The numbers are no longer projections. They are commitments.
Microsoft: $80 billion dedicated to AI data centre expansion in fiscal year 2025 alone.
Amazon: $86 billion allocated for AI infrastructure, with Anthropic receiving $8 billion in investment tied to AWS infrastructure.
Meta: Building GPU clusters with 600,000+ NVIDIA H100 equivalents for LLaMA model training.
Oracle: $300 billion five-year compute deal with OpenAI, beginning in 2027.
NVIDIA: Record quarterly data centre revenue of $46.7 billion (Q2 FY2026), up 56% year-on-year.
NVIDIA’s 2025
pre-orders for H100 and H200 GPUs tripled available supply.
SK Hynix
and Samsung control 95% of high-bandwidth memory (HBM3E) production a critical bottleneck for GPU manufacturing.
TSMC’s
advanced packaging capacity (essential for AI accelerators) is expanding at ~1.6x per year, barely keeping pace with demand.
Data centre
construction timelines are measured in years, not months while demand grows quarterly.
Power availability
is now the primary constraint on where new AI infrastructure can be deployed.
This imbalance
is structural, not cyclical. AI compute demand doubles approximately every 3.5 months. Even with aggressive supply expansion, the gap between what the market needs and what the market has is widening.
Demand is compounding. Supply is constrained.
The core dynamic of the AI infrastructure market is structural imbalance. Demand for GPU compute is growing exponentially driven by model training, enterprise inference, autonomous systems, scientific computing, and sovereign AI programmes. Supply is constrained by chip fabrication bottlenecks, advanced packaging limits, and the sheer physical challenge of building data centres fast enough.
GPUs are no longer components. They are infrastructure assets.
For decades, GPUs were components inside gaming PCs and graphics workstations. That era is over. Today, GPUs are the primary processing units in the most important and expensive infrastructure being built globally: AI data centres.
The GPU market alone is projected to grow from $40 billion in 2022 to over $400 billion by 2032, a compound annual growth rate of 25%. The broader AI infrastructure market reached $101 billion in 2026 and is projected to reach $202 billion by 2031.
This shift has changed how investors view GPU hardware. A survey by KPMG and Nuway Capital found that nearly 80% of investors see generative AI as the primary driver of GPU infrastructure investment. Over 70% of high-net-worth investors rank GPUs as a better investment opportunity than blockchain or quantum computing.
GPUs are now treated as long-term physical infrastructure
comparable to power generation assets, telecommunications equipment, or transport infrastructure. They have physical form, known useful lives, measurable throughput, and demonstrable revenue-generation capability.
Infrastructure is the enduring asset class of every technology revolution.
AI models will evolve. Some will succeed, many will fail. AI platforms will consolidate. Software applications will rise and fall. But the compute infrastructure underpinning all of it will persist and compound.
Over 70% of global corporate AI investment is spent on infrastructure. This is not incidental it reflects the reality that AI is a hardware-intensive discipline. Every model trained, every inference served, every enterprise AI workflow deployed requires physical GPU compute running in physical data centres.
For institutional investors seeking exposure to AI without the binary risk of picking individual models or applications, infrastructure represents the most defensible, asset-backed, and structurally supported position in the market
“Everything starts with compute. Compute infrastructure will be the basis for the economy of the future.”
— Sam Altman, CEO of OpenAI, September 2025
© 2026 Infrahub Compute. All Rights Reserved