Insights
By:
The Race for Scale
In the space of three short years, led by the world’s largest tech conglomerates, the buildout of AI infrastructure has become the dominant investment theme in global markets today.
Once upon a time, the Infrastructure Buildout theme was about the massive resources needed to fuel China’s urbanisation and the growth of its new mega-cities. As this theme matured, it evolved to encompass China’s Belt and Road Initiative, or the ‘New Silk Road’ as it was also known; China’s ambitious plan to develop land corridors and infrastructure in more than 150 countries. In the intervening years, the arrival of Barack Obama in the White House had expanded this theme to include the vast US investment required to update America’s crumbling infrastructure.
Today, the Infrastructure Buildout theme is about only one thing: the massive investment into building-out the infrastructure needed to deliver generative AI.
It’s been described as one of the greatest transfers of capital of all time; it’s a theme that captures everything from datacentres and semiconductors, and the technology giants racing to build both, to the energy companies being contracted to provide the immense level of power such computing requires.
A land of giants
In the west, the race to deliver AI is being dominated by the ‘hyperscalers’ – namely Meta (Facebook), Alphabet (Google), Amazon, Microsoft, OpenAI and Elon Musk’s xAI, which recently absorbed his X social-media platform (formerly Twitter).
The names they’ve chosen for their latest multi-billion datacentre investments echo the grandeur of their ambitions. Most recently, Mark Zuckerberg announced Meta’s new ‘titan clusters’, the five-gigawatt (GW) ‘Prometheus’ and ‘Hyperion’ projects; xAI has embarked on its ‘Colossus’ project while OpenAI, along with multiple partners, is developing its ‘Stargate’.
Other major US hyperscalers, that have seen less of the media spotlight in 2025, but which still represent global firepower in the war for AI include:
- Apple: its data centres already support a billion cloud customers via iCloud, Apple TV+, iMessage and Siri
- Anthropic: the developer behind the ‘Claude’ family of large language models
- IBM: which offers an evolving partner and product ecosystem with multi-cloud capabilities
- Oracle: which boasts an extensive network of data centres supporting its cloud computing services.
China’s hyperscalers have similar ambitions, but far lower budgets. Even so, the likes of Alibaba, Huawei, Tencent, Baidu and Kingsoft Cloud represent a challenge to the US-based AI players, especially after the success of China’s low-cost DeepSeek offering, which was enough to temporarily sweep the legs from beneath the AI investment rally back in early 2025.
Meanwhile, DeepSeek hasn’t cut its spending – the strongest indication that more capex is required. A recent Bank of America report estimated that combined government and business spending on AI in China is set to soar close to $100bn in 2025, a jump of just under 50% on 2024.
In most cases, when estimating purchasing power parity (PPP), it’s safe to multiply any dollar investment into China by three. The metrics are different in the case of datacentre investments due to the huge numbers of western chips the Chinese are importing as part of the process. Even so, given China’s far lower labour and land costs, and central party support (China’s government is expected to provide 70% of AI investment this year), it’s safe to assume that the value of these investments will be some way north of $100bn if it were spent on US-based datacentres.
Although the industry giants continue to build facilities of ever-greater size, at the other end of the spectrum, what are known as ‘edge data centres’ are also gaining ground and attracting private investors. These are literally, small neighbourhood computing centres, closely located to users, devices, and data sources. This minimalises latency, improves performance and enables the response times needed for real-time applications like streaming, the Internet of Things (IOT) or autonomous vehicles.
The war in numbers
Alphabet, Amazon, Microsoft and Meta alone spent more than $350bn on data centres in 2025 and this is already forecast to rise past $400bn in 2026. Meanwhile, total capex from these four companies is expected to sail past $500bn by 2030, effectively quadrupling what this cohort was spending when OpenAI released ChatGPT in late 2022.
It’s estimated by leading private equity players that it costs around $10bn to develop a 1GW US data centre, but a further $30bn in equipment costs, which illustrates the depth of the opportunity across sectors.
As a result, Goldman Sachs recently estimated that the growth of AI will see the total addressable market (TAM) for the software industry expand by ‘at least’ 20% pa between 2025 and 2029.
Meanwhile, Morgan Stanley estimates that between 2025 and 2029, global spending on data centres will hit almost $3 trillion, with just under half of this spending supported by capex from the leviathans of the tech market.
The rest, estimated to be in the region of $1.5 trillion, is expected to be financed by private investors, lenders, and developers.
AdobeStock_1127220670-recoloured
Powering the revolution
US electricity demand has risen for the first time in two decades, thanks to the soaring power demands made by data centres. Their mission-critical need for flexible and reliable power supply has put utility stocks firmly in the limelight.
