Layer 1. Compute
Does the thermal architecture follow the silicon, or does it try to lead it? Coral sources accelerators and immersion-native nodes rather than designing compute, and designs its thermal architecture around what the silicon vendors actually ship. The largest strategic error available to a modular infrastructure company is a thermal architecture that requires the silicon vendor to cooperate.
The engineering question
Does the thermal architecture follow the silicon, or does it try to lead it?
What Coral decided
Follow it, without exception. Coral sources accelerators and immersion-native nodes rather than designing compute, and designs its thermal architecture around what the silicon vendors actually ship rather than around what would be thermally convenient.
The largest single strategic error available to a modular infrastructure company is to design a thermal architecture that requires the silicon vendor to cooperate. They will not. Coral's architecture selection process starts from the published mechanical and thermal interface of the hardware the customer wants to run.
What is still unsolved
Whether immersion-native node design, where the board is laid out for the fluid rather than adapted to it, remains economically viable as rack-scale integrated systems become the dominant form factor. Coral's current position is that it does for the PCIe and eight-GPU segment and does not for rack-scale.
Where this appears in the pod
Request an architecture review.
A structured technical assessment of power, thermal capacity, density, jurisdiction and deployment sequencing for a specific site.