AI market needs to make $6 trillion a year by 2031 to fund its infrastructure habit

The AI industry must generate $6 trillion in annual revenue by 2031 to keep funding the infrastructure needed to meet anticipated demand, and this will require imaginative new uses beyond simply enhancing employee productivity. That $6 trillion figure comes from the 2026 annual Global Technology Report from management …

[breakfast]

A lot of the money they can account for seems to be coming from things it is increasingly clear AI can't do. Surely by this point anybody involved in investing around the sector must be doing everything with one hand on the escape-pod activation because this has to go and it's probably going to be quick once it happens.

[DS999]

It matters a lot of it is your own money, like it is for Google, Facebook, Amazon, and Microsoft. Less so if you are OpenAI or others living on borrowed money. But inflation is only a short term aid, because inflation raises interest rates so it will be reflected in your interest rates when you sell a 10 year bond next year, and the year after.

So no, inflation is not going to bail them out. It only would if it was a one time or otherwise short term borrowing, which they'd pay back later. Inflation benefits you if you take out a mortgage and pay it back over time. It doesn't benefit you if you take out a new mortgage every few years.

[mcswell]

Escape pod

An escape pod jettisoned during the fighting, but no life forms were on board.

[Jedit]

"no life forms were on board"

Thumbs up for the reference, particularly as these are most certainly not the droids we're looking for.

[drankinatty]

They are just praying the IPOs can be done quickly so the first-round investors can cash-out before the hype-party collapses along with the global economy that takes insurers and pension plans down with it. Just in time for the last batch of boomers to retire.

The costs and expenses are real, the revenues aren't. Vapor-ware and vapor-revenue go together. There is no magic-lamp or genie big enough to grow $40B in annual revenue to $6T in less than 5 years.

Buckle up...

[Eric 9001]

They want to US government to bail them out with everyone else's money - if successful, lack of profitability wouldn't be their problem.

Do they actually make 40 billion of revenue?

[Like a badger]

At the moment, the annualised run rate for the big AI players looks to be around $150-220bn (according to the Economist). I don't know what proportion of those revenues is real, paying end users as opposed to the AI sector's incestuous deals and unbacked promises. If we take the lower figure as more reliable, then the starting position is $150bn. Even that is problematic, because it isn't clear that (a) quite a lot of real revenues aren't cannibalistic (eg advertising revenues previously booked as search and placement, now with some weak AI link and fully booked as AI revenues), and (b) not clear that the end user is actually getting any value from what they're using/paying for - in which case the revenues aren't enduring.

There's nothing inherently wrong with the maths of what Bain have done here, they're doing some projecting forward from a range of current trends. It isn't really intended to be a projection, but to encourage debate and get the C-suite on the blower to Bain to get some answers. The analysis doesn't project the high likelihood that some of the AI players will fall out of the race, assumes that low cost or vastly more efficient alternative models don't disrupt the trends, that we don't suffer a global recession and/or see the AI bubble burst, and that $4.2 trn of revenue can appear like magic. There's also the challenges that if AI displaces meatsacks without creating an equivalent value in new employee roles then it'll result in social unrest and government intervention. It is possible to draw unwritten conclusions from the report - one of which is that AI is all a huge unsustainable bubble that won't ever be able to support the investment and debt that it needs. At the other end, there's the hope that somebody finds $4.2 trn worth of new and currently unforecast uses attached to paying customers.

[AVR]

Annualised run rate is so extremely manipulatible (for any given company they find their big shots of income, maybe encourage them to come at the same time with time-limited discounts & marketing, draw the narrowest possible date range which includes those, multiply the income as if those 13 days or whatever were representative of the year to get the ARR) that the real figure could be way under the low end of that range. Unless we got a real insight into the actual books treat that number as aspirational rather than real. Not saying the other factors you name aren't also issues, it's just that ARR is a terribad metric even before considering them.

[PghMike]

4.2T in new business

The thing is, when you look at Bain's guesses as to the sources of new AI revenue, many of them don't require large models running in data centers.

They're things like new scientific discoveries, self-driving everything, and AI in manufacturing.

Scientific discoveries, i.e. things like AlphaFold, aren't driven by LLMs, but by specially tailored models that are likely much cheaper to run. Self-driving and AI in manufacturing aren't going to be LLM-based and likely aren't going to be data center resident -- you don't want your car to crash when it enters a cell dead spot.

