The AI megacampaign "Stargate" hit a $500 billion snag: the conflict between OpenAI, Oracle and SoftBank is stalling the project

Short about the Stargate project

At the beginning of Donald Trump's second term, the megaproject Stargate was announced, aiming to invest several hundred billion dollars in building AI data centers across the United States. However, the project's implementation is slowing down due to disagreements among partners.

Participants and their commitments
Company | What they promised to do | Amount, USD
OpenAI | Build up to 10 GW of computing power within three years (plans are now on hold) | –
Oracle | Construct data centers with a total of 2 million accelerators, and sell OpenAI capacity for $300 billion over five years | $300 billion (sales), 2 million accelerators
SoftBank | Build a data center in Texas and lease it to OpenAI | –

All three companies announced an intention to invest at least $500 billion in developing AI infrastructure in the United States.

Why the “Stargate” project is slowing down
1. Disagreements over resource management

- Tom’s Hardware, citing The Information, reports that participants have not agreed on who will control the built computing capacity.

2. OpenAI’s strategy shift

- Initially the company planned to build a data center solely for its own needs, but the need for massive capital expenditures led it to seek partners.

- After negotiations with Oracle and SoftBank, the plans to launch 10 GW were called into question.

3. Negotiations with SoftBank

- Originally a joint 1‑GW data center in Texas was planned, but OpenAI switched to partnering with Oracle.

- In the end, the parties agreed that SoftBank would handle construction while OpenAI would lease equipment. As a result, SoftBank had to postpone buying cloud provider Switch for $50 billion due to regulatory obstacles.

4. Oracle’s financing

- In the last six months, Oracle issued bonds to raise the funds needed for the project.

Conclusion
At present, the Stargate project is “not standing still,” but OpenAI must admit that it cannot fully control all built data centers. Questions of responsibility allocation and resource management remain key obstacles to further AI infrastructure development in the United States.