Broadcom’s ‘Thor Ultra’ Chip Takes Aim at Nvidia — The Next Big Battle in AI Networking Begins
Broadcom has just pulled the curtain back on its latest AI-focused networking chip — the Thor Ultra, a high-performance processor designed to connect hundreds of thousands of AI chips together in sprawling data centers. The launch marks not just another product debut, but the beginning of a new chapter in the Broadcom vs. Nvidia rivalry — one that could reshape the infrastructure behind artificial intelligence.
This new chip is engineered for the era of massive AI models, like those powering ChatGPT and Google’s advanced machine learning systems. By enabling faster, denser networking between GPUs, Thor Ultra gives data centers the ability to scale AI workloads far beyond current limits.
A Strategic Move Against Nvidia
The timing couldn’t be more pointed. Just a day after announcing its 10-gigawatt custom chip partnership with OpenAI, Broadcom introduced the Thor Ultra — signaling a direct challenge to Nvidia’s leadership in AI networking.
While Nvidia’s networking hardware dominates many AI clusters today, Broadcom’s advantage lies in scale. The company’s chips already serve as the “invisible highways” of the internet, routing data across the world’s largest cloud platforms. Now, it wants to do the same inside AI supercomputing clusters, moving information between GPUs with even higher efficiency.
Executives say the Thor Ultra doubles the bandwidth of its predecessor, offering the backbone required for next-generation AI training systems.
The $90 Billion Opportunity
Broadcom CEO Hock Tan has made it clear: the company’s AI ambitions go far beyond one product line. He estimates that Broadcom’s total addressable AI market could reach $60–90 billion by 2027, spanning both networking chips and custom processors built for partners like Google and OpenAI.
In fiscal 2024, Broadcom’s AI revenue already reached $12.2 billion, and its newest chip represents a direct play to capture the expanding AI infrastructure boom — from data movement to model deployment.
Aoso Read: Windows 10 Support Ends — What Happens Next Might Surprise Millions of PC Users.
Why Networking Is the New AI Battleground
While GPUs power the computation, networking chips like Thor Ultra make large-scale AI possible. They’re the critical connectors that allow thousands of GPUs to act as one super-brain.
Broadcom Senior VP Ram Velaga described it best:
“In distributed computing, the network is what builds these large clusters. Anyone in the GPU business needs to be in networking too.”
The Thor Ultra will serve as that connective tissue — handling the immense data flow within AI data centers and ensuring minimal latency even at planetary scale.
Also Read: Sora 2 Is Coming to Android — And OpenAI’s Most Viral App Yet Faces a New Test.
Inside Broadcom’s AI Design Philosophy
Unlike Nvidia, which sells complete systems, Broadcom focuses on silicon design and integration for its cloud partners. The company’s engineers have worked closely with Google on multiple generations of Tensor Processing Units (TPUs), generating billions in revenue.
Each chip, Broadcom says, is developed alongside a reference system — an internal model that customers can use to build their own AI networking infrastructure. For every dollar Broadcom invests in its silicon, ecosystem partners invest up to ten, creating a multiplier effect that keeps Broadcom at the center of AI’s physical buildout.
The release of Thor Ultra solidifies Broadcom as one of the few companies truly competing with Nvidia in the AI hardware stack — not through graphics power, but through the architecture of connection.
In the AI gold rush, compute gets the headlines — but networking decides who wins the race.
Broadcom’s bet is simple: in a world where every tech company wants to train the next frontier model, the companies that move data fastest will lead the future.
And with Thor Ultra, Broadcom just turned up the speed limit.
Also Read: Instagram Just Went PG-13 — Inside Meta’s Biggest Shift to Protect Teen Users.
