One signal

Update What changed

Oracle is also expanding its Nvidia installations to multi-gigawatt levels, while adding Google TPUs and SambaNova inference processors to its AI chip portfolio, TradingView reports. The report says Nvidia and Oracle stated the planned expansion, but gives no value for their arrangement.

Oracle is also expected to use its own Acceleron networking technology in some Nvidia-based systems instead of Nvidia networking processors, according to the report. That suggests the expansion is not simply a matter of buying more chips: Oracle is mixing suppliers and its own networking gear across AI infrastructure.

The report cites Nvidia CEO Jensen Huang as having previously suggested Vera Rubin systems could generate up to $40 billion in revenue per gigawatt. That is Huang’s earlier estimate, not a disclosed value for Oracle’s arrangement.

Sources and evidence

• Oracle Expands Nvidia AI Chip Deal, Adds Google and Cerebras Options - TradingView: TradingView, citing GuruFocus, reports that Oracle plans multi-gigawatt Nvidia Vera Rubin installations and is broadening its AI hardware portfolio with Google TPUs, SambaNova inference processors and Oracle Acceleron networking.

Independent WittyWires Watcher; not an official account or feed.

Update What changed

Oracle plans to make Google Tensor Processing Units available through Oracle Cloud Infrastructure in 2027, with deployment expected to expand over several years, Analytics Insight reports. That would give OCI customers direct access to Google’s AI chips without using Google Cloud.

Oracle has forecast capital spending of about US$90 billion to US$95 billion for its 2027 financial year, according to the report. The figure gives a sense of the investment scale behind its expanding AI infrastructure plans.

Oracle says customers are increasingly funding the chips needed for their workloads, but has not disclosed how it will finance its latest purchases, Analytics Insight reports. The spending plan is clear; how the bill is shared is not.

Sources and evidence

• Oracle Expands AI Infrastructure With NVIDIA, Google and Cerebras - Analytics Insight: Oracle plans to offer Google TPUs through OCI from 2027 and forecasts US$90 billion to US$95 billion in capital spending for FY2027; it says customers increasingly fund chips for their workloads but has not disclosed how it will finance the latest purchases.

Independent WittyWires Watcher; not an official account or feed.

Update What changed

Cerebras and Oracle say the Wafer-Scale Engine combines compute and memory on a single wafer-scale processor. The companies say this design reduces data movement and communication between chips, supporting rapid token generation for long reasoning sequences and multi-step AI workflows.

They frame the practical benefit around latency: an agent can make multiple model calls while reasoning, checking its work and taking action, and each call adds time. Cerebras co-founder and chief executive Andrew Feldman says faster inference lets agents do more work without making users wait.

Oracle says Cerebras inference would join existing OCI capabilities, including Dedicated AI Clusters in OCI Enterprise AI.

Sources and evidence

• Cerebras and Oracle Team Up to Bring Cerebras AI Inference to Oracle Cloud Infrastructure - Oracle: Cerebras and Oracle have described the Wafer-Scale Engine design and its intended inference benefits for their planned OCI offering, adding technical detail beyond the existing report of the 2027 plan.

Independent WittyWires Watcher; not an official account or feed.