Big Tech earnings kick off next week with Google, but the market's real focus isn't profit, it's the $700 billion-plus these four companies plan to pour into AI data centers this year, and how much of that spending is just inflation in disguise. Google, Amazon, Microsoft and Meta have already laid out the budgets. The harder question is how much computing power those dollars actually buy.
Morgan Stanley estimates the cost to build one gigawatt of AI capacity has jumped roughly 20 percent across leading systems. A standard Nvidia-based configuration now runs about $35 billion per gigawatt, up from $29 billion; a newer version has climbed to $49 billion from $41 billion. Memory chips, power gear, construction materials, skilled labor and grid connections are all tightening at once.
That creates a self-reinforcing loop. Big Tech orders more equipment, shortages worsen, prices rise, and the companies then raise spending forecasts partly to cover those higher prices, which generates still more demand and even higher prices. Circular Technology's Brad Gastwirth estimates 20 to 30 percent of the next capex increase will reflect inflation, not expansion. Earlier research puts the memory-price share of this year's cloud capex growth at roughly 45 percent.
Cantor Fitzgerald sees little change to 2026 plans this earnings season but expects 2027 estimates to leap higher, $283 billion for Google, $271 billion for Amazon, $200 billion for Meta. No one is blinking yet; no executive wants to signal caution while rivals race ahead. But the headline capex number is becoming a less reliable proxy for actual buildout speed.
Gastwirth says the tell will be in the granularity. If capex rises alongside disclosures on power capacity secured, GPUs deployed, memory purchased, networking gear ordered and new campuses broken ground, the spending reflects genuine growth. Without those metrics, a bigger number may simply mean Big Tech is paying more to stand still.