The AI industry’s next challenge may be less about building compute capacity than generating enough economic value to support it. Bain & Company’s seventh annual Global Technology Report estimates that AI would need to generate about $6 trillion in annual revenue by 2031 to justify the infrastructure investment now underway, with existing consumer and enterprise applications potentially accounting for only $1.2 trillion to $1.8 trillion.
The artificial intelligence industry is entering a phase where infrastructure scale and economic returns are becoming increasingly connected.
Bain & Company’s seventh annual Global Technology Report, released September 29, says the AI industry would need approximately $6 trillion in annual revenue by 2031 to support the capital being deployed to satisfy growing compute demand. Existing consumer and enterprise AI applications could generate between $1.2 trillion and $1.8 trillion, leaving roughly $4.2 trillion that would have to come from new categories of products, services and business models.
The finding shifts part of the AI conversation away from employee productivity and toward whether entirely new markets can emerge quickly enough to absorb the infrastructure being built today.
Bain’s analysis identifies four areas that could contribute to that expansion: AI models replacing or augmenting search and advertising, autonomous systems, physical AI and entirely new products enabled by abundant machine intelligence.
AI infrastructure needs more than productivity gains
Much of the current enterprise AI discussion centers on automating software development, customer service, marketing, sales and IT operations.
Those applications are expected to become substantial businesses. But Bain’s research suggests they alone may not generate enough revenue to support the scale of infrastructure investment being contemplated.
The consulting firm estimates annual AI infrastructure spending could reach as much as $1.5 trillion by 2031, while cumulative data center spending could reach $5 trillion to $6.5 trillion by 2030.
That creates an economic requirement for AI applications to expand beyond existing enterprise workflows.
One potential source is the evolution of AI models into consumer-facing platforms that combine answers, discovery and advertising. Another is autonomous transportation and industrial automation, where AI becomes part of a physical product rather than an application running on a user’s computer.
Bain also points to physical AI, including robotics, simulations and digital twins. These systems could connect AI models to manufacturing, engineering and research processes, creating applications whose economic value is not captured by conventional software productivity measurements.
The fourth category is less defined: products and services that do not yet exist but become commercially feasible as intelligence becomes more abundant. Bain cites areas including drug discovery, mental health and energy generation as examples.
Hardware becomes a strategic layer
The report also describes an important change underneath the software boom: hardware and semiconductors are capturing more value as AI compute requirements increase.
Bain says hardware and semiconductor stocks grew at a 24% compound annual rate from 2020 through 2026, compared with 6% for software. High-bandwidth memory, advanced packaging and custom silicon are among the fastest-growing areas identified by the consultancy.
Custom silicon is particularly significant.
Hyperscalers and AI-native companies increasingly have enough scale in training, inference and agentic workloads to justify designing accelerators around specific computing requirements. That can allow companies to optimize performance, energy consumption or cost for workloads that run at very large volumes.
The shift is also changing semiconductor procurement.
Bain says companies are increasingly diversifying suppliers and production locations as natural disasters, geopolitical disruption and export controls introduce additional supply-chain risks. Long-term capacity agreements, supplier investments and multi-geography sourcing are becoming more prominent in technology procurement.
For AI infrastructure providers, that makes the supply chain part of product strategy rather than simply an operational concern.
Cybersecurity enters the AI infrastructure equation
The report also highlights a security problem created by the speed of AI adoption.
Bain estimates that AI has compressed the time required for a typical cyberattack from roughly four weeks to about 18 hours. The proliferation of AI agents adds another layer of complexity because enterprises increasingly have software systems capable of taking actions rather than simply generating content.
That changes the requirements for AI governance and cybersecurity.
Traditional vendor assessments based on periodic questionnaires may not be sufficient when models, software dependencies and AI capabilities can change throughout a contract. Bain says companies need greater visibility into how vendors deploy AI, including changes made during the contract lifecycle and risks introduced through fourth parties.
The report says leading organizations are increasing remediation budgets and, in some cases, redirecting 20% to 25% of cybersecurity personnel toward addressing alerts generated by AI-powered security scanning.
For enterprise AI platforms, security therefore becomes part of the deployment architecture rather than a separate compliance exercise.
AI absorption becomes a competitive variable
Another theme in Bain’s research is what it calls AI absorption speed: how quickly an organization can move from having access to AI models to generating measurable business outcomes.
The report says leading AI labs are investing up to $9.75 billion in forward-deployed engineering capabilities to help customers integrate AI into their operations. Vendors are also developing application and infrastructure layers that connect models with enterprise systems.
That suggests the competitive landscape is moving beyond model performance alone.
Enterprises still need to determine which model is appropriate for a particular task, how it should connect to internal data, what controls should govern its actions and how its output should enter existing workflows.
Bain does not expect large language models to follow a simple commoditization path. Instead, its analysis describes a potential segmentation between frontier models for difficult or emerging workloads and lower-cost, specialized models for more mature use cases.
That model segmentation could create a more complex AI software stack, with enterprises dynamically selecting models according to cost, capability, latency and reliability.
Software engineering illustrates the next bottleneck
Bain’s Tech and Engineering Survey of 293 senior technology leaders points to another consequence of rapid AI adoption.
Respondents expected release-cycle speed to improve by 148% and software developer productivity by 95% over the following one to two years. Bain says those expectations are considerably higher than the 20% to 27% productivity improvements organizations are currently capturing across key metrics.
The gap highlights a recurring problem: accelerating one part of a workflow can simply move the bottleneck elsewhere.
AI can generate code faster, but organizations still have to review, test, coordinate, secure and govern that code. Bain argues that companies need to treat the software development lifecycle as a continuously improving system rather than optimize isolated coding tasks.
That principle extends to AI deployment more broadly.
The infrastructure buildout can provide more compute, but realizing its economic value depends on the applications, workflows and entirely new products built on top of it.
For TechEdgeAI’s AI infrastructure market, Bain’s report therefore highlights a central transition: the next phase of AI growth will be measured not only by how much compute the industry can deploy, but by how many new sources of economic value that compute makes possible.
Market Landscape
Bain’s report arrives as AI infrastructure investment expands across data centers, accelerators, networking, memory and power systems. The consultancy estimates that supporting the current buildout could require approximately $6 trillion in annual AI revenue by 2031, while existing consumer and enterprise applications may generate only $1.2 trillion to $1.8 trillion.
The resulting gap places greater emphasis on autonomous systems, robotics, physical AI, scientific discovery and new consumer AI business models. At the infrastructure layer, custom silicon and advanced semiconductor technologies are also gaining importance as hyperscalers seek workload-specific performance and supply-chain resilience.
The broader market is consequently developing across multiple layers: AI chips and memory, cloud infrastructure, data centers, AI development frameworks, enterprise applications and AI agents. The economic question increasingly concerns how these layers translate compute investment into recurring revenue.
Top Insights
- Bain estimates AI infrastructure investment would require approximately $6 trillion in annual revenue by 2031 to support the current buildout.
- Existing consumer and enterprise AI applications could generate $1.2 trillion to $1.8 trillion, leaving a substantial innovation gap.
- Custom silicon, high-bandwidth memory and advanced packaging are gaining importance as AI compute demand expands.
- Bain estimates AI has compressed a typical cyberattack timeline from roughly four weeks to about 18 hours.
- Enterprise AI adoption increasingly depends on integration, governance, security and engineering processes rather than model access alone.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
