Broadcom targets $100 billion in AI chip sales by 2027

When the head of one of the world's largest semiconductor companies publicly names a figure of $100 billion, the market holds its breath. This is exactly what happened after Broadcom CEO Hock Tan's announcement: the company expects its revenue from AI chip sales to exceed this psychologically important milestone by 2027. For an industry where Nvidia seemed an unshakeable monopolist, this sounds like a declaration of war.

To understand the scale of Broadcom's ambitions, you need to recall where this company came from. Historically, Broadcom was known primarily as a manufacturer of network controllers, switches, and infrastructure chips — a reliable but not particularly visible player in the shadow of Intel and Qualcomm. Everything changed when Hock Tan, known for his aggressive acquisition strategy, transformed the company into a semiconductor giant with a market capitalization in the hundreds of billions of dollars. The acquisition of VMware in 2023 for $69 billion was only the most prominent of his deals. But Broadcom's true strategic pivot was toward custom AI chips — and it is here that the key to understanding today's announcement lies.

Unlike Nvidia, which sells universal GPU accelerators like H100 and B200 suitable for a wide range of tasks, Broadcom is betting on ASIC — application-specific integrated circuits designed for specific customers and specific workloads. Among its clients are first-tier technology giants: Google with its TPU (Tensor Processing Unit), Meta, and, according to analysts, a number of other large cloud providers. The logic is simple: when you train models at the scale of GPT or Gemini and spend billions of dollars annually doing so, even a ten-percent gain in energy efficiency or performance per watt pays for itself many times over.

A custom chip tailored precisely to your architecture can deliver this gain — and free you from dependence on a single supplier.

Fear of dependence on Nvidia has, perhaps, been the main driver of Broadcom's growth in the AI segment. Over the past three years, the world's largest technology companies have found themselves in a position they categorically dislike: they stand in line for GPUs, agree to Nvidia's terms, and build their infrastructure around a single CUDA ecosystem. For corporations accustomed to controlling every level of their technological stack, this is an unacceptable situation. Broadcom offers a way out — not a replacement for Nvidia as such, but an alternative path where the customer receives a chip designed for their needs, with their specifications, integrated into their infrastructure.

The $100 billion forecast is certainly ambitious, and it deserves a healthy dose of skepticism. According to analysts' estimates, in fiscal year 2025, Broadcom's AI revenue amounted to roughly $12–15 billion. Growth to $100 billion in two years means an increase of approximately seven times — a pace that requires not just organic growth, but fundamental expansion of the customer base and production volumes.

Nevertheless, there are grounds to believe that Tan is not bluffing. The total volume of AI infrastructure investments is growing rapidly: Microsoft, Google, Amazon, and Meta collectively plan to spend more than $300 billion on capital expenditures in 2026–2027, and a significant portion of these funds will go directly to chips. If Broadcom can capture even a third of this flow, the $100 billion figure stops seeming fantastical.

For Nvidia, this announcement is not a catastrophe, but a serious signal. Jensen Huang's company continues to control the overwhelming majority of the AI accelerator market and possesses a powerful software ecosystem that cannot be replicated in a year or two. However, the very fact that a competitor publicly declares billion-dollar prospects speaks to the fact that the AI chip market is ceasing to be a monopoly. It is becoming a battlefield where custom solutions from Broadcom, proprietary developments from Google and Amazon, as well as efforts by AMD and new startups like Cerebras and Groq gradually erode the dominance of a single player.

What does all this mean for the industry as a whole? First and foremost — healthy competition, which will inevitably lead to lower costs for AI computing. When customers have a real choice between universal GPUs and custom ASICs, prices begin to move downward and innovation moves upward. For companies building AI products, this means more accessible infrastructure. For end users — faster implementation of artificial intelligence in everyday services. Hock Tan's announcement is not merely a corporate forecast. It is a marker of a new stage in the development of the AI industry, in which monopoly over computing power gives way to a multipolar market. And this market is only beginning to form.