Oracle Expands Nvidia Chip Deal and Introduces Google TPU for the First Time
nashnova research
Oracle announced it will scale Nvidia's latest AI servers to multi-gigawatt levels while onboarding Google TPUs and chips from Cerebras and SambaNova for the first time — a clear signal that no single chip maker can feed the cloud giant's AI appetite amid a persistent silicon shortage.
What exactly is expanding in the Nvidia deal?
Oracle will deploy Nvidia's next-generation "Vera Rubin" AI servers at multi-gigawatt scale — This means → data-center power and compute commitments are climbing to power-plant territory.
At the same time, Oracle plans to swap in its own "Acceleron" networking chips in some Nvidia-based configurations, replacing Nvidia's networking silicon.
In plain terms = Oracle is still buying Nvidia's GPUs — the engines — but wants to build the highway connecting them itself.
Why bring in Google TPUs and other rivals now?
Oracle will run Google's TPU (Tensor Processing Unit — Google's custom AI chip) on its own cloud infrastructure. This is the first time a TPU has operated inside another hyperscaler's data centers.
Also joining the lineup: chips from Cerebras Systems and Intel-backed SambaNova Systems.
Oracle cloud infrastructure president Mahesh Thiagarajan said the chip shortage remains a persistent issue; offering multiple accelerator options helps ease the pressure. He stressed this is not a move away from Nvidia GPUs, calling them "the best and most versatile."
This reflects an industry reality: AI compute demand is so large that no single chip vendor can satisfy it alone. Diversification is not distrust — it is necessity.
What does the expanded OpenAI partnership mean?
Oracle is embedding ChatGPT and Codex into its software applications, letting users run database operations in natural language. Thiagarajan said this "lets anyone build, deploy, and manage applications."
OpenAI CEO Sam Altman appeared at Oracle's AI World keynote; co-founder and president Greg Brockman said "more and more enterprises are choosing OpenAI."
OpenAI's in-house Jalapeño chip will also run on Oracle's cloud — This means → OpenAI is becoming not just a software partner but a hardware supplier to Oracle.
What signal does Cerebras entering Oracle's cloud send?
Cerebras chips will run on Oracle's cloud infrastructure, roughly seven months after Amazon Web Services announced it would carry Cerebras hardware.
A Cerebras spokesperson called the partnership significant but declined to disclose specifics, citing a quiet period — a pre-IPO phase when a company cannot publicly share business details.
This reflects Cerebras's accelerating push into mainstream cloud platforms. Landing Oracle after Amazon signals a shift from niche alternative to mainstream supplier.
What are investors most worried about?
Oracle shares have fallen more than 30% this year, driven in part by doubts over how the company will finance its massive data-center pipeline.
A multi-billion-dollar data-center project in New Mexico, code-named "Project Jupiter," has faced local community opposition. Oracle last month issued a force-majeure notice — a legal declaration pausing contractual obligations due to unforeseen obstacles — to backer Blue Owl Capital, but insists the project remains on track.
Investors are waiting for an upcoming investor day, focused on two questions: whether Oracle can back up its FY 2030 revenue target, and whether cloud-contract margins are eroding as capital costs rise.
In plain terms = Oracle has drawn a very large AI blueprint. The market no longer wants more partnership announcements — it wants to see where the money comes from and whether margins hold.
市场有风险，内容仅供研究参考，不构成投资建议。