Bain & Company’s seventh annual Global Technology Report, published today, puts a number on the question hanging over the AI buildout. The number is $6 trillion. The AI industry must generate that much in annual revenue by 2031, Bain calculates, to justify the capital pouring into chips, data centers, and power. Existing consumer and enterprise AI applications can plausibly deliver $1.2 trillion to $1.8 trillion of it. Filling the remaining $4.2 trillion requires products and markets that largely do not exist yet.
“The debate today is fixated on employee productivity. The economics of AI infrastructure demand trillions in new revenue beyond productivity gains,” David Crawford, chairman of Bain’s global Technology practice, said in the report. “What the industry needs is a wave of innovation that will dwarf what mobile and cloud unlocked.”
The math behind the number
The $6 trillion figure rests on a simple ratio. Bain projects annual AI infrastructure spending of $1.5 trillion by 2031. That covers new data centers, compute capacity, and upgrades to installed GPUs, memory, and networking. Sustaining that spend requires an AI market approaching $6 trillion a year, assuming capital expenditure stays near 25% of revenue. Bain calls that ratio ambitious but reasonable for cloud providers.
Some of that revenue is already visible. Bain estimates consumer AI products, through subscriptions and advertising, could generate $200 billion to $400 billion by 2031. Enterprise AI could add another $1 trillion to $1.4 trillion for providers. Its reach spans software development, sales, marketing, customer service, and IT operations. The rest is the gap.
A requirement that tripled in a year
The bar has moved sharply in twelve months. Bain’s sixth report, released in September 2025, calculated that meeting 2030 compute demand would require roughly $2 trillion in annual revenue. It pegged yearly capital spending at $500 billion. Even assuming companies reinvested all their AI savings, an $800 billion shortfall remained.
One year later, the projected annual infrastructure spend has tripled to $1.5 trillion. The implied market requirement tripled with it.
The buildout behind those numbers keeps accelerating. Bain says hyperscaler capital expenditure could reach $780 billion in 2026, nearly five times the level of three years earlier. Leading-edge AI data centers are approaching 1 gigawatt today. Bain expects 2-gigawatt facilities by 2027 and 9-gigawatt campuses by the end of the decade.
Categories that fill barely a third of the gap
Bain names four categories meant to supply the $4.2 trillion in new revenue. Model providers replacing traditional search with advertising-funded chatbots could unlock $100 billion to $200 billion. Autonomous vehicles, trucks, drones, and industrial automation represent a $400 billion opportunity. Physical AI — simulations, digital twins, and robotics — could deliver $900 billion across automotive, electronics, semiconductors, and aerospace. Bain’s assumption there is a 10% cut in R&D and manufacturing costs.
Add those up, and the quantified categories reach roughly $1.5 trillion. Even in Bain’s own framing, about $2.7 trillion of the gap depends on products that do not exist today. The candidates include AI-driven drug discovery, always-available mental health support, and new forms of energy generation. The report is candid about the wager this represents.
Tripling the world’s compute in five years
On the supply side, Bain’s Data Center Model projects $5 trillion to $6.5 trillion of spending by 2030. That would build nearly 150 gigawatts of new capacity, almost tripling global capacity in five years. Four constraints bite at once.
Grid connections take more than four years, and critical equipment such as transformers is scarce. Chips are locked in years ahead through long-term supply agreements. Skilled trades are in short supply. Permitting is getting harder, not easier.
Public resistance is now a measurable economic force. Local opposition blocked or delayed at least 75 projects worth $130 billion in the first quarter of 2026 alone. That nearly matches the $156 billion disrupted in all of 2025.
In July, New York became the first US state to pause construction of the largest data centers. Fourteen other states have introduced similar legislation. Ireland, the Netherlands, and Denmark have restricted or paused new grid connections.
Hardware strikes back
The compute boom has reversed the tech sector’s value hierarchy. Hardware and semiconductor stocks grew at a 24% compound annual rate from 2020 to 2026, against 6% for software. Value has pooled in the bottleneck technologies.
The memory makers are the standout. SK Hynix and Micron have posted record gross margins of 75% to 85% for two straight quarters. The driver is high-bandwidth memory demand.
Custom silicon is the fastest-growing segment of data center compute. Hyperscalers and AI-native firms now run homogeneous workloads at volumes that amortize a custom design. That has fueled Broadcom’s ASIC business and carved niches for Groq and Cerebras.
One side effect is less benign. The major DRAM players have concentrated capacity on HBM, leaving conventional DRAM and NAND underinvested. Bain warns the shortfall is worsening shortages and raising prices for smartphones and PCs.
Cybersecurity at machine speed
The report’s starkest claim concerns attack speed. Bain finds AI has compressed the time a typical cyberattack takes, from about four weeks down to roughly 18 hours. The spread of AI agents widens the exposure. The context makes the figure credible.
Anthropic restricted its cybersecurity-focused frontier model, Claude Mythos, to a vetted partner program. A preview of the model had identified thousands of zero-day vulnerabilities across every major operating system and browser.
Defenders are retooling. Some companies deploying AI for vulnerability scanning have seen alerts increase by as much as eightfold. Leaders have raised remediation budgets by double-digit percentages. They have also redirected as much as 20% to 25% of cybersecurity staff to remediation work. The priorities are the oldest problems in the estate: legacy platforms, network layers, and SaaS vendors.
The productivity gap
For all the spending, measured productivity gains remain modest. Bain’s Tech and Engineering Survey of 293 senior technology leaders asked about expectations for the next one to two years. Respondents anticipate a 148% improvement in release-cycle speed and a 95% uplift in developer productivity. What they are capturing today is between 20% and 27% across key metrics.
The reason, Bain argues, is that AI shifts bottlenecks rather than removing them. Code arrives faster, then queues for review, coordination, quality checks, and governance. Independent telemetry from Faros — two years, 22,000 developers — found review time rising 91% as AI-generated code swells pull requests. Companies that redesign the whole development lifecycle capture far more of the gain. Amazon, for instance, now has AI handling up to 70% of its code reviews.
Editor’s Note
This article is based on a press release for Bain & Company’s seventh annual Global Technology Report, published September 29, 2026. TechRecast also reviewed the report’s published chapters, Bain’s prior-year report, and third-party coverage including Bloomberg via The Japan Times.
All figures are Bain estimates unless otherwise attributed. The $6 trillion revenue requirement, the $4.2 trillion gap, and the four new-revenue categories are projections, not confirmed outcomes. The Faros review-time statistic is Faros’s own telemetry. A summary circulated with the release cited a “21% more tasks” developer figure that does not appear in the release itself, and it has been omitted here.