> And I can't name several customers with very deep pockets working on their own chips to squeeze/compete with NVDA.

Google, Amazon, and Microsoft all have custom ASIC and TPU projects in their pipeline, and what holds true today might not hold true in 5 years.

A major reason Nvidia was able to do so well was because of technical outreach by donating their GPUs to programs all over the US, building a strong albeit self serving OSS relationship, the CUDA ecosystem, and the acquisition of Infiniband.

Much of these advantages can be nullified by competitive margins and pricing for hyperscalers designed and owned hardware.

Cisco was hit by this same situation as server vendors like Dell began integrating their own in-house networking functionality within their servers, and Dell itself was outcompeted by cloud vendors.