Artificial intelligence is attracting unprecedented levels of investment, but the technology industry is approaching a critical question: Can AI generate enough economic value to justify the enormous infrastructure being built to support it?
A recent analysis from Bain & Company suggests the AI industry may need to generate roughly $6 trillion in annual revenue by 2031 to support the scale of investment currently flowing into data centres, computing infrastructure, chips and power systems.
The challenge is significant because AI infrastructure is expanding at a pace rarely seen in the technology industry. Data centres are becoming larger and more expensive, while demand for computing power continues to climb.
AI needs more than productivity gains
Much of the discussion around AI’s economic potential has focused on improving employee productivity. Businesses are already using AI for software development, customer support, marketing, sales, research and IT operations.
But productivity improvements alone may not be sufficient to support the infrastructure investment taking place.
According to Bain’s projections, enterprise AI productivity applications could need to generate between $1 trillion and $1.4 trillion in annual revenue to make a meaningful contribution to the economics of the industry.
The much larger opportunity could come from products and markets that have not yet been fully developed.
Bain estimates that new AI-powered products and services could account for approximately $4.2 trillion in annual revenue by 2031. Potential areas include next-generation search and advertising, autonomous systems, physical AI and other applications that could create entirely new categories of products and services.
That distinction is important. The long-term economics of AI may depend less on making today’s businesses marginally more efficient and more on creating industries and services that would not otherwise exist.
New AI markets could become the biggest opportunity
AI is increasingly being incorporated into consumer products and professional services, but the technology’s future could extend far beyond today’s applications.
Potential markets include drug discovery, energy systems, scientific research and other highly complex industries where large amounts of information and computation can be valuable.
Consumer AI could also become a substantial revenue source. Bain estimates that subscriptions and advertising tied to consumer AI services could eventually generate $200 billion to $400 billion annually.
The underlying idea is that AI could become a general-purpose technology, similar in economic importance to previous waves of computing and internet infrastructure. If that happens, entirely new business models could emerge around increasingly capable AI systems.
The infrastructure bill is getting enormous
Building increasingly powerful AI systems requires enormous amounts of physical infrastructure.
Bain estimates that annual AI infrastructure spending could reach approximately $1.5 trillion by 2031. This includes new data centres as well as upgrades to GPUs, memory, networking equipment and existing computing facilities.
If infrastructure spending represents roughly 25% of industry revenue, Bain’s calculations imply an AI market approaching $6 trillion per year would be necessary to support that level of investment.
This creates a fundamental economic test for the industry: the technology needs to generate enough revenue to keep pace with the infrastructure required to operate it.
Data centres are scaling at extraordinary rates
The scale of AI data centres illustrates the challenge.
According to estimates cited from Epoch AI, Meta’s Prometheus data centre in Ohio was expected to grow from around 600 megawatts of capacity and an estimated $24 billion cost in 2025 to potentially 2 gigawatts and $80 billion by 2027.
The projected expansion does not stop there. Estimates put the facility at as much as 5 gigawatts and $175 billion by 2029, potentially reaching 9 gigawatts and $200 billion by 2030.
While individual projects vary considerably, the broader trend is clear: AI computing facilities are becoming dramatically larger and more capital-intensive.
Bain estimates that the size and cost of AI data centres are effectively doubling every 12 to 16 months.
That rapid expansion creates a secondary challenge: supplying these facilities with everything they need to operate.
Power could become one of AI’s biggest constraints
Computing capacity ultimately depends on electricity.
As data centres become larger, operators need access to substantial and reliable power supplies. Expanding the electricity grid can take years, creating a potential mismatch between the speed of AI development and the pace at which energy infrastructure can be built.
The challenges extend beyond electricity. AI companies also need increasingly sophisticated semiconductors, GPUs, networking equipment and specialized workers.
Competition for experienced technical talent is another consideration, particularly as companies attempt to build and operate increasingly complex AI infrastructure.
There are also environmental, regulatory and community concerns surrounding large data centres, including resource consumption, land use and noise.
Governments are treating AI infrastructure as strategic
The enormous infrastructure requirements have also changed the role of governments.
Countries and regions including the United States, European Union, United Arab Emirates, Saudi Arabia and South Korea are supporting the development of AI and data-centre capacity.
The reason extends beyond commercial opportunity. Advanced computing infrastructure is increasingly viewed as strategically important to technological development, economic competitiveness and national security.
As a result, governments are becoming more involved in attracting data-centre investment, expanding energy capacity and developing domestic technology ecosystems.
This could create opportunities for companies involved in semiconductors, electricity generation, networking, cooling systems, construction and other parts of the AI infrastructure supply chain.
The next phase of AI will be about monetization
The AI industry has already demonstrated that there is enormous demand for computing power. The harder question is whether that demand can translate into an equally enormous and sustainable revenue base.
To reach the scale projected by Bain, AI companies will likely need to move beyond today’s applications and develop entirely new products, services and markets.
That could include everything from autonomous machines and AI-powered scientific research to new consumer services and highly specialized enterprise applications.
The next stage of the AI boom, therefore, may not be defined simply by who can build the largest model or data centre. It could increasingly depend on who can turn rapidly expanding computing capabilities into products and services that businesses and consumers are willing to pay for at scale.
With infrastructure investment potentially reaching $1.5 trillion annually within the next five years, the economic stakes are becoming just as significant as the technological ones.
Bain's $6 trillion annual AI revenue need by 2031 is useful as a scale check against the capex going into data centres and power. I would extend it with the split they claim: only about $1-1.4 trillion from enterprise productivity apps, and roughly $4.2 trillion from products that mostly do not exist yet. That means the infrastructure bill is underwritten by markets that are still a slide deck. Activation of a GPU cluster is not the same thing as repeat paid demand at that size.