Amazon, Alphabet and Microsoft lead a roughly $1 trillion capex wave in 2026

Nine companies are expected to spend about $1 trillion on capital expenditures this year. Most of that money is headed into AI data centers, plus the chips and networking gear that fill them.

Three names carry most of the weight. Amazon, Alphabet and Microsoft together account for roughly $590 billion of the total. That leaves the other six companies splitting what remains, which is still a very large check.

Who is spending what

Capital expenditures, or capex, are what companies spend on long-lasting physical assets. Think buildings, servers and equipment, rather than salaries or electricity bills. When a tech giant builds a data center, that cost lands in capex.

Amazon sits at the top of the list. The company is forecasting approximately $220 billion in capital expenditures for 2026. Part of that figure is attributed to rising memory-chip costs, so not every extra dollar buys extra capacity. Some of it just covers pricier parts.

Alphabet, Google’s parent company, has kept its capex guidance at $195 billion to $205 billion. It also reported a Google Cloud backlog of $561 billion. A backlog is business customers have committed to but the company hasn’t delivered yet. It works like a restaurant’s reservation book: the tables aren’t full tonight, but the bookings are already on paper.

That backlog figure matters. It gives Alphabet a concrete argument that the spending is chasing real demand, not just hope.

Microsoft went the other direction, at least slightly. The company lowered its 2026 spending forecast to about $170 billion, down from an earlier estimate of $175 billion.

Meta and Oracle round out the group of major US hyperscalers driving the surge. Hyperscalers are the companies that run cloud infrastructure at massive scale and rent computing power to everyone else.

How we got here

The current totals are larger than earlier projections suggested. The combined spending of Amazon, Alphabet, Microsoft and Meta was previously estimated at about $720 billion to $745 billion for 2026.

The longer-term outlook is bigger still. Projections indicate the five major US hyperscalers could exceed $1 trillion in annual capex from 2027 to 2030.

What this means

The obvious beneficiaries are the companies selling the picks and shovels. Chipmakers, memory suppliers, networking vendors and data center builders sit on the receiving end of this capex. Amazon’s note about memory-chip costs hints that suppliers already have pricing power.

For the spenders, the math is less comfortable. Concerns have emerged about whether this pace is sustainable, and three pressure points keep coming up.

The first is depreciation. When a company buys a server, it doesn’t expense the full cost at once. It spreads the cost over the asset’s useful life, and that charge hits earnings every year. Spend heavily now, and those charges pile up later, whether or not AI revenue keeps pace.

The second is energy. Data centers need enormous amounts of power, and building one is pointless if the grid can’t supply it.

The third is chip supply. Even with budgets approved, companies can only build as fast as suppliers can deliver. Supply chain constraints could push spending into later years or drive up prices, which appears to be part of Amazon’s story already.

Investors are effectively being asked to accept a trade. Spend now, earn later. Alphabet’s $561 billion cloud backlog is the strongest evidence on the table that later is real. Microsoft’s small forecast cut is a reminder that even the biggest buyers are watching the bill.