Xsight Labs Raises $300M at $2.8B Valuation to Challenge Nvidia with Programmable Networking Chips

Xsight Labs Secures $300M to Challenge Nvidia's AI Networking Dominance

Xsight Labs, an Israeli chipmaker, has raised over $300 million at a $2.8 billion valuation in a funding round led by Fidelity Management & Research Co., marking a more than fivefold increase from its 2021 valuation1

1

2

2

. The round attracted participation from Intel Capital, T. Rowe Price, Battery Ventures, Valor Equity Partners, Aliya Capital Partners, Atreides Management, Artisan Partners, Diagonal Capital, Key1 Capital, Maverick Capital, Sienna, and Union Group2

2

. Investors are betting on AI data center networking ballooning into a $150 billion market by 2028, with Xsight Labs positioned to capture a meaningful slice through its power efficiency and programmability focus2

2

.

Programmable Networking Chips Target GPU Connectivity Bottleneck

While GPUs captured attention during the AI boom, the chips that connect thousands of GPUs are quietly becoming just as valuable, creating a critical data movement bottleneck in AI data center infrastructure1

1

. Xsight Labs builds the plumbing of the AI data center through two key components: the E1 data processing unit (DPU) and the X2 Ethernet switch, which shuttle data between servers, memory and networks1

1

2

2

. As AI models grow larger, moving data efficiently has become one of the industry's tightest bottlenecks, making the wiring between GPUs a prize of its own in the AI infrastructure stack1

1

.

Open Alternative to Nvidia's Closed Networking Architecture

Nvidia dominates not just the GPU market but also the networking layer through its Mellanox acquisition, creating a closed ecosystem1

1

. Xsight Labs pitches the opposite approach with open and programmable networking chips that run standard Linux and SONiC, allowing operators to rewrite the X2 switch in software after it ships instead of waiting for new silicon1

1

. Chief Executive Yossi Meyouhas stated the market wanted "a high-performance alternative to closed, legacy architectures," positioning Xsight among a wave of startups chipping at Nvidia's grip on the AI stack1

1

2

2

.

High Performance-Per-Watt Drives Competitive Advantage

The E1 data processing unit is the first 800-gigabit data processor to ship, running under 75 watts while delivering four-times the performance per watt of previous-generation designs1

1

2

2

. The chip pairs two 400G ports with 64 Arm-based Neoverse N2 cores built on a 5-nanometer process and 32 megabytes of shared cache, featuring PCIe Gen5 connectivity and encryption engines that run at line rate2

2

. Those 64 cores sit directly on the data pathway, so routing, telemetry and packet inspection occur on every data packet instead of being handed off to a smaller exception-handling engine, enhancing overall network bandwidth2

2

. The chip has achieved SONiC-DASH Hero 800G validation, sustaining more than 14 million connections per second with no dropped packets2

2

.

X2 Switch Delivers 12.8 Terabits with 40% Power Reduction

The X2 Ethernet switch moves 12.8 terabits per second while drawing 40% less power than older switches, providing full-duplex bandwidth at under 200 watts1

1

2

2

. It features first-bit-to-first-bit latency of less than 700 nanoseconds with 128 lanes of 100G PAM4 SerDes, plus a reprogrammable data plane that allows it to be reconfigured for different network applications on the fly2

2

. As AI's electricity bill balloons, every watt saved becomes a compelling pitch, particularly in AI data center racks where accelerator chips consume the bulk of the power budget and cooling capacity1

1

2

2

.

SpaceX Starlink V3 Satellites Validate Technology in Orbit

Xsight's technology is already deployed beyond lab environments, with several global network operators selecting its silicon, including SpaceX's Starlink program1

1

. Last week, Xsight's processors passed testing aboard SpaceX's new Starlink V3 satellites, which will be capable of moving more than 1 Tbps of fronthaul traffic and around 160 gigabytes-per-second of uplink capacity1

1

2

2

. Programmable chips suit space applications where engineers can fix faults through software rather than swapping hardware, and the X2 was chosen for its ability to operate in radioactive environments and thermal extremes while enabling in-orbit adjustments1

1

2

2

.

Veteran Team with Serial Exit Track Record

Founded in 2017 and based in Tel Aviv, Xsight Labs was established by veterans of EZchip, the Israeli networking firm Mellanox bought in 20161

1

. The company's chairman is Avigdor Willenz, a serial chip entrepreneur who has sold past companies to Marvell, Amazon and Intel, bringing both pedigree and investor confidence1

1

. The funding will be used to develop next-generation chips supporting the latest standards backed by the Ultra Ethernet Consortium, expand engineering teams across facilities in Israel, the U.S., Europe and Asia, and increase manufacturing and supply chain capacity for orders from Tier-1 hyperscalers1

1

2

2

.

Positioning Against Broadcom, Marvell in Expanding Market

Xsight's E1 and X2 have been deployed by multiple data center operators globally within edge and network infrastructure, and its products are now being evaluated by a number of Tier-1 hyperscalers2

2

. The startup is increasing its presence in AI data centers as Ethernet closes the gap with proprietary interconnects that link most AI chip clusters today2

2

. Last year, the Ultra Ethernet Consortium published its 1.0 specification defining remote-memory-access transport with multipath routing and congestion control designed for AI training and inference traffic, with Xsight saying the X2 is one of the first switches to meet the new standard2

2

. The question remains whether Xsight's promise of lower power consumption and competitive cost per token processing will hold up against incumbents like Cisco Systems, Broadcom and Marvell, who currently dominate large portions of the data center networking market with deeper customer relationships and broader product portfolios2

2

. The money pouring into AI infrastructure is no longer chasing only GPUs but everything around them, from power stations to the wires in the rack, with Xsight betting the winners in networking will be the open ones1

1

.