OpenAI and Oracle announced on October 27, 2026 that Oracle applications and data are coming into ChatGPT, and that Oracle is shipping new products built on OpenAI models. The same day, Cerebras said its fast inference will arrive on Oracle Cloud Infrastructure (OCI) in 2027. This post separates what the companies actually said from what is still a plan.
Quick reference
| Question | Answer |
|---|---|
| What is Oracle AI Workspace? | An Oracle product, in development, that brings Oracle apps and data into ChatGPT |
| What ships alongside it? | Oracle AI Security Mesh, Oracle AI SMB Bundle, Oracle App Composer |
| Can I buy OpenAI through Oracle? | Yes for US customers, via Oracle Marketplace and eligible Oracle Universal Credits |
| Oracle's own usage numbers? | 130,000 active ChatGPT users and 95,000+ active Codex users, as reported by Oracle |
| Cerebras on OCI? | Planned for calendar year 2027 |
| Primary sources | OpenAI, Cerebras press release |
What OpenAI says Oracle AI Workspace will do
According to OpenAI's announcement, the partnership aims at an "agent-ready enterprise where people and AI agents work across business applications in one connected environment." Oracle is developing Oracle AI Workspace so employees can describe what they need, find information, and complete approved business workflows without leaving ChatGPT.
The example OpenAI gives is a new finance team member who uses ChatGPT to investigate costs, understand purchasing decisions, and see what remains before month-end close, with information pulled from finance, procurement and planning applications. Access, the announcement says, would be governed by each organization's permissions and approvals, with unified sign-in and simpler administration.
Note the language. Phrases like "is developing", "the goal is" and "the intended experience would" signal a direction, not a shipped product. No release date or pricing for Oracle AI Workspace appears in the announcement.
The three Oracle products built with OpenAI
Alongside the workspace, Oracle announced updates and offerings that use OpenAI models:
- Oracle AI Security Mesh. Designed to help customers find and fix vulnerabilities in applications and infrastructure using OpenAI cyber models. Oracle manages the testing infrastructure, so customers run these capabilities within their authorized scope. This follows OpenAI's recent push on defensive cyber tooling, which also surfaced in its Daybreak work with Cloudflare.
- Oracle AI SMB Bundle. Combines ChatGPT Business and Codex with Oracle applications and recurring Oracle Cloud Infrastructure credits, aimed at small businesses that want to build products and manage operations.
- Oracle App Composer. Helps teams build apps with Codex and other AI tools, connect them to approved Oracle data and workflows, and share them across an organization. Oracle manages deployment on OCI, with company permissions and access controls built in.
For builders, App Composer is the most interesting piece: it points to a pattern where a coding agent builds internal apps against an enterprise system of record, with the vendor owning hosting and access control. The question every team will ask is how much of the agent's work is reviewed before it touches production data. OpenAI's own Codex Auto-review feature shows one approach to cutting approval prompts while keeping a reviewer in the loop.
Buying OpenAI through Oracle Marketplace
The announcement also says Oracle customers in the US can purchase access to the OpenAI API, Codex and ChatGPT Work through Oracle Marketplace, using existing procurement processes and eligible Oracle Universal Credits. For a procurement team, that matters more than any demo: existing Oracle spend commitments can be applied to OpenAI usage, avoiding a new vendor onboarding. It mirrors the way OpenAI models arrived on Google Cloud's Model Garden, where the real change was billing and procurement rather than the models.
Oracle's own numbers
OpenAI says Oracle reported 130,000 active ChatGPT users and more than 95,000 active Codex users, and that Oracle's recruiting team uses ChatGPT Work for market research that once took two to four days and now takes 15 to 20 minutes. These are customer-reported figures published in the vendor's own announcement. They show adoption inside Oracle but are not independent measurements, and the recruiting example is a single anecdote with no stated methodology.
Cerebras inference on OCI, planned for 2027
In a separate press release from Las Vegas on October 27, Cerebras and Oracle announced plans to make Cerebras fast inference available on OCI for AI labs, enterprises and developers. It builds on Oracle's existing Dedicated AI Clusters in OCI Enterprise AI.
