How Broadcom Competes with NVIDIA in AI Networking

Broadcom competes in AI networking by making special hardware. This hardware helps big data centers do hard jobs. The company works on custom ASICs and optical technology. These help save money and work better than regular GPU solutions. Broadcom is a top company in this area. It works with big customers like Google and Meta. Its new ideas help save power and move data faster. This makes Broadcom a strong rival to NVIDIA.

Evidence Description	Significance
Broadcom's 224G SerDes gives 1.6 Tbps speed. It can use up to 1,024 lanes on each chip.	This makes AI jobs faster and uses less energy.
Co-Packaged Optics technology cuts signal loss and power use by up to 65%.	More bandwidth helps process big data sets quickly.

Key Takeaways

• Broadcom makes special ASICs and uses optical tech to help AI networks. - Its chips use less power and cost less money. This makes them good for large data centers. - Broadcom's products move data quickly and work well. - The company works with big tech companies like Google and Meta. - Broadcom's solutions are a good choice instead of NVIDIA's GPUs.

Broadcom Competes in AI Networking

Networking Chips and Custom ASICs

Broadcom makes special chips for AI jobs. These chips are made to work fast and use less energy. They help big data centers move lots of data quickly. Broadcom has many types of AI chips for different needs.

• Tomahawk Ethernet switch silicon helps run big networks.

• Jericho router chips move data very fast.

• Optical interconnect solutions help servers talk to each other.

• Custom ASICs make AI computing better.

• Network Interface Cards (NICs) keep connections strong.

• Ethernet switches are set up for AI data.

Broadcom’s AI chips are better than normal networking chips in some ways. The table below shows how they are different:

Feature	Broadcom's Custom ASICs	Standard Networking Chips
Specialization	AI workloads, inference	General-purpose tasks
Efficiency	Higher	Lower
Cost-effectiveness	Greater	Variable
Performance in AI Inference	Superior	Inferior
Networking Standards	Open Ethernet	Proprietary solutions

Broadcom’s chips save more energy and money. The company makes chips for special AI jobs, like inference. Broadcom sells a lot of high-end data center chips. Its chips use open Ethernet, not closed systems.

New chips like Jericho3 and Ramon3 do cool things. They help with fast data movement and quick fixes if something goes wrong. The Jericho4 switch family keeps data moving without loss in big data centers. It can send data far and control traffic jams. The 7060X6 Series uses less power and saves energy. The 7800R4 Series grows big networks with two Jericho 3-AI chips on each card. The 7700R4 Switch sends data in one hop across many places. Meta uses this for Llama 3 training, which shows it works well.

Wired Connectivity and Optical Solutions

Broadcom also works on wired and optical tech. The company makes chips that move lots of data fast. Jericho 4 chips have fast memory and HyperPort tech. These chips help connect AI jobs in many places. They keep data moving quickly and cut down wait times.

Broadcom’s optical tech makes AI data centers work better. The company showed its new AI products at OFC 2026. These include XPU, Ethernet, Optics, SerDes, DSP, and PCIe. The Taurus 400G/lane optical DSP lets data centers get faster without using more power. Taurus makes each lane move twice as much data. This helps big AI clusters work better. Broadcom’s products give lots of data speed for small, low-power modules. They also have new 3.2T transceivers. These save power and keep things cool in crowded racks. Broadcom’s tech works for both 100T and 200T networks.

Broadcom started making co-packaged optics in 2021. The third Tomahawk ultra chip, Tomahawk 6 – Davisson (TH6-Davisson), is now being sent to early users. Broadcom worked with Micas Networks, TSMC, HPE, and others. TH6-Davisson is an Ethernet switch that can move 102.4T bps with optical power.

• Jericho 4 chips give fast and strong connections.

• Taurus optical DSP makes each lane twice as fast for AI.

• Tomahawk ultra switches give lots of speed and save power.

• Co-packaged optics help AI jobs grow in the future.

