Foxconn has priced a gigawatt-class NVIDIA Vera Rubin AI datacenter at $47 billion in capital cost, with annual power bills projected at $1.3 billion, according to Wccftech reporting.

The per-gigawatt figure covers the full hyperscale build-out — the building, the Vera Rubin compute stack, networking, cooling, and the electrical infrastructure for a 1 GW IT load. Foxconn, the Taiwanese contract manufacturer that assembles AI server racks for NVIDIA, is positioning itself as a turnkey contractor for the largest AI campuses.

What does the $47 billion per gigawatt figure cover?

The price scales linearly with compute capacity. Building 5 GW of Vera Rubin capacity would require roughly $235 billion in capital outlay, before operating expense.

The figure positions datacenter construction costs in territory previously reserved for national infrastructure projects.

What does the $1.3 billion annual power bill assume?

At one gigawatt of IT load, a facility running near full utilization pulls power continuously. With hyperscale PUE ratios of 1.3-1.5, the datacenter draws 1.3-1.5 GW from the grid, or roughly 11-13 TWh per year.

Foxconn's $1.3 billion figure implies an average industrial power price near $100-115 per MWh, consistent with rates at major US AI datacenter hubs in Northern Virginia, Phoenix, and Dallas.

What is Vera Rubin's place in the AI accelerator stack?

Vera Rubin succeeds NVIDIA's prior Blackwell accelerator generation. Each NVIDIA generation has roughly doubled or tripled compute density per rack. NVIDIA's GB200 NVL72 rack system pulls 120 kW per rack.

Vera Rubin configurations are expected to push well beyond that figure. At one gigawatt of total IT load, the facility would host several thousand racks.

How does this compare to traditional cloud datacenters?

A traditional cloud datacenter of the 2018-2020 vintage ran 50-100 MW of IT load at roughly $1 billion in build cost, or about $10-20 million per megawatt.

Foxconn's $47 billion per gigawatt figure translates to $47 million per megawatt, roughly 2-5x historical hyperscale build cost.

The premium reflects GPU rack density, liquid cooling, high-voltage DC distribution, and structural reinforcement required for multi-megawatt clusters.

What is Foxconn's role in the supply chain?

Foxconn has shifted from consumer electronics assembly into AI server manufacturing. The company operates dedicated production lines for NVIDIA rack systems including the GB200 and GB300.

Foxconn is now positioning for full datacenter contracts rather than rack-only assembly.

What does this mean for hyperscaler capex budgets?

Microsoft, Google, Amazon, Meta, and Oracle have guided to combined datacenter capex of well over $200 billion in 2024-2025. At Foxconn's $47 billion per gigawatt benchmark, 5 GW of Vera Rubin capacity would consume $235 billion — exceeding current annual hyperscale capex.

The numbers underscore why AI infrastructure deployment has begun competing with sovereign debt issuance for capital allocation.

Where can the next AI campuses be sited?

Power-grid availability, not accelerator supply, will increasingly gate where AI capacity gets built. A 1 GW facility needs utility-scale generation and transmission upgrades that typically take 3-5 years to permit and build.

Regions with stranded renewable capacity or behind-the-meter nuclear co-location will absorb the next wave of AI datacenter deployment first.