Cerebras CEO Andrew Feldman framed it around agents: every step an agent takes needs inference, so faster inference means more work done without users waiting. Mahesh Thiagarajan, president of OCI, called it "another ultrafast and deterministic option" in Oracle's compute portfolio. The release says Cerebras Inference on OCI is planned for calendar year 2027.
If you want the hardware background, our explainer on how Cerebras runs a 3-trillion-parameter model across 59 wafers covers why keeping weights in on-chip memory gives very high token rates. Cerebras is also already part of OpenAI's stack, as in the ultrafast GPT-5.6 Sol variant served on Cerebras.
Where this fits in OpenAI's enterprise push
The Oracle announcement is one of several enterprise moves OpenAI listed on its own site this month. Its related-posts list on the announcement page includes an expanded partnership with Atlassian on October 6 and Uber bringing ChatGPT Work to all employees on October 22. The common thread is that OpenAI is no longer selling only a chat window or an API: it wants ChatGPT to be the place where employees reach their company's software, with partners supplying the systems of record.
That strategy has a trade-off worth stating plainly. For the application vendor, being reachable inside ChatGPT widens distribution, but it also puts a third party's interface between the vendor and its users. For the customer, one conversational front end for finance, procurement and planning is convenient, but it concentrates permissions, logs and data flows in a new layer that security teams have to understand. Oracle's own framing, that access stays "governed by each organization's permissions and approvals", is the right promise; whether it holds in practice will only be visible once customers deploy it.
It also shows why Oracle bundles both sides of the stack. Oracle sells the applications, the cloud credits, and now the OpenAI access, so a customer's spend can stay inside one commercial relationship. That is attractive to procurement and consistent with Oracle's wider infrastructure ambitions, which we covered in the Tencent chip lease story. For OpenAI, the reverse holds: Oracle's installed base of enterprise customers is a distribution channel it does not have to build.
The chip claims: what we could and could not confirm
A CNBC report, as summarized in a headline on Techmeme, says Oracle also adds ChatGPT and Codex to its app stack and expands its OCI lineup with Google TPUs, OpenAI's Jalapeño chip and Cerebras hardware. We could not open the CNBC article itself, and OpenAI's and Cerebras's own pages do not mention TPUs or Jalapeño on OCI. So:
| Claim | Status |
|---|---|
| Oracle apps in ChatGPT via Oracle AI Workspace | Confirmed by OpenAI, in development |
| Oracle AI Security Mesh, SMB Bundle, App Composer | Announced by OpenAI |
| Cerebras inference on OCI in 2027 | Confirmed by the Cerebras press release |
| Google TPUs on OCI | Reported in a CNBC headline; not verified in a primary source we could read |
| OpenAI's Jalapeño chip on OCI | Same: reported in the headline, not confirmed by OpenAI's page |
Jalapeño itself is real. OpenAI has presented it as its first custom inference chip built with Broadcom, covered in our Jalapeño explainer. But whether it is offered to Oracle customers is a separate claim that needs a primary source. Until Oracle or OpenAI states it, treat it as reported, not confirmed.
What this means for what you build or pay
- If you run Oracle applications, expect ChatGPT to become another front end to them. The open questions are permissions, audit logs and what data leaves Oracle's environment, none of which the announcement answers in detail.
- If you are buying AI, the Marketplace route may let existing Oracle credits pay for OpenAI usage. Verify which credits are eligible and whether volume discounts apply.
- If you build agents, fast inference matters for multi-step workflows because latency compounds per call. Cerebras on OCI is a 2027 option, not something to plan a launch around today.
- If you care about security, AI Security Mesh uses OpenAI cyber models under Oracle-managed infrastructure and authorized scope. Ask how findings are validated before acting on them.
Open questions
- Pricing and availability dates for Oracle AI Workspace and Security Mesh.
- How ChatGPT enforces Oracle permissions at the data level.
- Which OCI regions get Cerebras first.
- Whether TPUs or Jalapeño appear in Oracle's own documentation.
- How the arrangement interacts with Oracle's much larger infrastructure commitments, such as its leasing deals for AI chips.
Related reading
Details are accurate as of October 27, 2026; product availability and plans may change.