Bain's enterprise numbers seem ambitious as well. $1.4T worldwide means about 900B in US revenue. Assume that half of the 130M jobs in the US are amenable to augmentation by AI. Then you're talking about $14K per employee per year in AI spending. Yet all the programmers I know who are using AI are either spending $240/year or $2,400 for Claude.

So, all in all, I don't even believe the current enterprise estimates are likely to be reached, much less the futuristic ones. Even the consumer ones look like about $200B/year in the US, which is $700 per year, or $58/month. I mean, it's not inconceivable, but when Google search is free and gives you AI summaries, I'm not feeling it.

Add in uncertainties from liability suits, copyright suits, competition from cheaper models, and it's hard to see a good ending here.

[Tron]

Not that bad.

Anyone picking AI shares for a pension plan should be preparing to spend the rest of their life in prison or in hiding. Pension plans invest in dull, rock solid, blue chip stocks. So that means nothing that comes close to AI. Most of this is fantasy money - promises of future cash to keep the scam going. GAFA didn't do AI themselves for a reason and will have insulated themselves from serious losses. The stock market will reset and there will be lots of panic on the news, but it will be nothing compared to cons like the sub-prime mortgages or Brexit.

[AVR]

Re: Not that bad.

In NZ, pension plans (Kiwisaver) which avoided AI were criticised for their relatively less impressive returns. The market leader one year being middle of the pack next after doing so, that kind of thing. Investment talking heads have zero shame so some of them are now mentioning the 'possible' AI investment bubble, but there will in fact be people whose pensions will be hurt by this.

[doublelayer]

Re: Not that bad.

Apple, and to a lesser extent Amazon, haven't spent as much of their cash on LLMs directly and might avoid this. Google did their LLM stuff within the Alphabet umbrella, and Microsoft has heavy investments in both OpenAI and the datacenters that they use. Those companies are not immune. Companies that are even less immune include Oracle, Nvidia, TSMC, and the RAM manufacturers. Of those, probably only Oracle is at risk of losing the company*, with the rest having somewhat wisely sold lots of shovels during the gold rush and saved enough of the money, but pension funds or any other investor still stands to lose because much of the value of their investments in those companies is speculation from other investors that the growth will keep going. Nvidia with all that money but without the guarantee that they can sell every GPU they could possibly manufacture at high margin will be worth much less to the funds and the people who rely on them. That's without also including Spacex, which wise investors would be advised to stay away from, but it's in tech index funds because a special exception was made for it, and it is heavily exposed to both AI risk and Musk risk.

[Anonymous Coward]

“ The AI industry must generate $6 trillion in annual revenue by 2031 to keep funding the infrastructure needed to meet anticipated demand, and this will require imaginative new uses beyond simply enhancing employee productivity.”

Revenue is not profit.

What is the bottom line needed??

I fear AI may exceed the planets GDP to keep it fed.

[Pete 2]

Peanuts

> The AI industry must generate $6 trillion in annual revenue

Considering the world's total GDP is $126 tn now, another $6 tn is little more than a rounding error.

It is also said to be increasing by 3.4% per year. Whether any / some / all of that increase is due to AI is unknowable, but it is a trivial exercise to arbitrarily say that $6tn is due to AI (or any other new development)

[Anonymous Coward]

Re: Peanuts

“GDP stands for Gross Domestic Product, which is the total monetary value of all finished goods and services produced within a country's borders during a specific time period.

What the Words Mean

• Gross: The total output before taking away anything for wear and tear on machinery or buildings.

• Domestic: Production that happens inside a country's physical borders, no matter who owns the company.

• Product: The actual final goods and services created and sold in the market.”

Not seeing any mention of Profit or ROI anywhere…. Just bullshit like ARR.

[Anonymous Coward]

Re: Peanuts

The total global economic output of ~$126.3 trillion can be categorised by how households spend money or by industry value. In economics, this is viewed through Personal Consumption Expenditures (Household Spending), which drives roughly 58% to 60% of the entire global GDP.

The remaining 40% consists of business investments, government spending, and international trade

Estimated Global Consumer Expenditure Breakdown

1. Housing, Shelter & Utilities (~20%–22% of Consumer Spending)

Real estate and shelter represent the single largest economic footprint in the world.

• The Breakdown: This includes gross rent, tenant utilities (electricity, water, gas), and residential construction.

• Economic Impact: The broader global real estate sector (including commercial property) accounts for an asset value that dwarf total stock and bond markets combined.

2. Food, Agriculture & Groceries (~15% of Consumer Spending)

Feeding the planet is one of the foundational drivers of global trade and employment.

