AI Data Centres: Who are the West’s Super-Investors?
With approaching $8 trillion projected to be invested in AI data centres and infrastructure over the next five years by the groups below alone, we look at the West’s top five financial giants behind this capital race.
Goldman Sachs projects a combined $5.3 trillion in capital expenditure on AI Data Centres between 2025 and 2030 from numbers 2 to 5 in our list. At number 1, Blackstone’s $100 billion commitment makes AI one of the largest voluntary capital allocations in economic history, equivalent to the GDP of Germany and France combined.
The four largest Hyperscalers are Microsoft, Amazon, Meta and Alphabet (Google’s holding company). In 2026, their spending is expected to reach $725 billion, up 77% on the £410 billion spent in 2025.
The fifth, in pole position, is the world’s single largest investor in Data Centres: Blackstone.
Between them, these companies’ strategies, financial clout and competitive ambitions are literally reshaping the physical landscape of the Western world.
Key Terms:
- GPU: graphics processing unit (a specialised electronic circuit for graphics used in gaming, video editing and AI)
- Training AI: needs larger centres and more power to instil data and run tests.
- Inferential AI: the service available once AI is ‘trained’. Co-location and speed of processing more of a priority.
1 Blackstone
- In April 2026, Stephen Schwarzman, Blackstone co-founder, Chairman and CEO, declared ‘”Blackstone has become the largest investor in AI-related infrastructure in the world”’
- By the end of this year, £100 billion will be committed to data centre investment.
- Operating companies under the Blackstone empire include QTS Realty Trust (acquired 2021); AirTrunk (acquired 2024); BREIT; Blackstone Digital Infrastructure Trust (BXDC).
QTS now operates the Cambois campus in Northumberland (£10 billion, 153 hectares), is developing major sites in Pennsylvania, Atlanta, Dallas, and across the Netherlands and Germany, and has a global pipeline of development commitments that dwarfs all rivals in the private market.
In addition, Blackstone acquired AirTrunk — Australia and Asia-Pacific’s largest data centre operator in 2024 for approximately $16 billion. This instantly created one of the most geographically diversified data centre portfolios in the world.
The firm’s competitive edge lies not just in its capital scale but in its operational model. Blackstone early saw the need for a power infrastructure to support power-hungry AI centres and deliberately ‘built big’, literally and through acquisition. This prescience has placed Blackstone at the core of AI data centre development. With the backlash against energy use, Blackstone has invested heavily in greener sources, reducing the carbon footprint of its newer facilities.
Blackstone does not simply finance development, it deploys proprietary deal-sourcing tools, manages power procurement and grid negotiation. It offers confidence through supreme financial strength to deliver multi-gigawatt campuses on multi-decade lease terms.
2 Microsoft
- Microsoft’s capital expenditure for 2026 is approximately $190 billion, the vast proportion of this for AI data centres and GPUs.
- Its infrastructure includes Azure cloud (which has grown 40%); OpenAI partnership; global data centre estate.
- In a single quarter of the 2026 tax year, Microsoft spent $30.9 billion on capital expenditure — more than most countries spend annually on research and development.
Uniquely, Microsoft is simultaneously a tech company, a cloud operator, and, through its partnership with OpenAI, the most direct commercial beneficiary of the revolution it is funding. The vertical integration model sees Microsoft designing its own AI chips (the Maia series), operating its own network infrastructure and co-developing the AI models that runs its hardware.
It has an estimated backlog of $80 billion in orders it cannot yet fulfil owing to power constraints.
In the UK, Microsoft has committed £10.5 billion to data construction over 3-5 years.
Globally, Microsoft is building or expanding campuses across the US, France, Germany, Sweden, Japan, Australia and the UAE.
Not only is Microsoft’s data centre programme a primary force behind the growth in demand for electricity in several US states, but it is also a primary customer for nuclear energy restart programmes, including the reopening of Three-Mile Island Unit 1 in Pennsylvania.
3 Amazon Web Services
- Capital expenditure in 2026 is approximately $200 billion, concentrated mainly on construction of data centres, networking infrastructure and GPU Procurement.
