OpenAI announced a partnership with Oracle to add 4.5 gigawatts of Stargate data‑center capacity in the United States OpenAI. The new build pushes the total under development to just over 5 GW and will run more than 2 million chips OpenAI.
The numbers sound massive, but they are part of a broader promise made at the White House in January: $500 billion to fund 10 GW of AI infrastructure over the next four years OpenAI. OpenAI now says it expects to exceed that commitment thanks to "strong momentum with partners including Oracle and SoftBank" OpenAI. If the math holds, the company could be looking at roughly $50 billion per gigawatt, a price tag that would dwarf most traditional data‑center projects.
Job creation is the other headline. OpenAI estimates the 4.5 GW expansion will generate over 100,000 jobs across construction, operations, and indirect supply chains OpenAI. That includes "direct full‑time jobs needed to operate Stargate data centers" and "short‑term construction roles like the many highly specialized electricians" OpenAI. The claim is ambitious, but the breakdown seems plausible: a gigawatt of AI‑grade compute typically requires thousands of specialized technicians, plus a larger pool of construction workers for the massive power and cooling infrastructure.
The partnership also brings hardware to the site. Oracle started delivering Nvidia GB200 racks last month, and early training and inference workloads are already running at the Abilene, Texas facility OpenAI. Those racks are the workhorses that will push "the limits of OpenAI's next‑generation frontier research," according to the company OpenAI. It's a reminder that the raw compute power is only useful if the software can actually harness it.
From a strategic angle, the move reinforces OpenAI's push to own more of the hardware stack. Microsoft still provides cloud services for OpenAI, but the Stargate platform is meant to be a "long‑term vision to deliver the benefits of AI to everyone" OpenAI. By stacking Oracle's data‑center expertise with SoftBank's financing muscle, OpenAI is trying to avoid bottlenecks that have plagued other AI firms.
I'm skeptical about the job numbers. The estimate lumps together construction, operations, and indirect roles, which makes it easy to inflate the total. Historically, large‑scale data‑center builds have delivered fewer permanent positions than projected, even if the construction phase is labor‑intensive. The real test will be how many of those "direct full‑time jobs" survive once the facilities are fully operational.
The partnership also raises questions about energy consumption. A 5 GW AI farm will draw roughly the same power as a small city. The announcement mentions "reindustrialization" but says nothing about renewable sourcing or carbon offsets. If the goal is to keep the U.S. competitive in AI, the energy strategy will matter as much as the compute capacity.
Bottom line: OpenAI's Oracle deal adds a sizable chunk of compute, bumps the total capacity past 5 GW, and promises a hundred‑thousand jobs. The numbers are impressive, but the real impact will depend on how many of those jobs are lasting, how clean the power is, and whether the extra chips translate into usable AI breakthroughs. Nobody knows how this will play out, and the next few quarters will be the real litmus test.
Sources
- OpenAI, "Stargate advances with partnership with Oracle," July 22 2025, https://openai.com/index/stargate-advances-with-partnership-with-oracle