Broadcom uses Ethernet and custom chips to compete. The company is a big rival to AMD and Intel in AI hardware. Big cloud companies want something better than NVIDIA’s GPUs because those cost a lot and are not always efficient. Broadcom’s chips use less energy and money, so they are good for big AI projects.

Broadcom keeps making new AI chips and tech. Its products help big data centers do hard AI jobs quickly and well.

NVIDIA Competition and AI Platforms

GPU Solutions for AI Models

NVIDIA is a leader in making powerful chips for AI. These chips help computers learn and make decisions. The DGX A100 and H100 chips are used for training AI. The H200, B300, and GB300 chips work with HGX servers. These servers use many chips to get more power. NVIDIA also has DGX Cloud, which lets companies use GPUs online. NVIDIA Dynamo is a tool that makes AI models faster, up to 30 times. RTX PRO Servers use RTX PRO 6000 Blackwell chips. They help big companies do lots of computer work.

• DGX A100 and H100 chips help train AI.

• HGX servers use many chips for more power.

• DGX Cloud lets companies use GPUs online.

• NVIDIA Dynamo makes AI models faster.

• RTX PRO Servers do big computer jobs.

NVIDIA competes by making strong chips for AI. The company is known for its powerful chips, not for networking.

Integration with Networking Infrastructure

NVIDIA connects its chips with strong networking tools. Spectrum-X Ethernet gives fast data speeds and keeps tasks separate. This tech helps many GPUs work together, from thousands to hundreds of thousands. Fast GPU-to-GPU talk is needed for big AI jobs. NVIDIA’s networking tools make this easier.

• Spectrum-X Ethernet gives fast data speeds.

• Many GPUs can work together easily.

• GPU-to-GPU talk helps AI work better.

NVIDIA and Broadcom both help AI data centers. NVIDIA makes strong chips. Broadcom builds tools for moving data. Broadcom’s Tomahawk 6 chip gives lots of speed for big networks. Broadcom owns the pipes that move data, helping NVIDIA’s chips work well. When Broadcom’s tools connect with NVIDIA’s chips, AI centers work better.

Feature	Broadcom's Networking Hardware	NVIDIA's AI Compute Engines
Primary Focus	Ethernet solutions for fast and efficient connections	Powerful chips for AI jobs
Key Technology	Tomahawk 6-Davisson with Co-Packaged Optics (CPO)	NVLink/NVSwitch and InfiniBand/Spectrum-X
Efficiency	Uses less power, about 70% less	High cost and special technology
Integration with AI	Good for chips not made by NVIDIA	Works best with NVIDIA chips and CUDA software
Market Strategy	Made for big data centers and AI jobs	Made for training huge AI models

NVIDIA uses strong chips, and Broadcom gives fast connections. Together, they make AI data centers work well and grow bigger.

Broadcom vs NVIDIA: Technical Differences

Scalability and Performance

Broadcom and NVIDIA are both important in making chips. They use different ways to make their products work well and grow bigger. Broadcom’s networking tools, like switch silicon and network adapters, help big AI jobs in huge data centers. The Tomahawk switches can handle lots of data and work fast. Broadcom’s EQDS UDP-based transport protocol makes queues better and helps switches work well. Most data centers pick Broadcom’s switch silicon because it helps them grow easily.

NVIDIA works on strong AI training with powerful GPUs. It uses special networking like NVLink and InfiniBand. These connect well with NVIDIA’s CUDA software. NVIDIA is best at training AI models. Broadcom focuses on saving energy and growing for AI inference and moving data. Both companies have seen their stock prices go up as more people use AI.

Custom Silicon vs GPU Networking

Broadcom makes custom silicon for networking. The Thor Ultra chip connects many AI accelerators in big data centers. This helps jobs grow and cuts down wait times for AI work. NVIDIA uses GPU networking with NVLink and InfiniBand. These work best with NVIDIA’s own GPUs. Broadcom’s open and strong chips let data centers build networks that are flexible and efficient. NVIDIA’s plan is to keep everything together in one system.

