In its seventh Global Technology Report, released on September 29, Bain & Company estimated that the AI industry will need about $6 trillion (roughly ¥941 trillion) in annual revenue by 2031.
However, if you add the upper limits of the existing uses and the three growth areas for which the report gives specific dollar amounts, the total reaches only $3.3 trillion. The gap to $6 trillion is about $2.7 trillion, or roughly 45% of the total. That gap would have to be filled by "new product development," for which no dollar figure is given, and by the upside in search and advertising, which Bain puts at "$200 billion or more."
The key point is that the $6 trillion figure is not a bottom-up sum of individual AI demand sources. Bain estimates that AI infrastructure investment in 2031 will need to reach $1.5 trillion, then works backward from the assumption that capital expenditure equals about 25% of revenue.
Last year's edition said $2 trillion in revenue would be needed to support $500 billion in data center investment. Compared with this year's pairing of $1.5 trillion and $6 trillion, capex works out to about 25% of required revenue in both cases. The main reason the required revenue rose from $2 trillion to $6 trillion is that the underlying investment figure grew.
Bain's "$6 trillion a year" estimate: existing uses are expected to reach at most $1.8 trillion
Bain believes annual AI infrastructure investment could reach $1.5 trillion (about ¥235 trillion) in 2031.
This figure covers not only new data centers and computing capacity but also the replacement of existing GPUs, memory, networking equipment, and other hardware.
Assuming that capital expenditure (capex) equals about 25% of industry revenue, working backward gives a required AI market of about $6 trillion a year. Yen conversions use ¥156.91 to the dollar as of 5 p.m. on September 30, 2026.
On the other side, Bain sizes AI uses that are extensions of today's at $200 billion to $400 billion for consumer subscriptions and advertising, and $1 trillion to $1.4 trillion for enterprise.
The enterprise figure assumes the portion of AI-driven productivity gains in software development, sales, customer service, and similar areas that accrues as revenue to AI service providers.
Together, the two come to $1.2 trillion to $1.8 trillion.
Bain names four areas to bridge the gap between $6 trillion and these existing uses: search and advertising, autonomy, physical AI, and new product development.
For search and advertising, it gives $100 billion to $200 billion or more, assuming the introduction of ads in AI chat services and partial substitution for traditional search.
For autonomy, it sees a $400 billion market opportunity, premised on higher utilization of vehicles, drones, industrial equipment, and the like, along with lower training and operating costs.
Physical AI is put at $900 billion, an estimate of the economic opportunity if R&D and manufacturing costs fall 10% in sectors such as automobiles, electronics, and semiconductors.
For new product development, Bain cites examples such as drug discovery, mental health support, and new materials, but gives no specific dollar amount.
Itemized figures reach at most $3.3 trillion; the breakdown of the remaining ~$2.7 trillion is not provided
The arithmetic is simple if you add up the upper limits of the items for which Bain gave specific amounts.
Add $0.2 trillion for search and advertising, $0.4 trillion for autonomy, and $0.9 trillion for physical AI to the $1.8 trillion upper limit for existing uses, and the total is $3.3 trillion.
The gap to $6 trillion is about $2.7 trillion, or roughly 45% of the total.
This $2.7 trillion is not a figure Bain presented directly. It is our own calculation, subtracting the amounts in the report.
The $6 trillion itself is an approximation ("about $6 trillion"), and search and advertising is described as "$200 billion or more," so $3.3 trillion is not necessarily a hard ceiling.
If search and advertising far exceeds $200 billion, the amount that new product development and other areas must supply shrinks accordingly.
Conversely, if existing uses come in at the low end of $1.2 trillion and search and advertising at $0.1 trillion, the total is $2.6 trillion and the gap to $6 trillion widens to $3.4 trillion.
For that reason, we do not treat this as a definitive shortfall. We simply note that, when the upper limits of the amounts Bain specifically gave are added up, a gap of about $2.7 trillion remains. At the same exchange rate, that is about ¥424 trillion.
In Bain's chart, the roughly $4.2 trillion gap beyond existing uses is to be filled by search and advertising, autonomy, physical AI, and new product development.
Of these, new product development is expected to make a large contribution among the four areas, yet no specific market size is given. Bain says these four areas are likely to support growth in the AI market.
One caveat is that not all of the combined $1.3 trillion for autonomy and physical AI will necessarily become revenue for AI companies.
Bain derives these "market opportunities" from gains such as higher utilization and cost reductions for the companies using AI, but does not specify how much of that would be captured as revenue by AI service providers.
In that sense, the portion of the $6 trillion with concrete evidence as AI company revenue could be read as smaller than $3.3 trillion.
Working backward from "capex is 25% of revenue": how this differs from last year's $2 trillion
The $6 trillion Bain presents was not calculated by adding up demand by use case.
