AI Infrastructure to Require $7tn by 2030, says McKinsey
Almost US$7tn will need to be invested in global data centre infrastructure to meet the rising demand for AI, according to a McKinsey report.
The report found that companies across the tech infrastructure ecosystem will need to invest US$6.7tn worldwide by 2030 to meet growing demand for AI and traditional IT applications. It also suggests US$5.2tn could be for AI-related data centre capacity alone, coming months after OpenAI and SoftBank announced their US$500 billion Stargate project to develop shared computing facilities.
Hyperscalers face tough investment decisions for future demand
The compute power value chain faces unprecedented investment decisions amid uncertainty about how AI development will impact future demand. Boston Consulting Group reports data centre investments have surged to US$50 billion in 2024, up from US$11 billion in 2020, driven by AI requirements.
- $3.1 trillion – The investment required by semiconductor firms and IT suppliers, marking the largest share of projected capital needs
- 20-30 megawatts – Power requirements for a typical AI-optimised data centre, compared to 5-10 megawatts for traditional facilities
- 70% – Proportion of future data centre demand projected to come from AI workloads by 2030
“Will hyperscalers continue shouldering the cost burden, or will enterprises, governments, and financial institutions step in with new financing models?” McKinsey asks. Companies face a dilemma: overinvesting risks stranding assets, while underinvesting means falling behind competitors.
McKinsey projects global data centre capacity could triple by 2030, with 70% of demand coming from AI workloads. This projection depends on viable AI use cases delivering business value and the pace of technology innovation. The report notes that growth will be uneven across regions, with highest concentration expected in areas with established technology ecosystems and reliable power infrastructure.
Chinese LLM company DeepSeek reported in February 2025 that its V3 model reduced “training costs by approximately 18 times and inferencing costs by about 36 times, compared with GPT-4o.” However, McKinsey suggests these efficiency gains might be offset by increased experimentation and training across the broader AI market, maintaining overall compute demand at projected levels.
McKinsey identifies key investor groups in AI infrastructure
The research categorises five investor archetypes with distinct challenges and opportunities. Builders – real estate developers and construction companies – are projected to invest US$800bn in AI workload capital expenditure.
Labour shortages present a significant challenge for builders, with the report noting that construction timelines are extending due to skills gaps across multiple markets. This challenge is driving innovation in construction methodologies, with prefabrication and automated building systems gaining traction.
Energisers – utilities and energy providers – face US$1.3tn in AI-related capital requirements. These companies are “making substantial investments in emerging power-generation technologies” while addressing grid weaknesses. McKinsey expects “renewables projected to account for approximately 45% to 50% of the energy mix by 2030, up from about a third today.”
- Builders: Real estate developers and construction companies investing $800 billion in physical infrastructure
- Energisers: Utilities and power providers committing $1.3 trillion to energy generation and distribution
- Technology developers: Semiconductor firms and IT suppliers requiring $3.1 trillion for computing hardware
- Operators: Hyperscalers and colocation providers optimising infrastructure utilisation
- AI architects: Model developers and enterprises building proprietary AI capabilities
The power requirements for AI infrastructure are substantial. A typical AI-optimised data centre requires 20-30 megawatts of power, compared to 5-10 megawatts for traditional facilities. This increased demand is straining existing grid capacity in many regions, creating bottlenecks for expansion.
Technology developers and designers – semiconductor firms and IT suppliers – face the largest investment need at US$3.1tn. “A small number of semiconductor firms have a disproportionate influence on industry supply, making them potential chokepoints in compute power growth,” the report states. This concentration creates significant market risks, with delays in chip production causing ripple effects throughout the AI ecosystem.
The remaining groups – operators (hyperscalers and colocation providers) and AI architects (model developers) – were not quantified in the research but play crucial roles in optimising infrastructure utilisation and computational efficiency.
Efficiency improvements shape computing requirements
Cost structures for AI are evolving, particularly for inference operations. “Models with advanced reasoning capabilities require significantly higher inference costs. For example, it costs six times more for inference on OpenAI’s o1 compared with the company’s non-reasoning GPT-4o,” the research reveals.
To address these costs, “leading AI companies are optimising their model architectures by using techniques like sparse activations and distillation” to reduce computational requirements. These optimisations could significantly impact the long-term infrastructure requirements for AI deployment.
McKinsey recommends companies focus on understanding demand projections amid uncertainty, innovating on compute efficiency and building supply-side resilience.
“Investing strategically is not just a race to scale data infrastructure – it’s a race to shape the future of AI itself,” the report concludes.
Explore the latest edition of Data Centre Magazine and be part of the conversation at our global conference series, Tech & AI LIVE.
Discover all our upcoming events and secure your tickets today.
Subscribe to our Weekly Newsletter for the latest insights, trends and expert analysis.
Data Centre Magazine is a BizClik brand