With spending for artificial intelligence (AI) infrastructure through the roof this year and only expected to grow in the coming years, Nvidia remains well-positioned. The company's graphics processing units (GPUs) are the main chips used to power AI workloads, and it has a more than 90% market share in the GPU space.
The company has carved out a nice moat, particularly with training large language models (LLMs), due to its CUDA software program. Nvidia seeded CUDA into universities and research labs that were doing early work on AI, which has led to a generation of developers being trained on its platform and virtually all foundational AI code being written on its software.
As the AI data center buildout continues, this helps ensure that Nvidia will be a nice winner. However, it's not the stock I'd be buying to play this spending boom. Instead, I think Broadcom (AVGO 1.12%) is the better buy.
NASDAQ: AVGO
Key Data Points
A networking and ASIC leader
With Broadcom, investors are getting exposure to two of the fastest-growing parts of AI infrastructure buildout: Networking and custom AI chips. Broadcom is a data center networking leader, where its Tomahawk Ethernet switches are the go-to option for large-scale data centers.
It also offers high-performance fiber optical interconnect solutions, digital signal processors (DSPs), and network interface cards (NICs), all of which help manage data flow and distribute AI workloads across servers. As AI chip clusters keep growing in size and complexity, networking becomes increasingly important. While not as large a market as the one for computing power, it should grow more quickly.
Broadcom's even bigger opportunity, meanwhile, is in custom chips. The company is a leader in ASIC (application-specific integrated circuits) technology, where it provides the building blocks to help customers turn their chip designs into custom chips that can be manufactured at scale. While AI ASICs lack the flexibility of GPUs, they tend to be very good at the tasks for which they've been developed. They're also more energy-efficient, which can help save on costs.
Image source: Getty Images.
Broadcom helped Alphabet create its highly successful Tensor Processing Units (TPUs), which should be a nice growth driver given the cloud computing leader's massive capex budget this year. Alphabet has also started letting customers use its TPUs through Google Cloud, opening up another opportunity. In fact, Anthropic has already placed a $21 billion TPU order through Broadcom to be delivered this year.
It's also leading to other companies working with Broadcom to develop their own custom chips. One of its biggest opportunities is with OpenAI, which has committed to deploying 10 gigawatts of custom chips. With each gigawatt of power equating to about $35 billion in chip costs (based on Nvidia GPU pricing), this is a huge opportunity.
With Broadcom generating slightly under $64 billion in total revenue last fiscal year, the company is set to see explosive growth in the coming years, making it a top AI infrastructure stock to own.
