Nvidia has been the dominant force in artificial intelligence infrastructure, with its graphics processing units serving as the primary chips for AI workloads. The company's CUDA software platform and NVLink interconnect system have created a strong ecosystem around its products.
Nvidia reported $57 billion in revenue last quarter, making it the world's largest company. However, customers are now seeking cheaper alternatives, particularly for inference tasks where cost-per-inference becomes a more important factor.
Broadcom is emerging as a key competitor in this space. The company is helping create custom AI ASICs, which are preprogrammed chips tailored for specific tasks. While less flexible than GPUs, these chips offer better performance for their designated functions and greater energy efficiency.
Alphabet's tensor processing units have gained momentum and are being offered to cloud computing customers. Broadcom played a significant role in turning Alphabet's chip designs into physical products and providing manufacturing support.
Anthropic recently placed a $21 billion order with Broadcom for Alphabet's TPUs to be deployed with Google Cloud. Other companies, including OpenAI, have turned to Broadcom for assistance with their own custom ASICs.
Citigroup analysts project Broadcom's AI revenue will climb from just over $20 billion last fiscal year to more than $50 billion this fiscal year, reaching $100 billion by fiscal 2027. These estimates do not include potential work with Apple, which Broadcom has reportedly been assisting with AI ASIC design.
Taiwan Semiconductor Manufacturing benefits from this competition regardless of which chip type dominates. The company manufactures both GPUs and ASICs, positioning it to gain from growth in either sector.
TSMC has projected mid-40% compound annual growth for AI chips over the next few years. The company is increasing capacity to meet this demand while maintaining its position as the only foundry proven to manufacture advanced chips at scale with few defects.
Broadcom's relationship with TSMC provides access to advanced packaging techniques and capacity, which is particularly valuable given current tightness in semiconductor fab capacity for advanced chips.