Artificial intelligence (AI) has been a dominant investment theme on Wall Street in the past couple of years. Considering the rapid and broad usage of these technologies and their transformational nature, many research firms have come out with estimates about the AI market opportunity. Mckinsey expects AI to add $13.6 trillion to $22.1 trillion in economic value to current global GDP by 2030, while PWC expects AI to increase economic activity by $15.1 trillion by 2030. Although there is significant variation in these estimates, most firms agree that the impact of AI on the global economy will be sizeable.
Microsoft CEO’s statement can adversely affect Nvidia
Nvidia (NVDA 4.45%) has already been one of the major beneficiaries of this fast-evolving trend, driven by extensive adoption of its AI-optimized chips.
With the supply of these chips falling short of demand, the company has also benefitted from significant pricing power. However, this dynamic seems on the verge of changing, as is evident by a statement from Microsoft’s (MSFT 1.14%) CEO Satya Nadella.
In an interview, CEO Satya Nadella claimed "Power yes...I am not chip supply constrained." This statement seems to hint at the possibility that the company has already built a large inventory of Nvidia chips or found alternatives in terms of competing or custom-made AI chips. In either case, this does not seem to be positive news for Nvidia shareholders.
However, there is one high-flying stock that can benefit from these developments.
Broadcom’s networking solutions may see growing demand
Looking at custom AI chip manufacturer and advanced networking solutions provider Broadcom (AVGO 0.25%), there is a lot to be excited about. A leading custom AI accelerators and AI-optimized networking solutions provider, the company posted a record revenue of $51.6 billion in fiscal 2024, implying a year-over-year jump of 44%. AI revenues surged at an even faster pace, of 220% year-over-year to $12.2 billion in fiscal 2024.
Hyperscalers require high-bandwidth and low-latency networking components to build large AI clusters for training and inferencing complex large language models. Broadcom’s three major hyperscaler clients are planning to scale the size of their AI clusters from 500,000 to 1 million XPUs (accelerators). Larger AI clusters will require higher silicon content for networking (Ethernet switches, high-speed interconnects). Furthermore, the networking specialist also expects resources expended on networking content to rise from a range of 5% to 10% to a range of 15% to 20% of computational content for increased AI connectivity. Hence, Broadcom's networking expertise is already being sought after extensively by many clients in large-scale AI deployments.
Against this backdrop, since Broadcom is a major networking player enabling the global AI infrastructure buildout, Microsoft may start using its networking components to expand its data center capacity. With Broadcom's expertise in scaling up and scaling out (interconnection inside data racks and amongst data racks), it can become a preferred partner for even more clients building large AI clusters.
Broadcom’s collaboration with OpenAI
Microsoft partner and ChatGPT developer OpenAI has collaborated with Broadcom and Taiwan Semiconductor Manufacturing to build first custom inference AI chip, which will help deploy advanced AI models in production environment. This, in turn, is expected to diversify OpenAI’s chip supply and optimise costs.
As Microsoft focuses on custom chip development (including in-house-designed Maia accelerators), the company may also seek out Broadcom’s expertise in specialized ASICs (application-specific integrated circuits). Broadcom has already demonstrated its technological prowess by developing cutting-edge XPUs with 3-nanometer process technology.
Unlike Nvidia which specializes in general-purpose GPUs, Broadcom works with its hyperscaler clients to develop customized solutions catering to their specific needs. Microsoft may prefer to work with Broadcom in long-term engagement to develop a multi-generational roadmap for their AI infrastructure requirements.
Furthermore, Broadcom may also see heightened demand for its AI-optimized server storage solutions, as Microsoft ramps up its data center capacity to meet increased demand. This may open up new growth opportunities for Broadcom in the coming years.
Conclusion
Broadcom is currently trading at about 21.1 times trailing -12-month sales, significantly higher than its historical three-year average price-to-sales (P/S) multiple of 13.04. However, considering the company’s robust growth potential, this rich valuation seems justified. Plus, Microsoft’s chip sufficiency may also prove to be a major tailwind for Broadcom in the coming years.
