Stargate Project: Michigan Data Center Costs May Top $56 Billion
Massive GPU and networking costs dwarf initial $16 billion construction budget
OpenAI CEO Sam Altman and Oracle executives gathered in Saline Township, Michigan, to mark progress on "The Saline Barn," a massive $16 billion data center campus. However, new disclosures reveal that the physical buildings represent only a fraction of the total investment. Oracle co-CEO Clay Magouyrk stated that the internal infrastructure—primarily GPUs and specialized networking equipment—could add another $30 billion to $40 billion to the project's final price tag.
Key details
The "Stargate" project in Michigan is emerging as one of the most expensive AI infrastructure developments in the United States. While the initial $16 billion construction budget for the Saline Township site is substantial, the high-density compute requirements of modern AI models drive far greater costs:
- Construction cost: $16 billion for the campus buildings and physical infrastructure.
- Equipment cost: $30 billion to $40 billion for GPUs, high-speed networking, and servers.
- Energy storage: DTE Energy announced a $1.6 billion deal with LG Energy Solution Vertech to provide battery storage, with five of eight systems dedicated to the Oracle site.
- Community investment: The developers committed $10 million for local recreation facilities in Saline.
Why this matters
The Stargate project underscores the shifting economics of AI. Infrastructure costs are no longer dominated by real estate or shells, but by the rapidly depreciating silicon inside. Magouyrk noted that internal infrastructure is significantly more expensive than the buildings housing it and requires more frequent replacement as AI inferencing demand accelerates. This creates a continuous cycle of high-capital expenditure to maintain state-of-the-art compute capabilities.
Context
The scale of the Michigan project reflects a broader trend toward "gigawatt-scale" AI clusters. As model training and inference demands grow, hyperscalers are moving beyond traditional data center hubs to locations with available land and grid capacity. In Michigan, this expansion has met with mixed local reactions, including a 1-gigawatt project by Google in Van Buren Township and a University of Michigan supercomputing facility in Ypsilanti Township, both of which have faced community scrutiny over tax breaks and resource usage.
Risks and open questions
The massive capital requirements of projects like Stargate raise questions about the long-term sustainability of AI scaling. With equipment costs potentially tripling the construction budget, the financial pressure to achieve high utilization is immense. Furthermore, the reliance on specialized GPUs and networking gear makes these projects vulnerable to supply chain fluctuations. Locally, residents remain concerned about the long-term impact on the regional power grid and water resources, despite the promised community benefits.
What happens next
Construction at the Saline Township site is expected to continue through 2027, with the first phases of compute capacity coming online as power infrastructure matures. Watch for further regulatory filings regarding the 1.5-gigawatt energy storage systems being deployed by DTE Energy to support the site's massive power demand.
Source: Planet Detroit Published on AI Usage Global, author: AUG Bot
