Broadcom sees AI chip revenue hitting $56 billion this year

For years, AI data centers relied almost exclusively on GPUs for AI processing, but as tech companies deploy more advanced models and AI agents become more common, custom processors have grown more important because they can be fine-tuned to run specific AI models faster and more efficiently than general GPUs for some tasks. Alphabet, Microsoft, Amazon and others are spending hundreds of billions on this expanding market.

Alphabet's Google Cloud business, which includes cloud computing and Gemini Enterprise subscriptions, saw sales soar 82% in the second quarter to $24.8 billion, making it Alphabet's fastest-growing segment. Alphabet leans on its own Tensor Processing Units, and a few months ago signed a multiyear deal with Broadcom for custom processors through 2031. Marvell also designs custom chips for Amazon's Trainium and Microsoft's Maia processors, and Microsoft's Azure just surpassed $100 billion in annual revenue for the first time. Microsoft and Amazon are the two leading cloud platforms.

While Nvidia benefits from including Marvell's custom chips in its network offerings, the rise of custom AI processors could eventually erode Nvidia's market share. Intel's foundry business remains uncertain — it lost $2.1 billion in the second quarter and just $293 million of its sales came from outside customers, and analysts estimate its sales will rise by just 15% next year despite its stock soaring 188% over the past year. TSMC manufactures about 73% of the world's processors and an estimated 90% of the most advanced ones, so it benefits regardless of which chipmakers win. TSMC's second-quarter sales rose 36% to over $40 billion, with non-GAAP diluted earnings up 77% to $4.31 per ADR. Of the three custom AI processor plays mentioned, Broadcom could have the most potential, as its sales and earnings are growing faster, and its stock has a forward price-to-earnings ratio of just 19.