Investment in artificial intelligence (AI) infrastructure would need an AI market worth about $6 trillion a year by 2031 to be sustainable, and consultancy Bain & Company only clearly sees between $1.2 trillion and $1.8 trillion. That’s the central thesis of its Technology Report 2026, published on September 29, 2026, which leaves a gap of roughly $4.2 trillion that, according to its authors, will have to come from markets that don’t yet exist or are only just emerging.
Bain’s AI funding report in 20 seconds
- Capex at the five biggest hyperscalers could reach $780 billion in 2026.
- Bain projects $1.5 trillion a year in AI infrastructure spending by 2031.
- Sustaining that would require a $6 trillion-a-year AI market.
- Consumer and enterprise use would contribute at most $1.8 trillion.
The report, authored by David Crawford, Kristie Tagawa, Cory Boles, and Tatum Quinn, starts from a simple idea: until now the conversation has focused on whether that much capacity can be built, but the relevant question is whether AI will generate enough economic value to pay for it. Its conclusion is that productivity gains from today’s applications, for both businesses and consumers, won’t be enough.
How Bain gets to $6 trillion
The starting point is spending. According to Bain, capital expenditure at Microsoft, Google, Amazon, Meta, and Oracle could reach $780 billion in 2026, almost five times the level of three years ago. Leading-edge AI data centers are already nearing a gigawatt (GW) of power, and the consultancy expects many facilities to approach 2 GW in 2027, with 9 GW campuses appearing by the end of the decade.
To illustrate this, the report uses Epoch AI estimates of the world’s largest AI data center at any given point in time. It takes as a reference Meta’s Prometheus facility, which came online in 2025 with about 0.6 GW and a $24 billion price tag, and projects a 4-to-5 GW campus costing $120 billion to $175 billion by 2029, and another one of around 9 GW and $200 billion by 2030. It’s worth reading this with caution: the chart’s own footnote clarifies that the 2027, 2029, and 2030 figures are extrapolated from a trend of doubling every 12 to 16 months, not from announced projects. It’s the same clash between power, chips, and money that we already described in the collision between AI, electricity, and chips.
With that curve, Bain calculates that annual AI infrastructure spending could reach $1.5 trillion by 2031, adding new data centers, compute capacity, and the refresh of already-installed GPUs, memory, and networking. It then applies a key assumption: that capex equals around 25% of sector revenue, a ratio the authors call “ambitious but reasonable” based on what’s seen among cloud providers. That’s how the roughly $6-trillion-a-year market figure comes about. If the ratio were higher, the required figure would fall; if lower, it would rise.
What’s already visible, and what’s missing
Part of that money is already taking shape. Consumer AI products, through subscriptions and advertising, could generate between $200 billion and $400 billion by 2031. Enterprise adoption would bring providers another $1 trillion to $1.4 trillion, thanks to productivity gains in software development, sales, marketing, customer service, and IT operations. In total, that’s $1.2 trillion to $1.8 trillion, leaving about $4.2 trillion still to cover.
To close that gap, Bain points to four categories:
- Search and advertising ($100 billion to $200 billion or more): ads embedded in chatbots and the replacement of a large share of traditional search.
- Autonomy everywhere ($400 billion): cars, trucks, drones, robotaxis, logistics, and industrial automation.
- Physical AI ($900 billion): simulations, digital twins, and robotics, including humanoids, across automotive, electronics, semiconductors, and defense. The figure assumes a 10% cut in R&D and manufacturing costs.
- New products: drugs for rare diseases, mental health support, new materials for batteries and semiconductors, or autonomous scientific research.
This is the report’s most important nuance. The three quantified categories add up to between $1.4 trillion and $1.5 trillion. That means roughly $2.7 trillion, more than half the gap, depends on the fourth category, new products, which Bain doesn’t put a number on. The authors themselves frame it as an open question: infrastructure is being built ahead of demand, and the issue is whether applications will arrive in time to pay for it.
From $2 trillion to $6 trillion in a year
The comparison with last year’s report helps put the figure in context. In September 2025, Bain estimated that AI would need $2 trillion in annual revenue by 2030 to fund the computing power expected by then, around 200 GW of additional capacity worldwide, and calculated an $800 billion shortfall even accounting for the savings AI itself would bring. The two figures aren’t directly comparable, since they change the reference year, the spending scope, and the methodology, but the scale of the problem has grown.
Bain doesn’t call it a bubble, but its diagnosis fits the debate that has run through the sector for months over whether AI can be real and still fuel a bubble, and with the volume of chip purchase commitments already piling up as hyperscalers commit nearly $2 trillion. The consultancy argues that enterprise productivity is only the first visible benefit of AI, and that a wave of applications comparable to the one brought by mobile and the cloud is still needed. By its calculations, financing the infrastructure sustainably would require adding roughly one percentage point to annual global GDP growth.
Frequently Asked Questions
How much money does AI need to generate, according to Bain?
Bain estimates that an AI market worth about $6 trillion a year by 2031 will be needed to sustain $1.5 trillion in annual infrastructure spending, assuming capex equals 25% of revenue.
Where would that revenue come from?
Between $1.2 trillion and $1.8 trillion from consumers and businesses. The remaining roughly $4.2 trillion would need to come from search and advertising, autonomous systems, physical AI, and new products.
How much are hyperscalers investing in 2026?
According to Bain, capex at Microsoft, Google, Amazon, Meta, and Oracle could reach $780 billion in 2026, almost five times what was invested three years earlier.
What did Bain’s 2025 report say?
That AI would need $2 trillion in annual revenue by 2030 and that there would be an $800 billion shortfall. The 2026 calculation uses a different year and scope, so it isn’t directly comparable.
Sources:
