AI Network Suppliers Face Delivery and Margin Tests as Data Centers Expand

AI networking suppliers face a timing test as data-center construction and equipment deliveries determine when accelerator-related demand turns into orders.

Demand estimates cover the globe, while the construction projection counts U.S. additions. Overseas projects may relieve pressure on U.S. power and land, but delayed facilities can push back server and networking purchases.

The central spending comparison is whether AI end users keep pace with infrastructure investment already underway.

The market for cross-data-center connectivity equipment is separate from Ciena’s revenue. Ciena would need to win orders and make sales to capture that demand. Pump lasers and ceramic packaging have constrained deliveries, making supply availability relevant to when the opportunity appears in company results.

Arista Networks could compete for cloud networking work as providers look for alternatives to proprietary systems tied to a single supplier. Arista’s established relationships with Meta and Microsoft include joint design work that could help it pursue projects. Adoption of open Ethernet and the orders Arista wins will shape investor views of the company.

Celestica’s earnings depend partly on which Google tensor processing unit (TPU) projects it retains as Google adds suppliers. The allocation of its TPU programs, especially higher-value racks with stricter delivery requirements, is a key question.

Celestica’s TPU work, AMD Helios and some 1.6T switch programs have different margin profiles, leaving project mix in focus. Investors will be looking for details on order allocation, project ramp-ups and the 2027 margin outlook at Celestica’s Oct. 27 investor day.