The Lamp restaurant in San Francisco on the 24th (local time). At a dinner arranged during President Lee Jae-myung's overseas trip, heads of Korean conglomerates and Big Tech leaders gathered in full force. During the roughly one-hour dinner, President Lee and Nvidia CEO Jensen Huang, seated across from him, led the conversation, while Samsung Electronics Chairman Jay Y. Lee, SK Chairman Chey Tae-won, Hyundai Motor Group Chairman Euisun Chung, Naver board chairman Lee Hae-jin, Deputy Prime Minister and Minister of Science and ICT Bae Kyung-hoon, and Rani Borkar, Microsoft's corporate vice president for Azure Hardware, chimed in at intervals.
At this gathering, there was a heavyweight who quietly ate a hamburger while listening intently to the conversation between President Lee and CEO Huang. That was Broadcom CEO Hock Tan. Both Jensen Huang and Hock Tan, Asian-American Big Tech leaders, sat side by side across from President Lee, but their styles were starkly different. As attendees noted that "Jensen did most of the talking with the president," Huang led the atmosphere, while Tan wore an almost expressionless face throughout. When the dinner ended, Huang exchanged greetings with the Korean conglomerate heads and government officials and waited until President Lee departed, but Tan was the first to leave his seat.
Unlike the Nvidia chief, familiar in Korea for his "gganbu meeting," Hock Tan is not well known in Korea, but at the AI Summit his standing was no less than Jensen Huang's. When taking a commemorative photo holding beer bottles before the meal began, Huang was to President Lee's right and Tan to his left. At the San Francisco AI Summit declaration stage held just before the dinner, the two men were seated on either side of President Lee. When entering the dinner venue, Huang came in separately with company officials, including his daughter Madison, while Tan entered together with the Korean conglomerate heads.
While attention was focused on the rivalry between Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman, who attended the AI Summit, a subtle tension also ran between Jensen Huang and Hock Tan. Nvidia leads at the top with a market capitalization of $5 trillion (7,315.5 trillion won), but Broadcom, at $1.82 trillion, is valued higher than AMD ($851.1 billion), considered Nvidia's biggest rival. Broadcom ranks right behind Apple, Google, Microsoft and Amazon, and counting only AI chip companies, Nvidia and Broadcom rank first and second. As SK Telecom (017670) agreed to be supplied next-generation graphics processing units (GPUs) from Nvidia to build an AI factory of up to 2 gigawatts (GW), Broadcom signed a memorandum of understanding with Samsung Electronics (005930) for a five-year, $200 billion (290 trillion won) supply of advanced memory semiconductors and foundry (contract chip manufacturing) cooperation. At the dinner venue, it was hard to see the two men in conversation.
The history behind today's Broadcom is somewhat complex. By name alone, it started from Broadcom, a U.S. fabless company founded in 1991, but its actual roots lie in the semiconductor product division that Hewlett-Packard (HP) operated from 1961. This organization continued for nearly 40 years before being absorbed into Agilent, established in 1999 when HP's semiconductor business was spun off. Kohlberg Kravis Roberts (KKR), the world's largest private equity fund, and Silver Lake Partners acquired Agilent for $2.66 billion in late 2005, launching Avago Technologies, headquartered in Singapore. Avago was listed on the Nasdaq in 2009 under the ticker AVGO. Avago acquired Broadcom for $37 billion in 2015 and changed its name to Broadcom, the acquired company. At the time, Avago's CEO was Hock Tan, who has been in charge of management for 20 years, first at Avago from 2006 and then at Broadcom.
While Jensen Huang is of Taiwanese descent, Hock Tan is of Malaysian descent. Born in Malaysia in 1953, he left for study at the Massachusetts Institute of Technology (MIT) in the United States at age 18 in 1971, and after passing through several tech companies, became a Big Tech heavyweight leading the U.S. artificial intelligence (AI) market. The growth engine of Broadcom, led by Hock Tan, was an aggressive merger-and-acquisition strategy. It acquired system semiconductor specialist LSI for $6.6 billion in 2014, Brocade, which makes network equipment such as routers and switches, for $5.9 billion in 2017, and virtualization platform company VMware for $69 billion in 2023. Avago, once at about $2.7 billion, became Broadcom and surpassed $1 trillion in market cap in 2024.
The time when Hock Tan drew particular attention was 2017. In the early period of the first Donald Trump administration, Broadcom sought to swallow Qualcomm for $117 billion, but the deal fell through by executive order of President Trump. This was due to concerns that, amid intense U.S.-China conflict, U.S. technology passing to a Chinese-affiliated company would pose a threat to national security. Then, a few months later, in November 2017, CEO Tan declared at the White House with President Trump that he would bring Broadcom's headquarters back to the United States. He decided to move the base in line with the Trump administration's stance seeking to relocate overseas headquarters to the U.S. and expand jobs. At the time of the announcement, CEO Tan drew a smile from President Trump by saying, "My mother would never have imagined that I would stand next to the U.S. president at the White House," and President Trump, watching the announcement from behind, responded, "My mother, too (would never have imagined that I would become president and enter the White House)," a scene that became a major talking point.
For Broadcom, which had steadily expanded its business, the AI boom was an enormous opportunity. Until then, Broadcom's mainstay had been its network business, but as AI model development and data center expansion began in earnest, Broadcom's chip design capability began to shine. Just as Britain's Arm dominated mobile chip design, Broadcom began to lead the application-specific integrated circuit (ASIC) market on the strength of chip development know-how accumulated since its HP days. As the AI market shifted from a training focus to an agentic AI focus, where inference is important, starting last year, demand for ASICs specialized in inference surged, and "love calls" poured in to Broadcom and rival Marvell, which design ASICs, to jointly develop custom chips.
While Nvidia dominates AI data centers with general-purpose GPUs, Broadcom serves as an ally for companies seeking to build their own chips to counter Nvidia. Broadcom takes on chip design in response to customer requests to help implement chips tailored precisely to their AI models. It oversees everything from silicon manufacturing, power and thermal management, and memory integration to final foundry production preparation. Because every process requires specialized technical expertise, it is difficult to entrust the work to any company that lacks more than 60 years of know-how like Broadcom.
Recently, the ASIC chip market cannot be discussed without Broadcom. Google, the strongest player in ASIC chips and at the center of the Nvidia exodus in the AI chip market, is a major Broadcom partner. Broadcom is participating in the development of Google's ASIC, the Tensor Processing Unit (TPU). This April, the two companies agreed to extend their TPU design partnership until as late as 2031.
It is not just Google. OpenAI announced last October that it would cooperate with Broadcom to develop a custom chip, and this June OpenAI's first inference-only chip, "Jalapeno," was unveiled. It usually takes more than two years from design until the blueprints are handed over to a foundry, but the time from the cooperation announcement last year to the chip's unveiling was just nine months. Broadcom, in effect, designed a chip precisely fitting OpenAI's request to make a chip specialized for its data centers. OpenAI's rival Anthropic also signed a contract this April with Google and Broadcom to secure GW-scale TPU capacity. This means OpenAI, Anthropic and Google—which lead the global AI model market with ChatGPT, Claude and Gemini—all rely on Broadcom.
The same is true for Meta, which is chasing the AI Big Three. Meta, which released its first custom chip in 2018, unveiled four products in the "Meta Training and Inference Accelerator (MTIA)" family this March—the MTIA 300, 400, 450 and 500. MTIA
