Bernstein stated that 1GW AI data center capital expenditures are approximately 34600000000 to 39500000000 USD
BIBIBI
AT A GLANCE
Bernstein's research report stated that building 1GW AI a data center requires approximately 34600000000 to 39500000000 USD in capital expenditures.
Article
Bernstein Breakdown AI Infrastructure costs: highest investment per gigawatt 39500000000 USD
PANews October 10 news, Bernstein research report shows that using different AI accelerator architectures, building 1GW a data center requires capital expenditures of approximately 34600000000 to 39500000000 USD. Among them, NVIDIA's Vera Rubin architecture has the highest construction cost, while OpenAI's in-house developed ASIC architecture, Jalapeno, is relatively less costly. Bernstein lowered its expected per-rack cost for NVIDIA Rubin NVL72 from the previous 9100000 USD substantially to 7520000 USD, a reduction of approximately 17%, mainly reflecting revisions to expectations for HBM prices and NAND storage capacity. The report also noted that the primary economic burden of AI data centers is not electricity costs, but rather substantial capital expenditures and the depreciation they generate.
PANews stated that Bernstein's research report estimates that using different AI accelerator architectures to build 1GW data centers entails capital expenditures of approximately 34600000000 to 39500000000 USD。
02
The report noted that NVIDIA's Vera Rubin architecture has the highest construction cost, while OpenAI's in-house developed ASIC architecture, Jalapeno, is relatively less costly.
03
Bernstein lowered its expected per-rack cost for NVIDIA Rubin NVL72 from 9100000 USD to 7520000 USD, a reduction of approximately 17%。
04
The report attributed the adjustment to cost expectations mainly to revisions to expectations for HBM prices and NAND storage capacity.
05
The research report noted that the primary economic burden of AI data centers is not electricity costs, but capital expenditures and the depreciation they generate.
AI-assisted interpretation
The following is analysis, separate from reported facts. Verify important claims independently.
Building AI a data center requires not only purchasing computing equipment, but also bearing upfront investments in supporting facilities and other items. Bernstein's estimates show that different accelerator solutions result in differing costs for projects of 1GW scale. Even if expected per-rack costs decline, the research report still views overall capital expenditures and subsequent depreciation as the primary economic burden.
Why it matters to readers
These estimates indicate that large-scale AI data centers may require substantial upfront funding, and also show that cost assumptions for accelerators, storage, and other items can affect project budgets.
Beginners should distinguish between capital expenditures and day-to-day electricity costs: the former are investments in construction and asset purchases, while depreciation is the accounting allocation of those asset costs over their useful lives.
Risks and unknowns
The original text is PANews' retelling of a Bernstein research report. As the original report was not provided, its complete methodology and assumptions cannot be verified.
It remains unclear 34600000000 to 39500000000 USD which equipment, facilities, and other costs are specifically included in the estimate.
The actual prices and capacities of NVIDIA Rubin NVL72、HBM and NAND were not disclosed in the evidence.
OpenAI Jalapeno ASIC The specific cost of the solution and the difference versus other architectures were not disclosed.
Related Developments
Loading event timeline…
Related concepts
GW
This term is not in the glossary yet. Browse related concepts in the glossary.
An integrated circuit customized for a single task and designed for maximum execution efficiency is typically more efficient than CPU or GPU in applicable mining networks.
