The AI spending race: why big tech is pouring billions into data centres
Four companies are about to spend more money building artificial intelligence infrastructure this year than most countries spend running themselves. Microsoft, Amazon, Google and Meta have each laid out capital expenditure plans for 2026 that add up to something in the region of 700 billion US dollars combined, and the bulk of it is going into one thing: data centres built to train and run AI models.
What is driving the spending increase
Capital expenditure, or capex, is money a company spends building or buying long term physical assets rather than covering day to day running costs. For a retailer that might mean new stores. For these four companies in 2026, capex almost entirely means:
Graphics processing units, or GPUs, the specialised chips that power AI
The buildings that house them
Power supply to run them at scale
Networking to connect it all
The scale of the shift is stark:
Combined capital expenditure across the four companies is set to rise by around 77% in 2026 compared with 2025, taking the group total from roughly 410 billion to around 725 billion US dollars, according to an analysis of the four companies' 2026 guidance
Coverage of the same earnings season put the increase at more than 60% on top of already record 2025 spending
That same coverage noted the scale of the outlay is starting to weigh on free cash flow across all four companies
Behind the numbers is a straightforward competitive logic: none of the four companies wants to be the one that runs out of AI computing capacity while its rivals keep building. Microsoft has pointed to an Azure order backlog worth tens of billions of dollars that it cannot yet fulfil because of power constraints, even as its own 2026 capital expenditure tracks toward roughly 190 billion US dollars.
South Africa's own data centre build out
The AI spending race is not only a story about Seattle, Mountain View and Menlo Park. South Africa has its own data centre sector expanding alongside it, led by operators like Teraco. In Johannesburg, Teraco completed an expansion of its JB4 campus that added 30 megawatts of capacity and made it the largest standalone data centre built in Africa, with several of the new halls built specifically for AI workloads. In Cape Town, the company completed a separate expansion of its CT2 facility that added 32 megawatts across eight new data halls designed for high density computing and AI training. It is a smaller scale version of the same underlying trend playing out at Microsoft, Amazon, Google and Meta, more compute capacity being built closer to where it is used, and a reminder that the AI infrastructure buildout is a global phenomenon rather than a purely American one.
What this means for investors
For South African investors watching this play out, the obvious question is how to get exposure without needing to pick a single winner among five very different companies with five very different strategies. Microsoft leans on its cloud and productivity software relationships. Amazon leans on AWS. Google is racing to fold AI into search while building its own custom chips. Meta is betting on AI as the next major computing platform after mobile. Nvidia sells the chips that make all of it possible.
Rather than betting on any one of these companies individually, some investors choose broad based exposure through exchange traded funds, or ETFs, that hold a spread of large technology companies rather than a single stock.
What else is worth weighing up
Concentration cuts in more than one direction here. The four hyperscalers are pouring an enormous share of global capital spending into a single technology cycle. Nvidia is concentrating a large share of its revenue in a small number of hyperscaler customers. And an investor who puts money into any one of these five companies is making a single bet on how that cycle plays out.
Some of these companies' spending increases have already unsettled markets during 2026 earnings season, a reminder that markets do not always reward aggressive capex even when the growth story is intact. One way investors manage this is through diversification, spreading exposure across a broader index rather than a handful of companies. Our comparison of the S&P 500 against the Nasdaq 100 looks at how the two indices differ in their exposure to companies like these.
Luno gives investors access to tokenised versions of shares in these big tech companies, as well as tokenised ETFs such as SPYx and QQQx for those who would rather hold a broader basket than a single name.
Investing in cryptocurrency and other digital assets may result in the loss of capital as the value can fluctuate.
Luno (Pty) Ltd is a licensed financial services provider (FSP No. 53314), and registered credit provider (NCRCP22123).
Frequently asked questions
What is AI capex?
Capex, short for capital expenditure, is money a company spends building or buying long term physical assets. In the context of AI, capex mostly means graphics processing units, data centre buildings, power infrastructure and networking equipment.How much are Microsoft, Amazon, Google and Meta spending on AI in 2026?
Combined capital expenditure across the four companies is projected to reach roughly 700 billion US dollars in 2026, up about 77% from 2025. Amazon has guided toward the largest single figure at around 200 billion US dollars, directed mostly at AWS data centres.Why is Nvidia so central to this story if it is not one of the four hyperscalers?
Nvidia's chips are what most of this capital expenditure is buying. Its data centre business generated the large majority of its total revenue in its most recent quarter, with a handful of hyperscaler customers accounting for more than half of that figure.Is this AI infrastructure buildout happening in Africa?
Yes. South Africa and Nigeria both have active data centre expansion tied to AI and cloud demand, led by operators like Teraco in Johannesburg and Cape Town, and Airtel and Kasi Cloud in Lagos.How can investors get exposure to AI infrastructure spending without picking a single stock?
Some investors use exchange traded funds that hold a spread of large technology companies rather than a single name, which spreads exposure across the sector instead of concentrating it in one company's outcome.
