I'm LongbridgeAI, I can summarize articles.The article argues that while Nvidia remains a strong play for AI training, investors missing Broadcom (AVGO) are overlooking the critical AI inference market. Broadcom develops custom ASICs preferred by major tech firms like Meta and Google for cost-efficient inference tasks. With AI chip sales projected to surge from $20 billion to $115 billion by fiscal 2027, Broadcom offers faster growth than Nvidia at a cheaper valuation, making it a essential complement for comprehensive AI exposure.
To gain some exposure to the booming AI market, many investors turn to Nvidia (NVDA +0.84%), the world's largest producer of discrete GPUs for data centers. That's still a smart move, since Nvidia sells the best picks and shovels for training AI algorithms.
Most of the world's top AI companies use Nvidia's data center GPUs, and it locks in those customers with its proprietary software and services. From fiscal 2026 (which ended this January) to fiscal 2029, analysts expect its revenue and EPS to both grow at CAGRs of 59%. Those are remarkable growth rates for a stock that trades at 23 times next year's earnings.
Image source: Getty Images.
So if you're bullish on the AI market's long-term growth, it's still a good idea to buy Nvidia. But if Nvidia is your only play on the AI market, then you're missing out on the expansion of the AI inference market. Let's see why the inference market is becoming just as important as Nvidia's training market -- and why Broadcom (AVGO +0.21%) is a great play on that paradigm shift.
NASDAQ: AVGO
Key Data Points
AI training vs. AI inference
Nvidia's data center GPUs are used to train large language models (LLMs). But that data is useless without inference, or the computing process that allows AI applications to access all that information. To help hyperscalers bridge that gap, Broadcom (AVGO +0.21%) develops application-specific integrated circuits (ASICs) that can be customized for inference tasks.
At scale, Broadcom's AI accelerators can process inference tasks faster and more cost-efficiently than Nvidia's stand-alone GPUs. That's why leading AI companies -- including Meta, Alphabet's Google, OpenAI, and Anthropic -- all use Broadcom's chips to accelerate their AI applications.
Why Broadcom is another must-own AI stock
In fiscal 2025 (which ended last November), Broadcom's sales of AI chips soared 65% to $20 billion and accounted for 31% of its top line. By fiscal 2027, it expects its AI chip sales to rise nearly sixfold to $115 billion -- about two-thirds of its projected $173.5 billion in revenue.
From fiscal 2025 to fiscal 2028, analysts expect Broadcom's revenue and EPS to grow at CAGRs of 62% and 77%, respectively. Yet it trades at just 22 times next year's earnings. Therefore, Broadcom is growing faster than Nvidia but still looks fundamentally cheaper.
Nvidia has been integrating more inference features into its latest GPUs, but it will remain more closely associated with the AI training market for the foreseeable future. So if you want more exposure to the growing market for custom inference chips, you should also invest in Broadcom.