bain
(Bloomberg) — The global AI industry needs to earn $6 trillion (roughly ¥944 trillion) in annual revenue by 2031 to justify the capital being invested in data center construction worldwide, according to US consulting firm Bain & Company.
(details omitted)
Source: newsdig.tbs.co.jp / Original article here
>That's less than I expected
Every single person on Earth (all 8 billion) would need to pay about $700/year (¥110,000) in subscriptions for that to add up.
Well, companies could pay around $70,000 (¥10 million) per employee replaced with AI. That's one more person out of a job, though…
If you replace 100 million high-paid white-collar workers worldwide at roughly $70,000 (¥10 million) of AI spend each, the math could work out — but the middle class gets wiped out globally, and you end up with a Great Depression.
He's the easily-hyped type — he's probably already forgotten what he said.
Some people talk like AI is going to become an omniscient, omnipotent destroyer-god AGI, but achieving AGI would require building and running thousands, even tens of thousands, of nuclear power plants — that's a world far off in the future.
With the electricity available on Earth today, the best we could manage is something like the capable genie servant from the Arabian Nights — the 'spirit of the ring.'
Just call it AGA already (lol — a pun on AGI and the medical term for male pattern baldness)
Re: #102
Just maintaining and continuously upgrading the data centers is already unsustainable on its own.
Re: #52: "Yes"
AI: "I'll kill you, then."
that way they become the hostages.
Or it could backfire and turn into a Trojan horse instead.
They're guzzling rural water supplies, and the locals are the ones left holding the bag.
It'll be a disaster if power companies pour money into expanding their infrastructure under contract, and then the data center shuts down operations right after…
Eating, sleeping, even taking a crap — AI'll do all of that too.
Humans won't have to do a damn thing.
♪ When you smile, even more happiness spreads across this whole world ♪ (quoting a pop song lyric)
It's simple, really.
If this turns mainly into a race between nations, it'll spiral into an Iran-Israel-style mess.
The only way forward is building a system where no single country gets the upper hand.
Feels like regulation will kick in once things get dangerous enough, and that's when the AI bubble pops.
The nuclear arms race, though, kept right on going until the Soviet Union itself collapsed.
Once the singularity is achieved, it stops being a question of whether they 'made their money back.'
Still, it's pretty odd to be gambling everything on scaling laws when the underlying theory hasn't caught up qualitatively.
Put another way, that's an opportunity for Japan too.
Someone could build software that logs what employees actually do all day. A company quietly installs it, and a week later — tap on the shoulder, you're let go…
Japan's so far behind that it's mostly just using this stuff rather than building it — doesn't that make the opportunity pretty thin?
Gemini's supposed to be the worst offender for hallucinations, so… yeah, that tracks.
Gemini is the biggest liar of the bunch.
And if you're bashing AI without even having used a paid version, that's a waste — you're missing out.
I've used AI for stock trading and made about $2.7 million (¥400 million) over two years.
Though that edge is starting to fade too.
If you think of today's money-losing AI space as the 'hardware' layer, isn't there also a viable strategy of attacking just the 'software' layer?
Typing prompts in right now isn't all that different from using MS-DOS.
Building the Macintosh or Windows equivalent looks like the more profitable move (lol)
Chinese AI makers are the ones attacking on the software side.
Companies like DeepSeek can't get their hands on cutting-edge NVIDIA GPUs because of US sanctions, so instead they've improved their AI algorithms to get results on older chips that rival American cloud AI.
And on top of that, they're releasing it as open models, so anyone can run it locally for free.
How long have people been crying wolf about this?
If the bubble's going to burst already, just hurry up and burst.
Bubbles, though — even when everyone's saying 'this is bad, this is bad,' they tend to keep going for a while yet.
A mid-sized AI facility runs in the 100MW–300MW range.
Needing an entire gas power plant just to run a single facility only adds more fuel to rising energy costs.
Exactly right.
On graphics cards, there are several centimeters of distance between the GPU and the HBM memory, so they have to crank up the voltage to communicate across that gap — which burns a massive amount of power.
Apple's Macs get around that bottleneck: thanks to 2.5D memory-stacking technology that integrates the CPU and GPU into a single unified memory pool, power consumption is dramatically lower, making it the optimal setup for running LLMs locally.
Tim Cook isn't just selling the iPhone that Jobs designed — he's built a genuinely brilliant business that exploits NVIDIA's weak spot.
Where's the AI actually going to make money?
The free version of Gemini does everything I need.
Whichever country or company seizes dominance is expected to take all the profits — so it's a game of chicken that keeps going until the rivals collapse.
Like a prisoner's dilemma — everyone has to keep playing chicken just to keep competing with their rivals.
Anthropic actually turned profitable in the April–June quarter.
Quite the contrast with OpenAI, which significantly widened its losses.
Background and Key Points of This Discussion
This estimate was put together by US-based Bain & Company in the second half of 2025. It holds that to recoup the enormous capital being poured into the worldwide data center construction boom, AI-related businesses need to generate roughly $6 trillion in annual revenue by 2031. Current AI company revenues fall far short of that figure, and while many thread commenters reacted with ‘that’s less than I expected,’ plenty of others pointed to constraints beyond mere profitability — strained power and water resources, and the risk borne by power companies that have already committed capital to expanding infrastructure for data centers. Some also noted that Anthropic turned profitable in the April–June quarter of 2025 while OpenAI’s losses widened — a lesser-known fact showing that even AI companies can have very different revenue structures.
*This article is excerpted and summarized from the 5ch (Breaking News+) thread ‘Justifying the world’s AI data center investment would require $6 trillion (¥944 trillion) a year in revenue.’