• The Breakdown: Raw agriculture itself makes up about 4.1% of direct global GDP, but when scaled to include food processing, shipping, grocery retail, and restaurants, the entire "food value chain" represents roughly $8 trillion to $10 trillion.

• Employment: Agriculture alone employs nearly 29.5% of the world's workforce.

3. Cars, Mobility & Transport (~10% of Consumer Spending)

Getting people and goods from point A to point B is a massive driver of industrial output.

• The Breakdown: The automotive manufacturing sector is roughly a $3 trillion global industry.

• The Multiplier: When including downstream transportation services (airlines, trains, freight cargo, auto repairs, and fuel), mobility accounts for a massive slice of the services and durable goods sectors.

4. Healthcare & Medical Services (~10%–12% of Consumer Spending)

As global populations age and middle classes expand, medical care has become a dominant force in modern GDP.

• The Breakdown: This tracks pharmaceuticals, hospital care, health insurance, and medical devices.

• Regional Divide: In advanced economies like the United States, healthcare consumption is much higher (~17% of national GDP) compared to developing nations where food and raw essentials take up a larger percentage.

5. Everything Else (~41%: Entertainment, Tech, Clothing, & Services)

The remainder of global output goes toward non-essential items, services, and lifestyle goods:

• Tech & Electronics: Smartphones, computers, data networks, and software.

• Apparel & Retail: Clothing, footwear, and consumer goods.

• Services: Financial banking, legal aid, tourism, education, and entertainment.

Asking AI of course / LOl.

[Pete 2]

Re: Peanuts

> Not seeing any mention of Profit or ROI anywhere

That is because the article is not about either of those things. You seem to be getting angry about something that isn't here.

[RLWatkins]

Re: Peanuts? Nope.

In the US that amounts to $19,000 for every adult and child in the country.

Divided between the US and the EU? That's $8,000 for every adult and child.

I'm not buying in. Nothing that "AI" does is worth that much to me.

[doublelayer]

Re: Peanuts

4.8% is not a rounding error. 4.8% of your annual income would be a substantial expense, though not your largest.

Arbitrarily declaring a number as AI-related is easy. It's also useless, since AI companies need hard cash because they have real bills. If they don't have it, someone else saying that some number is AI-related won't help them.

[Anonymous Coward]

Exactly. They seem to be equating LLMs with true AI, which isn't a thing!

[Jou (Mxyzptlk)]

Inflation.

The solution is inflation. Again. The bills they have to pay in ten years don't matter if you calc in the inflation. And since Petrol Dollar ist still alive...

Of course that does not help the normal people of the USA, especially not the lower income ones.

[Jam2]

Re: Inflation.

The current GDP for the U.S. is I think around 30T. So for ai/si to control 6T is I think, allot more integration than we have been seeing. I will be waiting patiently for the results. Also is 6T actually a reliable figure?

[Jou (Mxyzptlk)]

Re: Inflation.

You need to calc in the GFD, "Gross Federal Debt". 123% in 2025, and will be worst for sure in 2026. Was around 40% to 60% from 1940 to 2008 (before Bankster crash).

This ONLY works due to the insane inflation making the debt (a bit) smaller. And ONLY due to petrol dollar and a few other things, which DT is currently dismantling.

One of the many sources: https://fred.stlouisfed.org/series/gfdegdq188s

[Excused Boots]

Re: Inflation.

"Of course that does not help the normal people of the USA, especially not the lower income ones.”

I think the current administration works on the basis that the normal people of the USA are all tech billionaires. Everyone else is an anomaly.

[Anonymous Coward]

"Solution looking for a problem to solve"

Now with needing treble the money to just break even. And that much more pressure to find juicy moneyed problems to milk.

Expect much more aggressive pushing of AI everywhere, and prices of everything to go up accordingly.

Something something shoe event horizon. Best start growing wings today, eh.

[Excused Boots]

AI market needs to make $6 trillion a year by 2031 to fund its infrastructure habit

1) Shirley you mean SI - wouldn’t want to contradict the 'Great Leader’ would you?

2) Won’t happen.

[You aint sin me, roit]

Re: AI market needs to make $6 trillion a year by 2031 to fund its infrastructure habit

Won't happen? But, but, but... His Muskiness said the total addressable market for his SpaceX (an AI company, not a rocket company) IPO was $28tn.

That wasn't just a made up number was it?

[Cubbie Roo]

According to The Economist, Anthropic’s annualized run rate has been plateauing since June, even while they can pretty much manipulate that figure to their hearts desire.

So that looks like the Claude market is already topping out.

