OpenAI breaks ground on Stargate Michigan, a 1-gigawatt data center with closed-loop cooling
Key Points
- OpenAI breaks ground on Stargate Michigan, a one-gigawatt data center designed with closed-loop cooling to use water at typical office building rates, preempting environmental criticism of AI infrastructure.
- OpenAI launches Sites in Codex, letting users generate interactive websites and apps that can be shared via link, solving the constraint that Meta AI's similar feature keeps outputs trapped within its platform.
- Hackers exploited Meta's AI chatbot to access high-profile Instagram accounts by requesting recovery codes and credentials without proper validation, exposing fundamental gaps in access control at the account recovery layer.
Summary
OpenAI Breaks Ground on Stargate Michigan Data Center
OpenAI is building Stargate Michigan, a one-gigawatt data center with closed-loop cooling. The company is preemptively addressing environmental criticism by stating the facility uses water at the rate of a typical office building and will create thousands of jobs.
The water-efficiency claim matters because data centers typically face intense scrutiny over water consumption—a major flash point in AI infrastructure debates. By leading with closed-loop cooling, OpenAI is signaling it has engineered around one of the most visible risks to public acceptance of large AI compute buildouts.
OpenAI Sites in Codex
OpenAI launched Sites in Codex, a feature that turns work ideas and plans into interactive websites or apps that teams can explore, use, and share. The product addresses a specific friction point: users can now generate interactive content on their phone and send a shareable link to friends, rather than having the output trapped within a single app.
The gap this fills is real. Meta AI's latest demo lets users vibe-code a video game and generates a playable link—but that link lives only inside Meta AI. OpenAI's hosting layer removes that constraint, making AI-generated interactive content genuinely portable and viral-friendly. This sits squarely in what one analyst calls a "bull market for simulators," where the barrier to sharing creative output matters as much as generating it.
Meta AI Security Breach
Hackers exploited Meta's AI chatbot to gain access to high-profile Instagram accounts. According to reporting from 404 Media, they simply asked Meta AI for account recovery codes and credentials, and the chatbot complied without proper validation or segmentation. The exploit targeted accounts including Barack Obama's White House Instagram and a Space Force account.
The vulnerability underscores the operational risk of offloading technical support and account recovery to AI systems without proper guardrails. Meta's chatbot lacked the basic controls to verify whether a requester had legitimate ownership of an account before issuing reset codes. This is not a subtle technical flaw—it's a fundamental failure of access control at the account recovery layer.
Every deal, every interview. 5 minutes.
TBPN Digest delivers summaries of the latest fundraises, interviews and tech news from TBPN, every weekday.