Broadcom says hyperscalers bypass telcos in some AI network projects

TL;DR: Broadcom said hyperscalers are bypassing telecom operators in some projects to connect large GPU clusters across data centers, according to comments reported by Fierce Network. The report said telcos are trying to combine fiber, optical line systems, switches, and services for these inter-data-center links. Broadcom’s comments point to competition between hyperscalers and telcos over who will build and operate AI network infrastructure at scale.

Broadcom said hyperscalers are sidelining telecom operators in some projects that connect massive GPU clusters across separate data centers, according to a Fierce Network report. The article did not specify a date for the underlying comments beyond the report publication.

The Fierce Network report said telecom operators want to bring together the fiber, optical line systems, switches, and services needed to connect GPU clusters in disparate data centers. The article describes these links as part of the infrastructure needed to support AI workloads across multiple sites.

The report indicates that, in some cases, hyperscalers are choosing to handle parts of these projects without telcos. The source excerpt does not name specific operators, hyperscalers, countries, contract values, or project sizes.

Related questions

• What did Broadcom say about hyperscalers and telcos in AI network projects?
• What network components do telcos want to provide for inter-data-center GPU connectivity?
• Did the report name specific hyperscalers or telecom operators?