Where's the revenue when 'foreign' models just get cheaper and better, and smaller local models improve drastically?

[brainwrong]

Madness

"mental health support"

So it'll take your job, then help talk through your anger and unfulfilled life.

Sounds dreadfull.

[Anonymous Coward]

Re: Madness

Don’t forget the AI porn to give you something to do with your hands!

[Aladdin Sane]

Re: Madness

Your plastic pal who's fun to be with.

[I ain't Spartacus]

Re: Madness

Sounds ghastly!

[Andy 73]

We did the maths

We did the maths in the comment section a year ago - income from AI is in the billions, not trillions and most of the market at the moment is trying to reduce the subscription cost since the promised valuable use case hasn't emerged.

Instead it's acting like any other piece of productivity software with no significant moat. That means the price can't be magically inflated, and the real risk to the hyperscalers is that credible open weight models enable companies to run locally at a fraction of the price. There's a reason they're all calling for the market to be regulated - it's to protect their monopoly.

There's no doubt this is going to continue to have a catastrophic effect on employment, particularly amongst knowledge and creative workers. However, in the middle of a looming global economic crash, that doesn't free up enough money for AI services, and nor does it usher in some imaginary nirvana of productivity.

The question is really how long they can maintain the illusion that this is going somewhere that makes the IPOs and capital expenditure make sense.

[cyberdemon]

Re: We did the maths

This, absolutely this. Only three fucking orders of magnitude difference!

But what's a factor of a thousand between friends death-cultists, eh?

[Aladdin Sane]

Re: We did the maths

https://isaiprofitable.com/

[breakfast]

Re: We did the maths

Meanwhile Chinese models are getting better and staying cheap and local-models are getting towards where they can do most of the same work. The future from here is very unstable and the one projected here strikes me as deeply improbable.

[Groo The Wanderer - A Canuck]

Fantasies and stock market bubbles that are increasingly isolated to that of the Americans... classic pyramid scam bait.

[PghMike]

$6 trillion

Ladies and gentlemen, let's do some math.

Population of EU, JP and US is 900M people. Assume half of them are working, and perhaps half of those are working at a job that AI could meaningfully contribute to. So, 225M people in AI impacted jobs in those countries, which are most likely to be using AI extensively in the next few years.

Assume every single one of them is using AI. $6T/225M = $27,000 per year per employee.

Right now, a lot of people are paying $20/month plus API usage.

Even using today's numbers of about $500B to $1T per year, gives numbers in the range of $2,200 - $4,500 per year, which sounds pretty high for an industry where most people today are paying $240/year if they're paying for it at all.

And their competition, open weight models, are close to free. And if the NY Times wins their suit, these guys are going to have to retrain all their models on only data licensed for training. Not to mention that a lot of their training was done on whose initial copy was stolen.

I understand why these guys want to IPO quickly, before anyone manages to do any arithmetic. Because if they do, no way they're getting a $2 trillion market cap.

[Eric 9001]

Re: $6 trillion

The number works out to ~$26,666.66 - which is more fitting than rounding it up to 27k.

Still proprietary, but not SaaSS, local models aren't anywhere close to gratis - it takes very expensive hardware and a significant amount of power to run such models.

(Sure, there's a LLM model that can run on a CPU, but that's practically useless for anything but chatbot entertainment purposes (oh wait, that's all LLM's)).

You cannot steal data by copying it (as the one you copied it from still has their copy - there are now more copies - which is usually a good thing) - but you can carry out intentional copyright infringement and/or plagiarism, for profit.

Another malicious thing these companies do, is to scrape websites so hard that badly written software is DoS'd (but that's still not theft).

As for the NY Times case, unfortunately, the US court system could decide that making prohibited copies of works and then degenerating an obfuscated copy with a LLM is "fair use", just like how a US court decided that degenerating obfuscated copies of software without including the copyright information is "fair use" (see Doe v. GitHub).

[Anonymous Coward]

Re: scrape websites so hard that badly written software is DoS'd

sooooo.... Apache and Postgres?

I'll grant I didn't spend a lot of time tuning my Postgres cluster, but when the rate of requests due to probably-AI run a four-core machine from a baseline load average of not quite zero - including a long history of more conventional search-engine bot scrapers, which I've been fine with - up to a load average of 40+, and I can't use my own services on that machine because my attempts can't get a database connection because of the sheer volume of connections from slop machines, I'm not going to point the finger at anything on my end. There's literally no justifiable excuse for the type of traffic these training and inference systems are sending.

