Google’s newly unveiled open-source liquid-cooling design for artificial intelligence data centres could accelerate the commoditisation of some cooling equipment used in AI infrastructure, although it is unlikely to pose an immediate threat to established suppliers such as Vertiv and nVent, Bernstein analysts said in a research note.
Alphabet (NASDAQ:GOOGL)-owned Google recently released specifications for "Brazos", a liquid-to-air cooling distribution unit (CDU) designed for the Open Compute Project ecosystem, a consortium backed by major technology companies including Google, Microsoft and Meta.
Bernstein said the design was intended as a reference architecture for manufacturers rather than a product Google plans to build itself.
The analysts said the specifications appeared aimed at supporting AI inference workloads in existing data centres through relatively simple retrofits, rather than cooling the most advanced AI training systems.
At 60 kilowatts of cooling capacity, the Brazos design cannot support a single Nvidia Blackwell rack, which typically requires about 120 kilowatts of cooling capacity, Bernstein said.
Instead, the analysts believe the system is geared toward lower-density AI inference deployments that can be installed in legacy facilities that already have direct-current power infrastructure.
They said inference demand is expected to grow faster than training demand through 2030, making retrofit solutions increasingly attractive for hyperscale cloud operators seeking to bring capacity online quickly.
The brokerage said the broader implication for the cooling industry is that simpler, lower-specification cooling units could become increasingly standardised, potentially eroding margins for suppliers.
"There is real commoditization risk of the inference CDU ecosystem," the analysts wrote, noting that Brazos-class systems are less technically demanding than Google’s earlier "Deschutes" liquid-to-liquid cooling specification and may not require the same engineering expertise that has helped companies such as Vertiv and nVent command premium pricing.
Bernstein also flagged the possibility that some hyperscalers could shift spending toward retrofitting existing facilities instead of building new data centres if construction delays persist.
While the firm said it had not yet seen evidence of significant stranded capacity, it cited anecdotal reports of customers pushing back deliveries because projects were not ready. Standardised cooling systems such as Brazos could provide operators with greater flexibility to accelerate AI deployments in existing facilities, the analysts said.
source https://www.investing.com/news/technology-news/google-cooling-initiative-could-reshape-ai-datacentre-retrofit-market-bernstein-4751677