Efficiency in AI Workloads

Efficiency is very important for AI data centers. Broadcom’s custom silicon, like Tomahawk 6-Davisson, saves a lot of power. The company checks things like Model FLOPS Utilization (MFU), throughput, time to first token, and token throughput to see how well things work. NVIDIA checks GPU use and throughput to get the most out of their tools.

Technology	Description
Co-Packaged Optics (CPO)	Combines optics and silicon, reducing interconnect power and boosting efficiency.
Thor2 Network Interface	Delivers 50% lower power use while supporting high packet rates, ideal for AI workloads.
Tomahawk 6 Switch	Integrates CPO, cutting optical interconnect power by about 70%.

Broadcom’s Thor2 Network Interface Card can handle 250 million packets each second. It uses half the power of older cards. New switching products are 40% more efficient and use 50% less power. Tomahawk 6 helps AI clusters with 102.4 Tbps speed and less energy. These new tools help data centers handle more AI jobs and keep costs and power low.

Market Impact and Investor Implications

Broadcom’s AI Hardware Revenue

Broadcom’s AI hardware business is growing very fast. In the first quarter of 2026, Broadcom made $8.4 billion from AI. This was 106% more than last year. More people wanted custom AI accelerators and better networking tools. Broadcom’s total revenue for the quarter was $19.31 billion. This was higher than what experts thought it would be. Broadcom thinks its AI chip sales will reach $10.7 billion next quarter. CEO Hock Tan said the company wants to make over $100 billion from AI chips by 2027.

Customer Adoption and Market Share

Big tech companies use Broadcom’s AI networking products. Some of these customers are Google, Meta, ByteDance, OpenAI, Amazon, Microsoft, Apple, and Cisco. They use Broadcom’s custom ASICs to make their AI chips and data centers better. Broadcom has over $10 billion in orders for AI racks with its XPUs. The company has about 55% to 60% of the high-end custom ASIC market. Broadcom is likely to stay as a top company.

Company	Focus Area	Market Share/Contracts
Broadcom	Open standards, long-duration contracts	Multi-billion dollar contracts with Google, Meta, ByteDance, OpenAI
NVIDIA	Full-stack platform, high-performance GPUs	Dominant in AI computing with a strong software ecosystem

Future Outlook for AI Networking

The AI networking market is expected to grow a lot. Experts think it will go from $8.4 billion in 2023 to $143.3 billion by 2033. This means it will grow by about 32.8% each year. New technology, more data, and better network management help this growth. Companies also want AI automation, cloud solutions, and stronger security. Broadcom’s Tomahawk and Jericho chips and new optical tech help the company do well. Investors believe Broadcom and NVIDIA will keep making AI networks better.

Broadcom is an important competitor to NVIDIA in AI networking, but not directly. Broadcom works in AI infrastructure and gets benefits from NVIDIA’s strong GPU market. Its custom ASICs and Ethernet solutions give people another choice instead of just using GPUs. This changes how companies compete in AI networking. Experts say both Broadcom and NVIDIA help the AI world grow. Each company is strong in its own area.

Technology	Data Throughput	Description
Standard Ethernet	60%	Traditional networking in AI training
Spectrum-X	95%	Enhanced networking for AI workloads

The competition between Broadcom and NVIDIA will help make new things in AI networking. This will bring more value to everyone in the industry.

 

 

 

 

 

Written by Jack Elliott from AIChipLink.

 

AIChipLink, one of the fastest-growing global independent electronic   components distributors in the world, offers millions of products from thousands of manufacturers, and many of our in-stock parts is available to ship same day.

 

We mainly source and distribute integrated circuit (IC) products of brands such as Broadcom, Microchip, Texas Instruments, Infineon, NXP, Analog Devices, Qualcomm, Intel, etc., which are widely used in communication & network, telecom, industrial control, new energy and automotive electronics. 

 

Empowered by AI, Linked to the Future. Get started on AIChipLink and submit your RFQ online today!