摘要：TMTPOST -- In a dramatic shift in the tech sector, Broadcom, a leading U.S. AI network chip manufacturer, has emerged as Wall Stre

TMTPOST -- In a dramatic shift in the tech sector, Broadcom, a leading U.S. AI network chip manufacturer, has emerged as Wall Street’s new favorite. While competitors like AMD and others continue their uphill battle in the GPU race against NVIDIA, Broadcom has carved out its niche by focusing on custom AI chips.

Broadcom’s strategic move has not only attracted market attention but also sent its stock price soaring by 38% over two consecutive trading sessions, propelling its market capitalization above $1 trillion.

Broadcom’s meteoric rise positions it as the ninth global company by market capitalization and the third semiconductor firm to surpass the trillion-dollar mark. Meanwhile, NVIDIA—long regarded as the leader in AI technology—has seen its stock decline 1.7% to $132 per share on Monday. This marks an 11% fall from its November high, signaling a shift in market sentiment.

Hock Tan, the CEO of Broadcom, attributes the company’s success to its focus on custom ASIC (Application-Specific Integrated Circuit) AI chips.

Tan has recently revealed that Broadcom is collaborating with three major clients to develop these chips and estimated that the demand for custom AI chips could reach $60 billion to $90 billion by 2027. This demand is fueled by tech giants like Google, Amazon, Meta, and Apple, all of whom are seeking alternatives to NVIDIA’s GPUs to reduce dependency and optimize performance.

One of the most notable developments in this space is Apple’s reported partnership with Broadcom to develop AI server chips, underscoring the industry’s growing interest in customized solutions.

Unlike NVIDIA and Intel, which dominate the computing chip market, Broadcom has traditionally specialized in networking chips, earning its reputation as the world’s largest network chip manufacturer. Founded in 1991 by Henry Samueli and Henry T. Nicholas III, Broadcom initially focused on developing chipset solutions for network devices. Its breakthrough came with modem chips that provided faster internet services to cable TV systems, setting the stage for its rise in the communications industry.

Broadcom went public in 1998, riding the dot-com wave to financial success. In 2015, it was acquired by Avago Technologies for $37 billion, a move that transformed the company. Under Tan’s leadership, Broadcom shifted its strategy to focus on niche markets and create “local monopolies” by dominating specific verticals with cutting-edge solutions.

Today, Broadcom's product portfolio includes Ethernet switches, routers, wireless communication chips, Bluetooth, and Wi-Fi solutions. Its chips play a critical role in nearly all internet connections worldwide, giving the company a strategic edge as global demand for connectivity and data processing continues to grow.

Broadcom’s leap into the AI space has been significantly bolstered by its custom ASIC chips. These chips, designed for specific applications, offer advantages in performance, power efficiency, and cost. Unlike general-purpose GPUs or CPUs, ASICs are tailored to meet the unique requirements of tasks, such as data packet processing in routers or signal modulation in communication systems.

For AI applications, ASICs can deliver substantial improvements in computational efficiency, making them an attractive choice for companies seeking to optimize performance and reduce costs. For instance, Google’s Tensor Processing Units (TPUs), developed in collaboration with Broadcom, exemplify how ASICs can outperform traditional GPUs in specialized AI workloads.

Market forecasts show the growing importance of ASICs. Morgan Stanley projects the AI ASIC market to grow from $12 billion in 2024 to $30 billion by 2027, with a compound annual growth rate of 34%. Broadcom, alongside other players like TSMC and Cadence, is well-positioned to capitalize on this trend.

Broadcom’s latest financial results have demonstrated its strong performance. In its fourth fiscal quarter, the company reported revenue of $14.05 billion, a 51% year-over-year increase, and earnings per share (EPS) of $1.42, surpassing market expectations. Net profit under GAAP was $4.32 billion, while non-GAAP net profit reached $6.97 billion.

For the 2024 fiscal year, Broadcom achieved record revenue of $51.6 billion, a 44% increase from the previous year. Much of this growth was driven by its semiconductor solutions and infrastructure software businesses. The latter saw a staggering 196% year-over-year increase, contributing $21.5 billion to the company’s total revenue.

Broadcom’s AI-related revenue alone grew by 220% year-over-year to $12.2 billion, fueled by the surging demand for generative AI and the computational power required for training large models. As AI continues to reshape industries, the demand for high-performance chips is expected to grow, providing Broadcom with ample opportunities for future expansion.

Despite Broadcom’s impressive growth, analysts caution that ASICs may not immediately dethrone GPUs as the dominant AI hardware. NVIDIA’s GPUs remain the gold standard for AI training due to their unparalleled performance and versatility. However, as cloud providers and enterprises seek cost-effective alternatives, ASICs are becoming an increasingly attractive option.

Seasoned industry experts have noted that custom AI chips like ASICs can enhance inference efficiency and reduce operational costs for AI models. However, breaking NVIDIA’s stronghold on the market will require sustained innovation and execution.

Broadcom’s CEO remains optimistic about the future. If his projections for the ASIC market hold true, the company’s AI-related revenue could double annually over the next three years, signaling a potential inflection point for the semiconductor industry.

Broadcom’s recent performance has prompted bullish responses from market analysts. Goldman Sachs upgraded its 12-month price target for Broadcom to $240, citing new client acquisitions and the successful integration of VMware following its $61 billion acquisition. Barclays also raised its price target to $205, emphasizing Broadcom’s strong execution and growth potential.

However, some analysts note the risks associated with trend-driven investments. While Broadcom’s current trajectory is promising, the market’s focus on high-growth stocks could shift if better-performing alternatives emerge. Nonetheless, Broadcom’s robust fundamentals and strategic positioning make it a strong contender in the AI chip market.

Ji Yu, the founder of Chinese GPU chip company Xingyun Integrated Circuit, told AsianFin that today's NVIDIA products increasingly resemble AI mainframes. The so-called AI mainframe refers to the impact of the expensive GPU servers used today in the industry. By utilizing custom AI chips and delivering solutions in a CPU+GPU format, the inference efficiency of large AI models can be improved while cutting costs.

However, Ji also acknowledged that NVIDIA's GPUs are currently the highest-performing products in the market, making it difficult for competitors to challenge its dominant position with similar offerings. But whether for cost control needs or to enhance their own bargaining power, customers are motivated to find alternatives to reduce their reliance on NVIDIA.

来源：钛媒体