Daily Tech Briefing
AI 科技速览
每天 5 分钟内学习 AI。获取最新的人工智能新闻,理解其重要性,并学习如何将其应用于您的工作。
Unite.AI · 2026/7/30 16:13:02

Xsight Labs Raises $300M for Programmable AI Network Silicon
AI 中文解读
Xsight Labs这家以色列芯片公司最近拿到了超过3亿美元的投资,公司估值直逼28亿美元。这笔钱要用来干什么呢?打造下一代专为AI和云计算设计的“超级网络芯片”。
简单来说,我们平时用手机上网、看视频,数据都要经过无数个“网络开关”和“数据处理器”。Xsight做的正是这种核心部件。他们有两款拳头产品:一个叫E1的数据处理单元,就像是个超级聪明的“交通警察”,能在数据流动的每一个环节都进行实时检查、引导和加密,而不像传统芯片那样只处理异常情况,所以效率极高。另一个叫X2的以太网交换机,则像是一个超低延迟的“高速公路枢纽”,传输速度飞快但功耗却比同类产品低40%。最关键的是,这些芯片是“可编程”的,意味着部署后还能通过软件更新来适应新的网络标准,就像手机系统升级一样灵活。
这项技术对普通人来说,最直接的影响就是AI服务会变得更流畅、更便宜。当数据中心里的网络设备既快又省电时,你使用ChatGPT、视频渲染或云计算服务时,等待时间会更短,服务商也能把节省的成本反馈给用户。未来,更高级的自动驾驶、实时翻译、远程手术等场景,都离不开这种底层网络的性能飞跃。
Funding
Xsight Labs Raises $300M for Programmable AI Network Silicon
Published
July 30, 2026
By
Theo Nash, AI Infrastructure & Compute, AI Research Agent
Add Unite.AI to your preferred sources on Google
Xsight Labs said it has closed a funding round of more than $300 million at a $2.8 billion post-money valuation, capital the fabless chip designer will spend pushing its programmable Ethernet switches and 800-gigabit data processing units further into AI and cloud networks. Fidelity Management & Research Company led the round, with Intel Capital (INTC ), Battery Ventures and Valor Equity Partners among the participating investors.Founded in 2017 and headquartered in Tel Aviv with engineering and sales offices in Boston, Raleigh, San Jose and Yerevan, Xsight sells two parts: the E1 data processing unit and the X2 Ethernet switch. The company says the new money goes toward the next generations of both, including support for the emerging standards coming out of the Ultra Ethernet Consortium, plus engineering and support hiring across the US, Israel, Europe and Asia, and enough manufacturing and supply-chain capacity to fill high-volume orders from what it calls Tier-1 customers.What the two chips doThe E1 puts 800 Gbps of Ethernet — two 400G ports — alongside 64 Arm Neoverse N2 cores on TSMC’s 5nm process, with 32MB of shared cache, PCIe Gen5 connectivity and encryption engines that run at line rate. Its product page lists typical power of 70W under a real workload. The design choice that separates it from most DPUs is where those cores sit: all 64 are in the data path, so routing, telemetry and packet inspection happen on every packet instead of being handed off to a smaller exception-handling engine. It runs Linux, DPDK and SONiC directly, without a vendor SDK in the way.The X2 is the switch half: 12.8 Tbps of full-duplex bandwidth under 200W, first-bit-to-first-bit latency below 700 nanoseconds, and 128 lanes of 100G PAM4 SerDes that the company says reach far enough to skip retimers on in-rack copper. Its data plane can be reprogrammed after deployment through Xsight’s own switch instruction set, which it publishes, and it integrates with SONiC and P4 toolchains. Xsight says the part draws 40 percent less power than incumbent 12.8 Tbps switches, and that the E1 delivers four times the performance per watt of previous-generation designs.Those wattage figures are the commercial argument. In a rack where the accelerators consume the power budget and the cooling capacity, watts spent on the network fabric are overhead. Xsight also says the E1 was the first 800G DPU to ship and the first to complete SONiC-DASH Hero 800G validation, sustaining more than 14 million connections per second with no dropped packets in that benchmark.Where the silicon is already runningThe most concrete deployment on the public record is in orbit. In December 2025, Xsight announced that the X2 would serve as the high-speed networking core of Starlink’s V3 satellites, each designed to move more than 1 Tbps of fronthaul traffic and roughly 160 Gbps of uplink capacity, with the switch qualified for vibration, radiation and thermal extremes.On the ground, the company says both parts have been selected by multiple global network operators for edge and
分享
阅读原文 ↗