Recent years have seen a steady stream of deals between the US tech giants and power suppliers of every hue. Among other things, this has triggered a renaissance for US nuclear energy suppliers.
In March 2025, Meta signed a 20-year deal with Constellation Energy for its Illinois nuclear facility to supply 1.1 MW of emissions-free nuclear energy. It followed a similar deal between Constellation and Microsoft, penned in late 2024, for a 20-year power contract which would restart Three Mile Island's Unit 1, by mid-2028.
Meanwhile, Amazon has built on existing agreements with the US utility Talen Energy to expand their nuclear energy partnership with a guaranteed 1.9 MW of electricity until 2024 from its plant in Susquehanna, Pennsylvania to power Amazon Web Services (AWS) data centres.
The ready support offered by state and governmental bodies has helped to accelerate US energy production from coal, nuclear, geothermal, and natural gas, to meet AI’s burgeoning demands. Such efforts make it clear that with AI accounting for an ever-greater share of America’s energy consumption it’s destined to become the next US strategic national asset, in the same way as the US military or its nuclear weapons.
Key battlefronts
While the US majors have so far mostly self-funded their expansion into data centres, the vast level of capital required to realise generative AI means there’s a whole eco-system of lenders and private equity investors also feeding expansion in the data-centre segment, despite the potential risks of obsolescence. The race for scale by participants in every part of the AI ecosystem has led to what are called ‘build-to-suit’ developments.
Oracle has been a trailblazer here. It has leased a 2GW data centre in Abilene, Texas, to OpenAI, as part of a $30bn a year deal to provide 4.5GW of computing power for the latter’s US Stargate data centre project. Meanwhile, the $15bn Abilene site is being developed by external partners using a combination of investment and borrowing that won’t appear on either Oracle’s or OpenAI’s balance sheets.
Thanks to the clamour for data-centre assets, third-party developers with ‘shovel-ready’ sites with worthwhile power and water connections are flourishing, for the moment, as they offer the potential for accelerated delivery over the typical three years it takes to develop a site.
Elsewhere, companies with ready-built facilities, such as CoreWeave, are leasing these, and the Nvidia chips they hold, to external clients.
When the music stops
Although the likes of Alphabet, Amazon, Meta, OpenAI and Microsoft are leading the goldrush to build-out AI data centres capable of training large language AI models – such as Google’s Gemini, Amazon’s Olympus, Meta’s Llama or Microsoft’s MAI – industry observers are already fretting as to the obvious risk of overcapacity.
Especially once the balance of AI shifts from training ever-smarter models to inference – namely using those models to respond to user queries, review new, real-world data, recognise images, or translate languages. As inference requires far less computing power, such AI training facilities could quickly lose their sheen. They also risk being left behind by subsequent advances in semiconductors and the cooling and power technologies that support them.
Even so, data is fast becoming the world’s most valuable resource meaning that the facilities that store, process, and protect it are on course to become some of the most accretive assets on the planet.
For now, the Infrastructure Buildout theme has become a race dominated by the largest companies on the planet. While some of the Magnificent Seven stocks stand to make great gains from the race to the summit, and are well placed to shoulder any shortfalls in expected returns, the breakneck speed of development means capital is also being attracted to far newer market entrants with weaker balance sheets and competitive positions.
Playing the Infrastructure Buildout theme
We have invested significantly into this theme through those companies we see as being best-in-class providers with strong competitive moats.
Among the Magnificent Seven mega-cap US tech stocks we own Alphabet, Amazon, and Microsoft. Each enjoys dominant positions in the cloud market having made enormous investments into building out their data-centre presence, and developing leading-edge large language models.
We also own the world’s most advanced photo-lithography business, ASML of the Netherlands. It enjoys a monopoly on the technology required to fabricate the world’s best semiconductors, which sit at the heart of the AI revolution.
Elsewhere, we own Amphenol, a global leader in the design, manufacture, and sale of interconnect solutions, including connectors, fibre-optics, cables, and electronic components. Its hardware solutions are central to the global buildout in AI infrastructure and high-speed data connectivity.
Our Infrastructure Buildout theme holdings are rounded out by Eaton Corporation, a leading international power management company. It provides mission-critical solutions for managing electrical, hydraulic, and mechanical power more efficiently. It offers intelligent power-management solutions for high-performance computing environments like data centres, while its smart power solutions enable the increased take-up of AI-driven automation.
1 Bank of America/South China Morning Post June 2025.
2 Bank of America/South China Morning Post June 2025.
3 Financial Times 14 August 2025.
4 Financial Times 14 August 2025 (based on 10-K filings, S&P Global Market Intelligence).
5 Financial Times 14 August 2025.
6 Reuters 11 June 2025.
