AI Chip Is Facing An Inflection Point, So Is Broadcom?

Since October 2024, investors have been reallocating funds from semiconductor stocks, with Broadcom underperforming its tech sector peers. While NVIDIA has remained resilient, peers with significant non-AI exposure, such as Broadcom, face more skepticism about their recovery prospects.

Broadcom is set to report its fourth-quarter earnings on December 12th. Following a notably disappointing quarter, investors are expected to scrutinize the semiconductor leader's AI revenue prospects.

Broadcom's CEO, Hock Tan, expressed his optimism at a conference in September 2024, highlighting the surge in AI infrastructure capital expenditures by leading hyperscale providers (Microsoft, Amazon, Google). This momentum has been observed to continue, as seen in the recent third-quarter earnings season. Consequently, hyperscale providers have bolstered confidence in aggressive investment to maintain a competitive edge in the rapidly evolving AI race. Morgan Stanley expressed its confidence that hyperscale providers' AI capital expenditures could soar to $300 billion by 2025. NVIDIA's recent earnings commentary emphasized the frenzy demand behind the shift from Hopper architecture to Blackwell AI chips.

Thus, the question arises whether Broadcom's AI revenue forecast for the next year could significantly exceed the $12 billion expected for the fiscal year 2024. As large-scale AI clusters (>100K GPUs) take center stage, they are increasingly becoming the bet in the race for AI dominance. Therefore, it is not misleading to see AI clusters in the hundreds of thousands of GPUs being built. However, AI companies and hyperscale enterprises are increasingly facing pressures to address the challenges of AI capabilities and scaling costs. It has been reported that despite AI companies' eagerness to improve their cutting-edge LLMs, the pace of AI scaling is slowing down, leading to increased focus on custom AI chips.

As AI companies and hyperscale enterprises reassess their financial capacity to invest in emerging AI clusters, Broadcom's expertise in custom AI chips is expected to lead the next phase of growth. Given the scale required for designing and producing custom AI chips, it is anticipated that the demand for custom chips will significantly increase, with Broadcom's expertise and experience playing a crucial role.

In September, Chen stated that custom chips account for two-thirds of its AI revenue. Moreover, Broadcom has established solid partnerships with three hyperscale enterprises. ByteDance's geopolitical headwinds are expected to prompt the world's most valuable startup to collaborate with Broadcom in developing custom chips, as obtaining NVIDIA's leading AI chips becomes increasingly complex. Additionally, the expansion of OpenAI makes the collaboration with Broadcom for custom chip development a viable business, potentially enhancing its cost efficiency.

Therefore, it is not unreasonable to believe that Broadcom sees custom chips potentially taking a more prominent role in hyper-scale workloads over the next five years. AWS's experience with its customers using its custom AI chips has proven the feasibility and sustainability of the custom chip business, as hyperscale enterprises compete for opportunities to improve cost competitiveness.

Nevertheless, NVIDIA's ability to expand its network business to align with its commercial AI chip business is noteworthy. As NVIDIA CEO Jensen Huang pointed out, the company's earnings conference utilized NVIDIA's Spectrum-X networking technology for xAI's Colossus 100K Hopper supercomputer. Huang also emphasized the increasing competitiveness of Spectrum-X against more "traditional" Ethernet technology.

Although Broadcom is a leader in the Ethernet market, it may face more significant challenges as NVIDIA seeks to develop more cost-effective bundled products to support the adoption of its networking products. Nevertheless, Broadcom remains confident that it can maintain industry standards as it demonstrates its scalability and energy efficiency. Therefore, the competition between NVIDIA and Broadcom is expected to remain exceptionally fierce, although Broadcom's diversified revenue departments should mitigate unexpected weaknesses from its emerging but growing AI business.

Broadcom's valuation relative to its tech sector peers is not low (D- valuation grade). Thus, given its B- growth rating, it is clear that Broadcom is expected to continue performing well, highlighting the market's high expectations for the company. In other words, not-so-strong future performance guidance could dampen buying sentiment for the stock, affecting its buy-on-dip momentum.

 

Therefore, despite increasing caution about the sustainability of the surge in AI infrastructure investment, Broadcom must continue to excel in its AI narrative. Moreover, recent weakness in the non-AI revenue departments of the semiconductor industry could undermine Broadcom's bullish argument.

Wall Street remains optimistic about Broadcom's valuation, supporting its bullish inclination. However, the expected $12 billion in AI revenue for 2024 is expected to account for less than 25% of its total revenue base that year. Therefore, investors must cautiously consider Broadcom's AI growth narrative as the company navigates short-term headwinds from other business departments.