Why Hyperscalers Want Custom Silicon
At enormous cloud scale, small efficiency gains are valuable.
A cloud provider may optimize silicon specifically for:
- networking;
- storage;
- security;
- virtualization;
- AI data movement;
- infrastructure control.
Custom silicon can reduce unnecessary general-purpose functionality.
What an IPU Does
Infrastructure processing units can offload infrastructure tasks from the main server CPU.
Examples include:
networking
storage
security
virtualization
telemetry
That allows the host CPU to spend more cycles on customer workloads.
Google Collaboration
Intel and Google announced expanded co-development of custom ASIC-based IPUs in 2026. [S22]
This is strategically different from simply selling Google a catalog chip.
It is a deeper design relationship.
Why Custom Silicon Fits Intel
Intel can potentially combine:
CPU architecture expertise
+
ASIC design
+
foundry manufacturing
+
advanced packaging
inside one customer engagement.
That is precisely the kind of system-level relationship Intel wants Foundry to enable.
NVIDIA Collaboration
The 2025 NVIDIA partnership also includes custom x86 CPU design for NVIDIA AI infrastructure. [S31]
That reinforces the idea that Intel's x86 capability can be sold as a customizable building block rather than only a standard Xeon product.
Economics
Custom silicon can produce:
- large strategic customer relationships;
- longer design cycles;
- lower unit fungibility than merchant products;
- potentially stronger customer lock-in.
But customer concentration can also rise.
Intel disclosed that its top three customers represented 43% of 2025 net revenue. [S1]
Strategic Role
Custom silicon gives Intel another path to remain relevant as hyperscalers verticalize.
Instead of resisting customer-designed chips, Intel can participate in designing and manufacturing them.