Wccftech reports Foxconn Chairman Young Liu cited capital costs for NVIDIA's Vera Rubin AI datacenters as high as $47 billion per 1 GW installation. Per Wccftech, Young Liu estimated a 1 GW facility would contain about 3,557 racks, with each rack costing roughly $9.1 million, and an annual electricity bill near $1.3 billion. Wccftech also reports Young Liu saying hardware depreciation would be approximately six times the power bill, and cites a Morgan Stanley Research bill-of-materials for related Vera Rubin servers. First Vera Rubin systems are being delivered to major cloud providers for validation and testing, per Wccftech.
What happened
Wccftech reports Foxconn Chairman Young Liu gave example figures for building AI datacenters anchored on NVIDIA's Vera Rubin. Per Wccftech, Young Liu estimated a 1 GW installation could require up to $47 billion in capital, contain about 3,557 racks, cost roughly $9.1 million per rack, and incur an annual electricity bill of about $1.3 billion. Wccftech reports Young Liu saying hardware depreciation would be approximately six times the power bill, and cites a Morgan Stanley Research bill-of-materials for VR200 NVL72 servers.
Technical context
These reported figures reflect infrastructure line items that typically scale nonlinearly with density: facility construction, power delivery and cooling, rack-level interconnects, and specialized chassis. For practitioners, the headline numbers imply very large capital and operating expenditures when designing for sustained, dense GPU or accelerator deployments expected for agentic workloads and large inference or training clusters. A separate Morgan Stanley estimate (PC Gamer, May 2026) put the per-rack bill of materials for a VR200 NVL72 at approximately $7.8 million - a figure representing what cloud service providers pay for components, excluding integration and facility costs.
Context and significance
Public reporting frames Vera Rubin as NVIDIA's next-generation architecture that cloud providers are now validating, and the Foxconn figures place a spotlight on the downstream cost structure for hyperscale AI. Industry observers have previously documented that power and facility costs often dominate total cost of ownership for AI clusters; the Wccftech-reported numbers quantify that effect at extreme scale.
What to watch
For practitioners and infrastructure planners: monitor independent vendor bill-of-materials estimates, published rack-level power density figures from NVIDIA and OEMs, and early cloud-provider deployment reports for Vera Rubin. These indicators help translate headline capital figures into per-model and per-inference cost estimates relevant to capacity planning and pricing.
Scoring Rationale #
Foxconn's per-gigawatt cost estimates for Vera Rubin datacenters provide meaningful context for infrastructure planning and total cost of ownership analysis, corroborated by independent Morgan Stanley bill-of-materials research. The story is important for practitioners planning GPU-dense deployments but is single-source trade-press reporting on illustrative cost figures, placing it at the upper edge of 'notable' rather than 'major'.
Practice interview problems based on real data
1,500+ SQL & Python problems across 15 industry datasets — the exact type of data you work with.