Fidelity led the massive funding round, which values the company at more than five times its 2021 worth. The investment underscores a shifting reality in the tech industry: the chips that wire thousands of GPUs together are quietly becoming just as valuable as the processors themselves. As AI models grow exponentially, moving data between servers, memory, and networks has become a critical industry bottleneck.
Challenging the Nvidia Mellanox monopoly
Nvidia dominates not just the AI accelerator market, but also the networking layer that connects those servers, a position it solidified by acquiring Mellanox. Xsight is betting that hyperscalers want an alternative to closed, proprietary hardware. The company builds the plumbing of the AI data center with two core components: the E1 processor and the X2 switch.
Unlike legacy systems, Xsight’s data center chips are open and programmable. They run standard Linux and SONiC, an open-source network operating system originally spearheaded by Microsoft. This approach allows operators to rewrite the X2 switch in software after it ships, deploying updates and fixing bugs without waiting for new silicon tape-outs. The market demanded “a high-performance alternative to closed, legacy architectures,” chief executive Yossi Meyouhas explained.
Efficiency is another major selling point as AI electricity consumption balloons. When GPUs sit idle waiting for data to arrive over the network, hyperscalers waste millions in hardware and energy costs. The E1 is the first 800-gigabit data processor to ship, running under 75 watts. Meanwhile, the X2 switch moves 12.8 terabits per second while drawing 40% less power than older switches. Every watt saved is now a critical pitch to data center operators facing grid constraints.
Proven deployments in space and cloud
Xsight is not just a lab experiment. Several global network operators have already adopted its silicon, including SpaceX’s Starlink. Last week, Xsight processors passed a critical test aboard SpaceX’s new Starlink V3 satellites. Programmable switches are particularly well-suited for space deployments, where engineers can patch software faults remotely rather than replacing physical hardware in orbit.
The ability to reprogram networking gear post-deployment also appeals to massive cloud providers who need to adapt to shifting AI workloads without ripping and replacing racks. Xsight joins a growing wave of startups chipping away at Nvidia’s grip on the broader AI stack, offering flexibility that closed systems cannot match.
A serial exit machine behind the tech
The startup carries a deep pedigree in Israeli semiconductor engineering. Veterans of EZchip, a networking firm Mellanox bought in 2016, founded Xsight in 2017. Its chairman is Avigdor Willenz, a serial chip entrepreneur who has previously sold companies to Marvell, Amazon, and Intel. His presence signals to investors that Xsight has the leadership to navigate the complex semiconductor market.
That track record drew heavyweight investors into the $300 million round, including Intel Capital, T. Rowe Price, Battery Ventures, and Valor Equity Partners. The fresh capital will fund the development of next-generation programmable switches and accelerate a sales push targeting Tier-1 hyperscalers.
What Happens Next
As AI models grow exponentially, the bottleneck is shifting from raw compute power to data movement. Expect hyperscalers to aggressively diversify their networking suppliers to avoid vendor lock-in and manage soaring power bills. The money pouring into AI infrastructure is no longer chasing only GPUs; it is chasing everything around them, from power stations to the wires in the rack.
If Xsight can secure design wins with major cloud providers, it could force Nvidia to lower prices or open up its own networking ecosystem. The battle for AI networking infrastructure is expanding beyond the GPU, and open, programmable switches are the next frontier. Watch for Tier-1 hyperscaler announcements in the coming year as companies look to balance performance with energy efficiency.
— David Kim, technology desk, AXO News