- Key infrastructure: AWS cloud; 33 operational regions globally, US Project Rainier.
- Project Rainier in the US is a hyperscale cluster developed in partnership with Microsoft. It uses custom Trainium chips and is one of the largest AI training facilities ever built, spanning multiple sites in northern Virginia.
With approximately 30% of the global cloud market share, Amazon Web Services (AWS) is the world’s largest cloud computing provider.
Each of the 33 cloud regions comprises multiple data centres. Expansion has been announced in the UK, Thailand, Malaysia, New Zealand, Mexico, Saudi Arabia and Germany.
In the UK, Amazon has committed to spend £8 billion on cloud infrastructure in three years to 2027.
The scale of Amazon’s global logistics network gives it infrastructure procurement advantages no plain data centre company can match.
Amazon’s competitive position rests on the breadth of its cloud services, its custom silicon programme (Trainium for training, Inferentia for inference).
Like Microsoft, it is also a major buyer of renewable energy and nuclear power, with long-term power purchase agreements signed across the US, Europe, and Asia-Pacific.
4 Alphabet (Google)
- Capital expenditure for 2026 set to be $175–190 billion. Up from $85 in 2025.
- Key infrastructure: Google Cloud; Google DeepMind; custom TPU chips.
- Google DeepMind, whose headquarters are in London, is one of the world’s leading AI research organisations. Its findings drive demand for the infrastructure Google builds.
The doubling of investment shows the acceleration of AI demand. It also reflects Google’s resolution to maintain its early lead in AI to Microsoft or Amazon given its history of research into the potential of AI.
Alphabet was the first hyperscaler to make AI the explicit centre of its data centre strategy. Now it their sixth generation, Google’s Tensor Processing Units (TPUs, custom chips designed specifically for AI training and inference workloads). These give Google a technical advantage in cost-per-compute that allows it to operate AI services more efficiently than rivals relying on third-party GPU supply.
Part of a £1 billion UK investment commitment, in 2025, Google opened a new data centre in Hertfordshire.
Globally, Google’s 2026 construction programme includes major expansions in Ohio, Texas, Iowa, South Carolina, Virginia, Singapore, Taiwan, Finland, and Belgium.
Google has also been the most aggressive of the hyperscalers in securing long-term nuclear power agreements, signing contracts for advanced small modular reactor capacity in the United States.
5 Meta
- Meta’s 2026 capital expenditure is approximately $115–135 billion.
- Key infrastructure: Meta AI; Llama models; proprietary GPU clusters
- Meta is willing to build at the 5 GW scale where a single facility requires power equivalent to the needs of a small city’s entire power grid. This is a new threshold in data centre ambition that even its hyperscale rivals have not yet matched.
Meta’s planned capital expenditure is particularly remarkable given that, unlike the previous companies, it does not sell cloud services to third parties. Every dollar Meta invests in data centre infrastructure is used for its own AI research, its social media platforms (Facebook, Instagram, WhatsApp) and its increasingly ambitious AI assistant products.
CEO Mark Zuckerberg framed AI as competitive necessity for Meta. The Llama family of open-weight large language models has become the foundation for thousands of third-party AI applications globally, creating a demand that Meta itself must supply.
To this end, Meta is also building “the world’s longest subsea cable”. This is a 40,000-kilometre fibre loop encircling the globe to give its data centres private, low-latency international connectivity free from dependence on shared infrastructure.
Sites include a 1 GW data centre in Ohio and a facility in Louisiana that could eventually scale to 5 GW.
In Europe, Meta operates large data centres in Denmark, Ireland, and Sweden, all powered by renewable energy, and has committed to significant UK expansion.
Conclusion
It’s a race. Whoever builds the most and fastest wins the AI race. 
Whoever wins that cleans up financially from the biggest technological ‘advances’ of the century. Analysts predict the small will go to the wall, as happened before in the tech evolution wars.
Sadly, walls will also cover a lot of our landscape, and, as we see elsewhere in this newsletter, the natural world will not recover in a lot of human life times. It’d better be worth it.