Broadcom, a major network chip manufacturer that supplies customized ASICs to hyperscalers, has soared in share price as the latest AI model "Gemini 3" of Google parent company Alphabet has received great reviews. CNBC reported on the 24th...
Broadcom, a major network chip manufacturer that supplies customized ASICs to hyperscalers, has soared in share price as the latest AI model "Gemini 3" of Google parent company Alphabet has received great reviews.
CNBC reported on the 24th that Google is a major customer of Broadcom’s ASIC business. Broadcom assists Google in designing the "Tensor Processing Units" (TPUs) it develops, which are mainly used for internal AI infrastructure and are regarded as competitors of NVIDIA (Nvidia) graphics processors (GPUs).
Melius Research analyst Ben Reitzes reiterated Broadcom's investment rating as "buy" on October 27, and raised the target price by US$60 to US$475.
Reitzes stated through the report that Google and its partner Broadcom have been working together to develop customized ASICs since 2016 and are now in the seventh generation. For AI workloads, in addition to NVIDIA GPU, TPU can be said to be the ASIC that has gained the most empirical evidence, and now has the strongest momentum.
Reitzes believes that although Alphabet has a promising future, Broadcom may benefit more, because in addition to Alphabet will contribute a large amount of AI revenue, many other partners are also interested in Broadcom's design expertise and want a share of the pie. He added that TPU is quickly becoming more important to Alphabet's growth strategy.
Broadcom surged 11.1% to close at US$377.96 on the 24th, setting a new closing high since October 29 and the largest single-day increase since April 9. Broadcom was also the best-performing component stock of the SPDR Technology Stock ETF (Technology Select Sector SPDR Fund, code XLK) that day. Year to date, Broadcom has soared 63.03%.
D.A. Davidson analyst Gil Luria published a research report on September 2 stating that some cutting-edge artificial intelligence laboratories (such as xAI) seem to have a strong interest in TPU, but only if Google is willing to sell these chips externally. Google TPU is an ASIC designed specifically for machine learning tasks and is available through the Google Cloud platform.
Luria stated through a report that Google TPU is still the best alternative to NVIDIA, and the gap between the two has narrowed significantly in the past 9 to 12 months. During this period, the market's evaluation of TPU became increasingly positive. The performance of Google's next-generation TPU has not only been expanded to 42.5 Exaflops, the high-bandwidth memory (HBM) capacity has jumped, and the cost-effectiveness has also been greatly improved. This is the main reason for attracting the attention of cutting-edge laboratories.
Further reading: Why NVIDIA and Broadcom are competing for CPO dominance? Comprehensive analysis from the perspective of speed, power consumption and reliability Loyal customers bite back on NVIDIA? Google launches its own TPU to enter the small cloud market AI chip transmission to CPO card position battle: What exactly are NVIDIA and Broadcom competing for? (Down)
