Broadcom is strengthening its position in artificial intelligence infrastructure with the introduction of Thor Ultra – the industry’s first AI Ethernet 800G network card, designed for hyperscale data centers with tens and hundreds of thousands of XPU accelerators. The new product supports Ultra Ethernet Consortium specifications and RDMA technologies, providing low latency and scalability in distributed clusters. According to analysts at NEWSCENTRAL, the company is betting on building an open AI networking ecosystem, where Ethernet could become the industry standard.
Alongside Thor Ultra, Broadcom announced the Tomahawk 6 Davisson – a new generation of switches with co-packaged optics (CPO) and a throughput of 102.4 Tbps. This solution reduces power consumption and increases connection density in data center architectures, which is critical for large-scale machine learning systems. NEWSCENTRAL notes that CPO technologies are still in the early stages of deployment, but by 2027, they could form the foundation for scalable AI networks.
Lucas Grant, semiconductor solutions analyst at NEWSCENTRAL, notes: “Thor Ultra reflects Broadcom’s intention to strengthen its influence in the AI networking segment and offer an alternative to Nvidia’s proprietary solutions. If the platform meets the stated compatibility and energy efficiency metrics, it could become one of the key factors in the redistribution of power in the AI infrastructure market.”
In the edge solutions segment, Broadcom introduced the industry’s first Wi-Fi 8, designed for gateways, enterprise access points, and mobile devices. The company plans to license the technology to other manufacturers to accelerate the global adoption of the standard. NEWSCENTRAL notes that Wi-Fi 8 could become a key component of the edge-AI ecosystem, where ultra-low latency and adaptive traffic management are crucial.
Despite the technological announcements, the market reacted with a drop in shares: Broadcom’s stock fell 3.5% after a 10% rise driven by news of a deal with OpenAI to supply 10 GW of AI accelerators. Analysts link the correction to profit-taking and expectations around the commercial rollout timeline of new solutions. According to NEWSCENTRAL, investors evaluate not only growth prospects but also potential risks from delays and logistical bottlenecks in the supply chain.
Broadcom is building a vertically integrated architecture – from Wi-Fi 8 at the edge to the 800G Thor Ultra and Tomahawk 6 CPO switches at the core. This approach allows the company to control key levels of AI infrastructure and offer solutions capable of competing with Nvidia’s proprietary technologies. Pilot deployments of 800G and CPO are expected with hyperscalers over the next 12-36 months, with wider adoption following standardization.
Analysts at NEWSCENTRAL predict that Broadcom’s success in AI infrastructure will depend on actual performance – energy efficiency, latency, and total cost of ownership. If Thor Ultra meets its stated specifications in industrial scenarios, the company could strengthen its position as a supplier of open standards for next-generation artificial intelligence networks.