Broadcom’s latest challenge to Nvidia is not another processor designed to replace its market-leading GPUs.
It targets the network that enables thousands of those processors to operate as a single computing system.
DriveNets on July 1 unveiled two AI-networking platforms built around Broadcom’s Tomahawk 6 switch chip.
The systems promise faster connections with fewer networking layers, potentially reducing delays, power consumption and spending on optical equipment.
For AVGO investors, the attraction is straightforward as the company can capture more AI-infrastructure spending without needing to defeat Nvidia in GPUs.
A new system turns Tomahawk 6 into a commercial catalyst
DriveNets’ new 2600SL and 2601S platforms each provide 102.4 terabits per second of switching capacity across 64 ports running at 1.6 Tbps.
The liquid-cooled and air-cooled systems are scheduled to begin shipping during the third quarter of 2026.
Those specifications matter because training advanced AI models requires enormous numbers of accelerators to exchange data quickly.
A network bottleneck can leave costly GPUs and custom processors sitting idle, reducing the return on a data centre’s investment.
Broadcom says Tomahawk 6 can connect as many as 128,000 accelerators through only two switching tiers.
Bob Wheeler, an analyst at LightCounting, said in Broadcom’s March product announcement that using fewer tiers can reduce latency, simplify congestion control and cut the number of optical links required.
The DriveNets launch also marks a commercial step forward.
Broadcom began shipping Tomahawk 6 in production volumes in March, less than three quarters after the chip began sampling.
In Broadcom’s March product announcement, Dell’Oro Group vice-president Sameh Boujelbene said the company was “translating its roadmap into real-world deployment” by moving Tomahawk 6 into production shipments.
Also read: Broadcom extends Apple chip partnership through 2031, stock climbs 5%
Broadcom does not need to replace Nvidia’s GPUs
The more contrarian investment argument is that Broadcom can benefit even when customers continue buying Nvidia processors.
Nvidia’s advantage extends well beyond GPUs. Its NVLink, InfiniBand and Spectrum-X Ethernet products allow the company to control more of the system linking accelerators together.
That integrated approach can deliver strong performance, but it also makes customers more dependent on Nvidia’s hardware and software ecosystem.
Broadcom is attacking that control through Ethernet. Tomahawk switches can support networks containing Nvidia GPUs, Google TPUs and other custom accelerators.
That gives cloud operators greater freedom to combine products from several suppliers rather than buying an entirely proprietary system.
The broader market is already moving towards Ethernet.
Dell’Oro said sales of Ethernet switches used in AI back-end networks more than doubled during the first quarter of 2026 and represented about two-thirds of switch sales in AI clusters.
Boujelbene said Ethernet maintained a “clear lead” despite a recovery in InfiniBand demand.
JPMorgan analyst Harlan Sur expects Broadcom to retain about 70% of the AI Ethernet switching-silicon market, citing its rapid product cycle and the technical barriers facing competitors.
Sur estimates Broadcom’s AI-networking revenue could more than double to at least $45 billion in fiscal 2027.
Mizuho analyst Vijay Rakesh has also identified scale-up Ethernet as a potential growth engine.
MarketWatch reported that he believes it could eventually contribute about one-quarter of Broadcom’s networking revenue and help the company compete with NVLink.
Mizuho raised its Broadcom price target to $530 from $480 following the June earnings report while retaining an Outperform rating.
The post Broadcom stock’s latest catalyst targets Nvidia where it looks strongest appeared first on Invezz