I could give this VM the entire 48 vcores and 128GB (modulo a little for Proxmox) and if I hadn't started bouncing certain useragent patterns to a "Bad Bot" page, I'd still be seeing issues.

[Anonymous Coward]

Re: $6 trillion

What is the profit in this ???

We can see the cost and questionable ROI ….

You’d be better spending the money on solar/wind/hydro/nuclear/thermal to remove costly and volatile Carbon fuels from the economy … and that’s before you account for the pollution and climate benefits.

[VoiceOfTruth]

Consumers, the public in general

>> via subscriptions and advertisements

People I speak to are fed up with both.

Let's take that $1.2 trillion low end figure for a moment, and divide that by the population of the USA. That's about $3,500 per person, young or old. Each year. Meanwhile, a full Adobe subscription costs about $840 a year. We know already that babies are not going to subscribe. Uncle Bob is tapped out. So that means higher per person costs. Where are these people who will fork over another $3,500 a year? Some will, sure. But if you think a new iPhone every year is not for you, then how about twice that?

[EricB123]

Mental Health Support?

So the solution to this problem is the same thing that largely made it happen? Sounds circular to me.

[mihares]

Cloud AI is just

Going to die. Edge models can already execute on hardware very motivated consumer can buy and are already “not bad” if you are in that sort of things.

With another couple of generations nobody is going to bother paying the must-be-ever-increasing bill for cloud AI.

I predict that by 2031 most of the scum bAIgs will have kicked the bucket hard.

[Professor Woozle]

Fantasy economics

The figure about how much they're going to need to bring in is probably about right, but the stuff about where the revenue's going to come from is total fantasy, and the whole AI house of cards is probably going to come crashing down next year - if either company go for IPO it'll be immediately apparent how ludicrously bad their finances are and also just how much they're on the hook for all the unbuilt data centres they've contracted for with no way to pay for them. I strongly advise anyone who wants to know exactly how bad things are to read Ed Zitron's blog https://www.wheresyoured.at/ , the latest piece (Dead Money) is a good start but there's a shedload of detail and analysis in the other pieces, showing strong parallels to both Enron and the subprime mortgage crisis.

Basically, the AI fanboys don't want to admit that the emperor is not only wearing no clothes, he also has appalling diahorrea ...

[Thousandwich]

Re: Fantasy economics

Always enjoyable to read Ed Zitron's material as he has been warning everyone about this for years.

There is no real demand for AI services that can't already be satisfied with existing infrastructure. Everything else is overblown hype.

We really should stop referring to this stuff as AI as it isn't, and will never be, "artificial intelligence".

[Bebu sa Ware]

"the kind of answer that your granny could have come up with."

Not mine, not even when she was alive.

She could have competed in the Doolally Olympics and robbed the US of gold all round.

It has been said that the Soviet Union couldn't match US military expenditure and ultimately collapsed trying. Ironically one might imagine the US is doing the same thing to itself with AI (Supremely Imbecilic) expenditure.

The only brains in the game appear to be the Gen-α whose current put·down "That's so AI" speaks volumes.

[Tron]

Tulips, railway mania, the South Sea Bubble and AI.

Trump needs AI to keep going because it covers up the recession. He can ban Chinese AI and order his 'allies' to ban it too, as the US did with Huawei, but I'm not sure there are enough rich mugs in the world to last that long.

In the 80s we had Expert Systems. AI is an attempt to create a Universal Expert System. Not possible. Not even by adding a natural language interface to make it look like magic.

They needed to be a lot fussier about source material and needed to accept the limitations. An expert system full of builder's regs is fine, but history is contested, religion likewise. Different people accept different 'truths'. In science too. Even in America, there are plenty of people who believe evolution to be a fake scam.

The other issue is that all of this stuff is obsolete. At best you are freezing human knowledge, and making it subject to national and religious ideologies to flog it. All great scientific achievements come by disproving accepted science facts: It was an accepted scientific fact that God made the world in 7 days and the universe revolved around the Earth. Bake that into your LLM to keep the punters happy, use it in the classroom and at Uni, and no Darwin, no Galileo.

So, small scale expert systems are fine for bulder's regs, and even code development, but take it any further and your market shrinks and demands that it embraces their prejudices.

And you will not get anything approaching an expert system by trawling social media posts and scanning/destroying old text books.

Is any of that worth x billion? No. As a product it is part fantasy, part scam. Get your cash out asap. Because when this goes, all the apps will lock up and the stock markets will freeze trades. And you will be screwed.