The four largest data center operators on the planet have collectively committed to spending nearly $2.4 trillion in the coming years. Amazon, Microsoft, Alphabet, and Meta are locked in an infrastructure spending race, and the reverberations are being felt well beyond Silicon Valley.
S&P Global estimates that data center capacity will need to expand by 55 to 60 gigawatts to meet surging demand, with associated spending landing in the $1.8 trillion to $2.4 trillion range through 2030. The near-$2.4 trillion commitment from these four giants sits right at the top of that range.
The numbers are getting absurd in the best possible way
Combined capital expenditure guidance for the four hyperscalers sits at roughly $725 billion for 2026 alone. That represents a 77% year-over-year increase.
Between 2026 and 2027, the quartet is expected to deploy approximately $1.5 trillion. Goldman Sachs has projected that total spending could reach $5.3 trillion or more by the end of the decade.
Cumulative big-tech AI infrastructure spending has already surpassed $1 trillion since 2023, with each quarterly earnings update revealing yet another upward revision to capital expenditure plans.
Crypto mining sites are getting a second life
Former Bitcoin mining facilities are being repurposed for AI workloads, a trend that’s gained significant momentum as hyperscalers scramble for sites with existing power infrastructure. Crypto mining operations were purpose-built to consume enormous amounts of electricity in locations with reliable grid connections, making them attractive candidates for AI data centers, which have similarly voracious power appetites.
Cipher Mining’s $5.5 billion contract with AWS stands as perhaps the most prominent example of this trend. Significant deals involving hundreds of millions of dollars for power access and facility conversions have become routine in the space.
For crypto miners, the value of their physical infrastructure—the land, the power purchase agreements, the cooling systems—may now be worth more as AI real estate than as Bitcoin mining operations. Several publicly traded miners have already pivoted or announced dual-use strategies, effectively becoming landlords to the AI economy while maintaining some mining capacity.
What this means for energy, semiconductors, and your portfolio
55 to 60 gigawatts of new data center capacity requires power generation equivalent to roughly 55 to 60 nuclear reactors. That demand is already straining grid capacity in key markets like Northern Virginia, where data center density has created genuine concerns about electricity availability for residential and commercial users.
Every new data center needs chips, and the most advanced AI accelerators from Nvidia, AMD, and increasingly custom silicon from the hyperscalers themselves are in perpetual short supply. A spending commitment of this magnitude essentially guarantees that semiconductor demand will remain elevated for the foreseeable future.
As hyperscalers aggressively lock up power capacity through long-term contracts, the available supply for mining operations could tighten, potentially increasing marginal production costs for Bitcoin. Conversely, miners sitting on valuable power contracts find themselves holding assets that appreciate in value regardless of Bitcoin’s price, with the optionality to mine Bitcoin when it’s profitable or lease capacity to AI companies when it’s not.