Broadcom is a US semiconductor and infrastructure-software group headquartered in Palo Alto, California. The company designs, develops and supplies a wide range of semiconductor and software products for data center, networking, software, broadband, wireless, storage and industrial markets.
AI relevance
Broadcom is central to AI infrastructure in two ways: first, it co-develops custom AI accelerators, so-called XPUs or ASICs, with large cloud operators. According to Wikipedia, Broadcom has contributed to Google's Tensor Processing Units (TPU) and Meta's Training and Inference Accelerator (MTIA), and has recently begun working with OpenAI on a custom inference chip. Second, Broadcom supplies the networking chips that carry traffic inside AI data centers. Its Tomahawk family of Ethernet switches reached 102.4 Tbit/s with Tomahawk 6, which reached the market in 2025 per trade outlet The Register. Broadcom also offers the Jericho4 Ethernet fabric router, which per Network World links more than one million AI accelerators across multiple sites, supports up to 36,000 HyperPorts at 3.2 Tbit/s each, is built on a 3-nanometer process, and complements the Tomahawk Ultra and Tomahawk 6 chip families.
Business model
Broadcom has grown through numerous acquisitions into a broadly diversified group that reports in two segments: Semiconductor Solutions and Infrastructure Software. In fiscal year 2025, according to the financial results announcement from December 2025, about US$36.86 billion, or 58 percent of group revenue, came from the semiconductor segment and about US$27.03 billion, or 42 percent, from the software segment, which grew 26 percent year over year. In semiconductors, Broadcom is also active in conventional networking, broadband, wireless and storage. With the 2023 acquisition of virtualization provider VMware, valued at roughly US$69 billion, Broadcom substantially expanded its software business around virtualization and data centers. This combination produces broadly diversified revenue.
Key figures
For fiscal year 2025, Wikipedia cites group revenue of roughly US$63.89 billion, up 24 percent year over year, and about 33,000 employees. In the same fiscal year, Broadcom spent roughly US$10.98 billion under GAAP on research and development. AI-related semiconductor revenue rose 74 percent year over year in the fourth quarter of 2025. For the first quarter of fiscal year 2026, CEO Hock Tan guided for AI semiconductor revenue to roughly double year over year to about US$8.2 billion. Broadcom first surpassed a market capitalization of US$1 trillion in December 2024; according to stockanalysis.com, its market cap stood at roughly US$1.87 trillion on July 13, 2026. The shares trade on Nasdaq under the ticker AVGO.
Market position
In AI accelerators, Nvidia's conventional graphics processors remain dominant. According to InvestorPlace, GPUs currently account for roughly 90 to 95 percent of AI accelerator revenue, with custom XPUs such as Broadcom's co-developed chips making up only about 10 to 20 percent, a share that could rise to 25 to 30 percent by 2030. In custom AI silicon and networking chips, Marvell Technology is Broadcom's main competitor. In June 2026 Marvell introduced its Teralynx T100 switch chip, also rated at 102.4 Tbit/s, to rival Tomahawk 6; Broadcom, per The Register, had already been shipping since 2025. Cisco, too, already has comparable AI networking solutions in production, per The Register. The business depends heavily on a small number of very large cloud customers.
This profile is a neutral description and is not investment advice.