Israel-based Xsight, which makes server networking and storage chips, raised $300M led by Fidelity Investments at a $2.8B post-money valuation
Graphics processing units are all the rage, but startups developing the networking equipment that connects GPUs to one other are getting some love from investors, too.
Context & Ripple Effects
Xsight’s round extends a run of investment in data-center components beyond compute chips. Fidelity also led Celestial AI’s $250M round, backing a company focused on faster data transfer inside servers, while Israel-based Pliops previously raised capital for flash-memory data-center acceleration.
The common thread is that AI-scale systems depend on the movement and storage of data as well as processing. Xsight’s financing gives that networking layer a more prominent place in the infrastructure funding cycle.
First-order effects
- Xsight gains $300M to develop and commercialize its server networking and storage chips, with a $2.8B post-money valuation setting a new financing benchmark for the company.
- Fidelity deepens its exposure to data-movement infrastructure after its earlier investment in optical interconnect developer Celestial AI.
Second-order effects
- Other startups building networking, interconnect, storage and data-processing silicon may find investors more receptive to the case that GPU deployments require complementary hardware.
- Established suppliers of server networking and storage components face a better-funded specialist competitor, increasing pressure to demonstrate performance and integration advantages.
Third-order effects
- If capital continues to flow into the links between accelerators, storage and servers, AI infrastructure value capture could broaden from compute chips into the bottlenecks that govern how systems scale.
- The pattern favors more vertically coordinated infrastructure stacks, though Xsight’s eventual influence will depend on product adoption rather than funding alone.
The trend: AI infrastructure investment is widening from GPUs to the networking, interconnect and storage layers needed to keep accelerator-heavy systems utilized.