Nvidia sits at the center of the AI boom, but the cloud giants that buy its chips are quietly trying to escape its orbit. Amazon, Google and Microsoft are racing to design their own AI silicon and pouring staggering sums into data centers — a coordinated push to reduce dependence on the world’s most valuable chipmaker.
The custom-silicon surge
The hyperscalers are surrounding Nvidia with their own chips: Amazon’s Trainium for AI training, Google’s TPUs, and Microsoft’s Maia, plus custom Arm-based CPUs for their data centers. Amazon’s custom-silicon business — Graviton, Trainium and Nitro — topped a $20 billion annual revenue run rate in Q1, and CEO Andy Jassy says it would be a $50 billion business if it stood alone. That is no side project; it is a strategic pillar.
The spending is staggering
The scale of investment defies comprehension. Amazon, Alphabet, Microsoft and Meta are on track to spend roughly $725 billion on capital expenditures in 2026 — up about 77% from last year — with the vast majority going to AI chips, servers and data-center infrastructure. Microsoft alone expects around $190 billion in capex, even as its Azure cloud runs capacity-constrained through year-end.
Why escape Nvidia?
The motivation is cost and control. Nvidia’s chips are expensive and scarce, and paying it enormous margins on every AI workload eats into the hyperscalers’ profits. Designing in-house silicon lets them tailor chips to their own workloads, cut costs at scale, and reduce reliance on a single supplier that also sells to their rivals. Owning the chip is owning the economics.
What it means for Nvidia
Nvidia remains dominant — its hardware-plus-CUDA-software moat is formidable, and the same companies building alternatives are still buying its GPUs at scale. But custom silicon chips away at the edges: every workload that runs on Trainium or a TPU instead of an Nvidia GPU is revenue Nvidia does not capture. Over time, that could compress its extraordinary margins and market share.
The inflation wrinkle
The spending has side effects. Surging demand for memory and components has pushed up prices industry-wide — Microsoft attributed $25 billion of its budget to rising chip and memory costs — and analysts note the AI buildout is now a measurable contributor to broader inflation. The boom is so large it is moving the macro economy.
The bottom line
With $725 billion in capex and a full-court press of custom chips, Big Tech is making its most serious bid yet to loosen Nvidia’s grip on AI. Nvidia is not dethroned — but the era when it was the only game in town is ending, and the economics of the entire AI industry hinge on how the chip war plays out.
Photo: MDGovpics / BY via flickr