[News] Broadcom’s New Networking Chip for AI Reportedly Built on TSMC’s 3nm; Full Shipments Expected in July

TSMC Chairman C.C. Wei said at yesterday’s shareholder meeting that AI demand remains strong despite tariff concerns—and the momentum is clear. On June 3, major client Broadcom announced it has begun shipping its latest AI-focused networking chip, the Tomahawk 6, according to its press release.

As highlighted by Reuters, the Tomahawk 6 is being manufactured using TSMC’s cutting-edge 3nm process. Meanwhile, the chip launch also drew praise from major IT channel vendors led by AMD, as per CRN.

According to the Economic Daily News, Senior Vice President Ram Velaga didn’t reveal the first customers using Tomahawk 6 but said they include major cloud service providers and networking equipment makers integrating Broadcom’s tech into their products.

New Chip May Cost Nearly Double Its Predecessor

Citing Velaga, Bloomberg reveals that the company started delivering the Tomahawk 6 switch chip last weekend, with full shipments expected by July. The new chip packs the power of six older versions, with a price tag nearly twice as high—but still under $20,000 each, Bloomberg suggests.

Notably, Reuters also points out that the Tomahawk 6 is the first in its lineup to adopt a chiplet architecture—combining multiple chips in a single package, a design approach popularized by companies like AMD.

Tomahawk 6: Designed to Meet AI’s Growing Demands

Broadcom claims the Tomahawk 6 to be the world’s first single-chip switch to hit 102.4 terabits per second, doubling the bandwidth of any current Ethernet switch.

Designed for next-gen AI networks, Tomahawk 6 offers massive scale, high energy efficiency, and advanced AI features, with support for 100G/200G SerDes and co-packaged optics (CPO) for maximum flexibility, according to its press release.

Reuters suggests that Broadcom’s networking chips are playing a bigger role in the AI boom, as building AI data centers requires linking together thousands of chips. That demands high-performance networking gear—exactly where Broadcom’s Tomahawk series comes in, the report adds.

Citing Velaga, Reuters notes that with Tomahawk 6, Broadcom has doubled performance and enhanced capabilities, making it powerful enough to support massive AI clusters—sometimes connecting over 100,000 GPUs.

According to CRN, the switch chip’s design supports unified AI training and inference networks at massive scale. Features like advanced telemetry, dynamic congestion control, and fast failure detection—powered by Broadcom’s Cognitive Routing 2.0—make it possible, the report indicates.

Read more

• [News] TSMC Fast-Tracking 3nm Progress in U.S., Equipment Installation Reportedly Starting Mid-2026
• [News] Intel’s Comeback? NVIDIA and Broadcom Reportedly Eye 18A, with a Boosted Version in the Works

(Photo credit: Broadcom)