1Finity Unveils Optical NIC for Long-Distance Data Links
The 1Finity Ultra Optical System N100 connects data centers up to about 100km apart at 800Gbps.
Updated on Nov. 18, 2026 in Data Centers

1Finity, a Fujitsu company, has developed the 1Finity Ultra Optical System N100, an optical network interface card that links remote data centers with direct, low-latency optical connections. The company will begin accepting advance orders in November 2026.
Why it matters
Data centers face limits on power and physical space, which is pushing operators to spread computing resources across multiple sites. Conventional Ethernet suffers latency and packet loss over long distances, which can leave GPUs idle, and the N100 is positioned to address that gap.
The N100 has two 400Gbps QSFP56-DD optical interfaces, delivers up to 800Gbps over roughly 100km, and is PCI Express compliant. Its proprietary xRDMA transfer technology is manufacturer-reported and has not been independently verified.
The players
1Finity
1Finity is a global network solutions provider and a Fujitsu company, based in Japan.
The details
The card is installed in servers at each data center location to connect remote servers over optical links. Its proprietary xRDMA technology is designed to offset the effects of communication latency and packet loss across long distances. The product was developed using technologies from the NEDO Green Innovation Fund project 'Development of Next-Generation Green Data Center Technologies'.
Timeline
November 2026: Advance orders for the 1Finity NIC begin.
April 2027: Global availability of the 1Finity NIC begins, with gradual rollout ahead of that date.
The Tech Race
The N100 is an attempt to rethink how distant data centers are joined, moving past the latency limits of conventional Ethernet interconnects. It places Fujitsu's 1Finity in a contest to supply the networking hardware that increasingly ties together distributed AI computing.
For operators planning multi-site AI infrastructure, the N100 offers a hardware option for linking facilities at up to 100km with lower latency than standard Ethernet. Advance orders open in November 2026, with global availability following in April 2027.
The takeaway
Distributed data center designs depend on the network between sites as much as on the servers themselves. Teams evaluating multi-site AI capacity should weigh interconnect latency and packet loss alongside power and space constraints.
Further reading
For more on how operators are rethinking facility design, see Data Centers.
Source note: This article includes information reported by Thefastmode.



