OpenAI × Oracle’s reported $300B ‘Stargate’ deal, explained

TL;DR

The Wall Street Journal reports OpenAI signed a five-year, ~$300B cloud deal with Oracle, starting in 2027. Oracle and OpenAI have not formally confirmed terms. Reuters relayed the WSJ reporting.

Separately, OpenAI and Oracle announced 4.5 GW of new “Stargate” data-centre capacity in the US, bringing >5 GW under development and >2 million chips planned. Parts of Stargate I in Abilene, Texas are already live with GB200 racks and early workloads. 

Microsoft remains a cloud partner for OpenAI as this goes multi-cloud for raw capacity. 

What happened

The deal: WSJ says OpenAI will buy ~$300B of compute from Oracle over ~5 years from 2027. Reuters cites the same report and notes Oracle declined comment. Treat figures as reported, not company-confirmed. 

The capacity build: On 22 July, OpenAI said it will add 4.5 GW with Oracle. That brings >5 GW of Stargate capacity under development, targeting >2 million chips. 

Abilene status: OpenAI says parts of Stargate I are up. Oracle started delivering Nvidia GB200 racks. OpenAI is running early training and inference workloads there. 

Multi-cloud: OpenAI says Microsoft will continue to provide cloud services, including through Stargate. 

Why it matters

Model pace is now a power and supply problem. 4.5 GW is massive. It accelerates training throughput and shortens iteration loops, which can pull forward next-gen model timelines. 

Cost curves shift with scale. If Oracle commits huge capex and long-term capacity, unit costs for training and serving can drop, changing API pricing and margins across the ecosystem. (Inference based on scale economics; monitor Oracle’s RPO and capex.) 

Vendor risk concentrates. A single mega-buyer can reshape chip allocations and lead times for everyone else. Watch GB200 delivery cadence and interconnect buildouts. 

What changes for you (operators, agencies, buyers)

Budgeting: Expect more stable access to high-end GPUs for Oracle OCI regions. Lock pricing and SLAs early for Q4–Q1 projects. 

Multi-cloud strategy: Keep portability. OpenAI confirms Microsoft remains in the mix. Design for Azure + OCI where possible. 

Latency and regions: Track where Stargate capacity lands first. Abilene is already partial. Plan data residency and peering. 

Hiring: OpenAI estimates large job creation tied to the new 4.5 GW. Talent will cluster near build sites. Adjust recruiting. 

Numbers to know

$300B / 5 years. Reported value and term. Not company-confirmed. 

4.5 GW new capacity with Oracle. >5 GW under development. >2M chips targeted. 

GB200 racks already arriving at Abilene. Early workloads are running.

Action plan for this week

Capacity hedge: If you run heavy training or high-QPS inference, open an OCI channel partner ticket and pre-negotiate allocations for 2026–2027. Reference Stargate timelines. 

Portability audit: Confirm your stack runs on both Azure and OCI equivalents. Prioritise container images, storage classes, networking, and observability that are cloud-agnostic. 

Chip planning: GB200 availability is staging in phases. Design rollout plans that can tolerate staged capacity. 

Data strategy: Map datasets to regions likely to get early Stargate buildouts to cut latency and egress. 

FAQs

Is the $300B figure official?
It is reported by WSJ and relayed by Reuters. Oracle declined to comment. OpenAI did not respond at time of reporting. No joint press release with dollar figures yet. 

Is Microsoft out?
No. OpenAI states Microsoft will continue to provide cloud services, including via Stargate. This is a multi-cloud capacity play. 

What exactly is Stargate?
OpenAI’s infrastructure program spanning multiple partners and sites. The July update details power targets, chip counts, and job estimates, plus momentum with Oracle and SoftBank.