
Fetching...

19 November 2026
For decades, computing progress followed a familiar path: faster processors, more storage, bigger data centers. Artificial intelligence has broken that pattern. Training and running modern AI models requires a fundamentally different kind of infrastructure, and the world is now rebuilding it at record speed.
The scale of the shift is visible in the numbers. Global spending on AI infrastructure reached $89.7 billion in the first quarter of 2026 alone, up 33% from a year earlier. The largest US technology companies are on track to spend more than $700 billion on capital expenditure this year. And demand for data center capacity is running well ahead of supply.
This is the AI compute supercycle: a multi-year build-out spanning chips, servers, cloud platforms, data centers and the power that runs them. In this blog, we look at how the industry is evolving, where the capital is flowing and what lies ahead.
Traditional computing was built around the central processing unit (CPU), which handles tasks largely in sequence. AI workloads work differently. Training a large language model or running complex simulations requires thousands of calculations to happen simultaneously. That has shifted the industry toward accelerated computing: graphics processing units (GPUs) and custom chips (ASICs) designed for parallel processing.
The shift is already visible at the very top of the market. On the November 2025 TOP500 list of the world's fastest supercomputers, 255 of the 500 systems used accelerators, up from 237 just six months earlier. Supercomputing is also no longer the preserve of governments and research labs: the world's fifth-fastest system, Eagle, is operated by Microsoft Azure, a commercial cloud.
Even the architecture underneath is changing. In the first quarter of 2026, spending on ARM-based accelerated servers reached $53.0 billion, overtaking x86 systems at $34.6 billion for the first time. After decades of x86 dominance in the data center, this marks a notable turning point.
The clearest signal of the AI compute supercycle is where the money is going. In Q1 2026, global AI infrastructure spending reached $89.7 billion, up 33.1% year on year. Servers accounted for 97.6% of that total, or $87.6 billion. The capital is flowing overwhelmingly into physical hardware.
The world's largest technology companies are leading this build-out. Amazon, Microsoft, Alphabet and Meta are collectively on track to spend roughly $700–725 billion on capital expenditure in 2026, up about 75% from a record $410 billion in 2025.
| Company | Projected 2026 capex |
|---|---|
| Amazon | ~$200 billion |
| Microsoft | ~$190 billion |
| Alphabet | $180–190 billion |
| Meta | $125–145 billion |
Looking further out, annual AI-related capital expenditure is projected to exceed $1 trillion, with most of it directed toward data centers, power and compute capacity. For the companies that build and operate this infrastructure, that spending translates directly into demand.
The cloud market of the last decade was built for websites, enterprise software and storage. AI has created demand for something more specialized, and a new class of provider has emerged to meet it.
Neoclouds are cloud operators built specifically for AI. They offer on-demand GPU capacity, bare-metal hosting and infrastructure designed for training and deploying AI models. Gartner predicts that Neoclouds will capture 20% of a $267 billion AI cloud market by 2030, positioning them as a significant competitive force alongside the established hyperscalers.
This reflects a broader change in how compute is bought. Instead of purchasing and owning servers, companies increasingly rent GPU capacity as a service. The Neocloud market is projected to reach $53.4 billion by 2030, growing at roughly 58% a year, while the GPU-as-a-Service market is expected to grow from $6.1 billion to $37.1 billion by 2035.
Building AI infrastructure is not just a matter of buying chips. AI hardware places very different demands on the data center itself.
Legacy data centers typically operate at 5 to 15 kW per rack. Modern AI and high-performance computing clusters require 40 kW to upwards of 130 kW per rack. That step change is turning data centers from server storage facilities into industrial-scale compute facilities, with advanced liquid cooling and specialized power delivery.
Power has become the core bottleneck. New greenfield grid connections can take three to five years to secure, which makes energized capacity one of the most valuable assets in the industry. Companies that already hold large, grid-connected sites have a clear speed-to-market advantage.
Supply is struggling to keep pace with demand. North American data centers hold roughly six months of available capacity in reserve, while new capacity typically takes two to five years to bring online. That gap is one of the strongest indicators that the build-out has years left to run.
Industry forecasts point to sustained, multi-year growth across the AI compute value chain. The fastest growth is expected in the newest segments, Neoclouds and data center GPUs, while the broader high-performance computing market continues to expand steadily.
| Segment | Projected market size | Expected growth (CAGR) |
|---|---|---|
| Neocloud market | $53.4 billion by 2030 | ~58% |
| Data center GPUs | $22.7 billion (2026) to $159.3 billion (2033) | 32.1% |
| GPU-as-a-Service | $6.1 billion to $37.1 billion by 2035 | 22.3% |
| Global HPC market | $91.4 billion by 2030 | 10.4% |
Taken together, these projections describe a structural shift rather than a short-term cycle. As AI moves from experimentation into everyday business operations, the infrastructure behind it is becoming a core part of the global economy.
For investors, the AI compute supercycle presents both an opportunity and a challenge. The opportunity is clear: record capital spending, rapidly growing markets and a long runway of demand. The challenge is precision. AI compute spans hardware makers, specialized cloud providers and infrastructure operators, many of which are still in the middle of transitioning their businesses.
The Indxx High Performance Computing Index was designed to address that challenge. It offers rules-based exposure to the companies enabling accelerated computing, across two core sub-themes: HPC infrastructure and hardware, and Neoclouds and specialized AI compute services. Companies are classified by how much of their revenue comes from these themes, as pure-play, quasi-play or marginal-play, so the index can capture both established leaders and companies pivoting into AI compute.
The index currently comprises 27 constituents and serves as the underlying index for the Grayscale AI Compute ETF (GCPU).
The infrastructure behind AI is still being built. The companies building it are where the next chapter of this supercycle will be written.