If capex is assumed to equal 25% of revenue, every dollar of capex requires four dollars of revenue.
Dividing the $1.5 trillion of 2031 AI infrastructure investment by 0.25 gives $6 trillion.
Naturally, changing this ratio changes the required revenue substantially. At a 20% capex ratio, it would be $7.5 trillion; at 30%, $5 trillion.
Bain characterizes the 25% level as an "ambitious but reasonable" assumption based on cloud providers' behavior.
In last year's edition, Bain estimated that new data center investment needed in 2030 would be $500 billion, and that $2 trillion in revenue would be required to support it.
This year's edition puts 2031 AI infrastructure investment, including the replacement of existing equipment, at $1.5 trillion and the required revenue at $6 trillion.
In both cases, capex works out to about 25% of required revenue.
However, last year's edition did not explicitly state the 25% ratio; it is implied by the figures of $500 billion and $2 trillion.
The target years and the investment scope also differ. Last year's edition centered on new data center investment, whereas this year's includes the replacement of existing equipment.
It is therefore not appropriate to say that required revenue "tripled from $2 trillion to $6 trillion," as if the same metric had simply expanded.
Bain estimates that 2026 capex by five companies, Microsoft, Google, Amazon, Meta, and Oracle, could reach $780 billion.
That is about five times the level of three years ago, and already exceeds the $500 billion in new data center investment that last year's edition said would be needed in 2030.
That said, the $780 billion is company-wide capex for the five firms, so its scope differs from the $500 billion, which covers only new data centers.
Under this backward-calculation method, if 2031 infrastructure investment exceeds $1.5 trillion, the required revenue at a 25% capex ratio would also exceed $6 trillion.
If the size of existing uses stays at $1.8 trillion, the portion without specified uses grows by the same amount.
At Amazon, CEO Andy Jassy raised the 2026 capex outlook from $200 billion to $220 billion, citing factors such as rising memory prices.
Microsoft has also explained that its $41 billion in April–June capex included the effect of higher component prices.
If rising component prices inflate capex itself, the required revenue implied by the 25% ratio also rises.
Bain also points out that memory shortages could worsen further and lead to higher prices for smartphones and PCs. Massive AI infrastructure investment may affect not only data centers but also the prices of devices Japanese consumers buy.
For Alphabet, Meta, Microsoft, and Amazon, April–June 2026 capex totaled $171.2 billion against combined revenue of $471.2 billion, a capex ratio of 36.3%.
That is above 25%, but the revenue denominator includes non-AI businesses such as advertising and retail. Because this differs from the basis of the "AI industry revenue" Bain has in mind, the figure cannot be used to simply recalculate the $6 trillion.
Could rapid growth at AI companies shrink the remaining ~$2.7 trillion?
The major counterargument is that AI companies' revenue is already growing rapidly, so Bain's $1.2 trillion to $1.8 trillion estimate for existing uses may prove too conservative by 2031.
Citing a New York Times report, Axios reported on September 18 that Anthropic's revenue has reached a pace of more than $100 billion on an annualized basis.
On September 29, Axios also reported, citing people familiar with the matter, that OpenAI's annual recurring revenue (ARR) is approaching $70 billion.
Simply added together, the two companies come to about $170 billion, equal to roughly 14% of the $1.2 trillion low end Bain gave for existing uses in 2031.
However, current annualized figures and ARR are defined differently from a 2031 market-size forecast. Nor does the current rapid growth necessarily mean the same pace will continue through 2031.
Still, if the AI services market grows faster than Bain assumes, existing uses alone could exceed $1.8 trillion, narrowing the gap to $6 trillion.
The roughly $2.7 trillion shown in this article does not deny the room for AI market growth. It is simply the gap that remains if you add up the items for which Bain's current report gives specific dollar amounts.
Because the required revenue itself is determined by the assumed infrastructure investment, delays in data center construction that shrink investment would also shrink the $6 trillion figure.
In its analysis of data centers, Bain says that in the January–March 2026 quarter alone, at least 75 data center projects worth a total of $130 billion were canceled or postponed due to opposition from local residents and others.
David Crawford, who leads Bain's Global Technology Practice, has said that AI infrastructure is being built ahead of demand growth, and that sustaining the investment would require economic benefits large enough to lift annual global GDP growth by about one percentage point.
To see whether this view changes, the key will be Bain's next Global Technology Report.
The past two editions were published on September 23, 2025, and September 29, 2026. Points to watch in the next edition are whether it uses the same 2031 framework, whether it raises the size of existing uses substantially from this year's $1.8 trillion upper limit, and whether it puts a specific dollar figure on new product development.
Bain believes that justifying massive AI investment will require a new wave of applications comparable to mobile and cloud.
How the remaining gap of about $2.7 trillion gets filled can only be verified once those new uses show up as actual revenue.