Alpha Galactica logo Alpha Galactica Technology Investment Insights

In no theme: tracked for the companies it names

Custom silicon is reducing dependence on merchant GPUs

Large compute buyers may be moving more production workloads onto their own accelerators.

NVDA AMD AVGO GOOGL AMZN MSFT

Current assessment

No material update

Evidence supporting the claim is broadening from single-firm internal use to commercial distribution. Google supplying TPUs to Oracle Cloud, alongside Oracle adopting Cerebras and OpenAI custom inference chips, directly challenges the assumption that proprietary silicon remains confined to internal hyperscaler workloads. This expands options for buyers seeking alternatives to merchant GPUs. However, merchant GPUs retain deep software ecosystems and dominance in frontier training, leaving open whether multi-cloud custom silicon can win large external developer share.

Latest development: On October 27, 2026, Oracle announced plans to offer Google's TPU accelerators on its public cloud, marking the first time Google has supplied its chips to another hyperscaler, while also adding Cerebras accelerators and OpenAI's Jalapeño inference processor.

Last material update
28 Oct 2026
Last checked
4 Nov 2026

Development history

3 updates

  1. On October 27, 2026, Oracle announced plans to offer Google's TPU accelerators on its public cloud, marking the first time Google has supplied its chips to another hyperscaler, while also adding Cerebras accelerators and OpenAI's Jalapeño inference processor.

    Evidence supporting the claim is broadening from single-firm internal use to commercial distribution. Google supplying TPUs to Oracle Cloud, alongside Oracle adopting Cerebras and OpenAI custom inference chips, directly challenges the assumption that proprietary silicon remains confined to internal hyperscaler workloads. This expands options for buyers seeking alternatives to merchant GPUs. However, merchant GPUs retain deep software ecosystems and dominance in frontier training, leaving open whether multi-cloud custom silicon can win large external developer share.

    • Supports the claimSiliconANGLE AI
      Oracle brings Google’s TPUs to its cloud, adds ChatGPT integrations
      Oracle Corp. today announced plans to make Google LLC’s TPU artificial intelligence accelerators available through its public cloud. The initiative marks the first time the search giant is supplying the chips to another hyperscaler.

  2. On October 12, 2026, CNBC reported that Microsoft plans to increase its reliance on homegrown AI processors, aiming to reduce dependence on Nvidia chips, expand profit margins, and strengthen its internal AI stack.

    Evidence of hyperscalers moving towards in-house silicon continues to accumulate. Microsoft's reported push to substitute merchant GPUs with its own accelerators follows Alibaba's custom chip deployments, pointing to a broader pattern among tier-one cloud providers seeking to curb merchant chip costs. The development reinforces expectations of procurement shifts to improve operating margins. However, merchant GPUs still dominate frontier training clusters, and it remains an open question whether homegrown silicon can efficiently support third-party cloud customers at scale or will remain focused on internal workloads.

  3. Alibaba unveiled a new proprietary AI chip to support its global data center expansion, adding another major cloud provider to the list of hyperscalers deploying custom silicon.

    Large compute buyers continue to advance their proprietary hardware to reduce reliance on merchant GPUs. Alibaba has introduced a new proprietary AI chip alongside plans to significantly expand its global data center capacity. This deployment by a major cloud provider strengthens the trend of hyperscalers migrating workloads to in-house accelerators, potentially redistributing chip spending across the semiconductor supply chain.