Artificial intelligence capabilities are advancing faster than the infrastructure, governance systems and organisational changes needed to support them, creating bottlenecks and risks that could slow the technology’s economic impact, according to new research by the McKinsey Global Institute (MGI).
In a discussion paper published on September 10, 2026, titled “The AI Economy: Interconnected Forces, Feedback Loops, and Speeds of Change”, MGI argues that understanding AI’s future requires looking beyond technological advances to the interaction between infrastructure, investment, institutions, businesses and people.
The report says AI capabilities have been improving exponentially by some measures, while building data centres, power grids, chip factories and other physical infrastructure takes years. Organisations may take even longer to redesign their operations and working practices around the technology.
“Risks and bottlenecks arise because different parts of the AI system are moving at vastly different paces,” the report says.
It adds that leaders must learn to manage these different speeds because progress can slow when infrastructure or organisations fail to keep pace, while risks increase when AI adoption advances faster than safeguards, governance and public acceptance.
Infrastructure and skills emerge as constraints
The report traces some of the early constraints on AI development to shortages of advanced chips and computing capacity. As those bottlenecks eased, electricity supply, grid connections and data-centre construction became more prominent challenges.
It cites the example of AI chip demand outstripping the capacity to package processors with the memory required to operate them in 2023. The report says buyers reportedly faced lead times of up to 11 months for finished servers based on Nvidia’s H100 chips.
Electricity supply has since emerged as another constraint, with data centres requiring substantial power and utilities needing to build new lines and substations before some facilities can operate.
The report notes that the median new power plant in the United States takes about five years from an interconnection request to commercial operation, compared with less than two years in 2008.
It also identifies data-centre construction, access to copper and other critical minerals, and shortages of engineers and other skilled workers as potential constraints on further expansion.
Meanwhile, AI capabilities continue to improve rapidly. According to the report, the complexity of tasks AI can reliably perform has doubled approximately every seven months since 2019 and every four months since 2023, based on measures comparing AI task performance with the time human experts take to complete similar work.
However, the report cautions that increasingly capable technology does not automatically translate into higher productivity across businesses and economies.
Organisations struggle to redesign work
MGI says many organisations have adopted AI without fundamentally changing how their operations work, limiting the potential benefits.
A McKinsey survey of 1,719 respondents across 97 countries, conducted in May and June 2026, found that nearly nine in 10 respondents said their organisations regularly used AI in at least one business function.
However, outside a small group classified as AI high performers, representing 6 per cent of the sample, only about one-quarter reported redesigning workflows around AI rather than inserting the technology into existing processes.
The report also identifies gaps in AI governance and strategy. A separate survey of about 500 organisations, conducted in December 2025 and January 2026, found that only about 30 per cent had reached level three or higher on a four-level AI governance maturity scale.
It further says that only 22 per cent of operations leaders reported having a fully developed and implemented enterprise AI strategy.
The researchers warn that unclear accountability across information technology, legal, human resources, data, risk and strategy teams can leave organisations without adequate oversight of AI systems, including decisions about which actions autonomous systems can take and when human approval is required.
Cybersecurity is another concern. The report says roughly one in three organisations still lack a formal process for checking AI systems for security risks before deployment.
AI’s economic impact remains uncertain
The report presents AI as an interconnected system with three broad layers: foundations, transmission and outcomes.
Foundations include AI capabilities, research and development, physical resources and institutional rules. Transmission refers to adoption and adaptation by individuals, businesses and governments. Outcomes include productivity growth, innovation, employment changes, shifts in investment and effects on society.
According to MGI, changes in one part of the system can trigger feedback across the others. Better models may encourage wider adoption, which attracts investment and expands infrastructure and research. Conversely, power shortages, weak governance or declining public confidence may slow deployment.
The report says the economic gains from AI will depend on more than the ability to automate individual tasks. Businesses must redesign workflows, develop new products and spread productivity improvements across operations. At the economy-wide level, those changes must extend across industries.
It also warns that rapid growth in AI capabilities does not guarantee equally rapid growth in revenue or profits for companies supplying or using the technology.
MGI estimates that global investment in data centres could reach US$7 trillion cumulatively between 2025 and 2030, much of it driven by AI. The report cautions that investment could outpace demand, leaving excess capacity and financial losses, as happened during earlier railway and internet investment booms.
Employment is another uncertain outcome. The report says technological change can eliminate some roles while creating new occupations and demand for different skills. Whether AI ultimately creates more jobs than it displaces will depend partly on the emergence of new kinds of work and whether demand grows quickly enough to offset losses from automation.
Call for coordinated responses
The researchers argue that businesses, investors, policymakers and individuals need to monitor the entire AI system rather than focus solely on the technology.
For businesses, the report recommends developing new products and markets, adapting workflows more quickly and ensuring employees, customers and communities see tangible benefits from AI adoption.
For investors, it advises monitoring emerging constraints, distinguishing between rapid technological progress and slower business returns, and assessing risks shared across AI infrastructure, software and related industries.
For policymakers, the report recommends addressing infrastructure and skills constraints, supporting competition, adapting regulation as evidence changes, building public trust and encouraging investment in computing infrastructure, energy, research and talent.
Individuals, meanwhile, are encouraged to develop AI fluency while retaining human judgment and being careful about sharing sensitive personal information with AI tools.
The report concludes that AI’s trajectory will not be determined by technological advances alone, but by how effectively infrastructure, institutions, organisations and people adapt to the changes.
“The task is not to predict a single AI future, but to understand how the pieces interact, where mismatches may emerge, and how the choices people make can alter what happens next,” the report says.