AI infrastructure investment 2050 is set to reach 31.6 trillion dollars, according to PwC. That is the total money the world will spend on building data centres for AI through the year 2050. PwC shared the number on September 2, 2026, in its Global Data Centre Outlook. It is the first long range forecast of its kind.
The yearly numbers are huge too. Data centre spending will rise from about 800 billion dollars a year in 2026 to 1.8 trillion dollars a year in 2050. To put it in scale, the total is more than the full yearly output of the United States economy in 2025. AI is no longer a small side bet. It is one of the biggest building projects in human history.
PwC asked Oxford Economics to build the model behind the report. The study covers 46 countries and territories. Together they hold most of the world economy and digital investment. PwC says no one has made a spending forecast this far out before. The range is wide. The full picture could land anywhere between 22 trillion and 50 trillion dollars. The 31.6 trillion number sits in the middle.
The United States will take almost half of the money. PwC expects America to capture 48 percent of AI infrastructure investment 2050, about 15.1 trillion dollars. The country sits at the heart of the advanced chip world. Asia Pacific comes next with 8.2 trillion dollars, led by China and India. Europe and the Middle East will take most of the rest. Many governments there are pushing their own sovereign AI plans.
The strange part is where the money goes. Unlike old building booms, this one will not slow down after the first wave of construction. Chips and other computer gear need fresh upgrades every few years. So recurring chip upgrades, not new buildings, will drive most of the spending. Computer equipment will grow from 70 percent of investment today to 93 percent by 2050. Chip makers are already trimming designs, and Tesla cut its Tesla AI5 chip memory in half to save cost.
Power will decide where the money flows. Affordable, reliable and clean electricity at large scale is the hardest thing for many countries to offer. Data centres drink huge amounts of power. So places with cheap and steady energy will win the investment. Links, security, clear policies and local support matter too.
Trade fights could cut the total. PwC tested two future paths. In the first path, tighter export controls break chip supply chains. Yearly investment falls to about half the main forecast by 2030, then slowly recovers. The AI infrastructure investment 2050 total drops to about 25.5 trillion dollars, around 6 trillion less. US export controls already force GPU export paperwork on American buyers. In the second path, countries push for digital sovereignty. They build their own trusted AI at home. The world total falls only a little, but the money moves to countries with strong local demand and weak data centre capacity.
The big question is where all this money comes from. AI labs report fast growing revenue, but even big numbers look small next to this bill. Amazon raised its 2026 spending plan to 220 billion dollars and ordered 2 million more Nvidia chips, according to Telecoms. Both Amazon and Google turned cash flow negative in the second quarter. AI now shapes even defense plans, and Musk joined a Pentagon AI warfare study this year. Clara Cutajar, global infrastructure leader at PwC Australia, called it one of the defining capital allocation challenges of the next generation.
The MES Times Take. This report is a wake up call for everyone. AI is not just software on a screen. It is steel, concrete, copper and power lines. Countries that sort out cheap clean energy today will own the AI economy of 2050. The rest will just pay rent. For young readers, the message is hopeful. The world will need millions of engineers, electricians and builders for this boom. The future is being built right now, one data centre at a time. And it belongs to those who prepare for it.
