BroadcomThis has significantly raised the long-term outlook for the custom AI accelerator and data center networking businesses. As more companies seek to design their own AI chips, this signals that not only NVIDIA’s general-purpose GPUs but also custom semiconductors and networking equipment are becoming key pillars of AI infrastructure investment.
During a conference call held after its earnings announcement, Broadcom projected AI semiconductor revenue for fiscal year 2027 to be approximately $115 billion...was the forecast. This figure is higher than the previously projected $100 billion, and for fiscal year 2028, it is expected to be approximately $230 billionI predicted that it could double again.
However, The figures for 2027 and 2028 are not actual revenue that has already been generated, but rather the company's long-term projections based on customer investment and securing supply. We need to look at current performance and future guidance separately.
| Third-quarter AI semiconductor revenue: $16.7 billion
Broadcom's total revenue for the third quarter of fiscal year 2026 was $29.591 billionThis represents an 86% increase compared to the same period last year. Of this total, revenue from AI semiconductors was $16.7 billionThis represents a 221% increase from a year ago and a 54% increase from the previous quarter.
The company had previously forecast third-quarter AI semiconductor revenue of approximately $16 billion. The actual results exceeded that level, and in the fourth quarter, AI semiconductor revenue is expected to $21.7 billionIt is expected to increase by 236% compared to the previous year.
AI semiconductors are also rapidly accounting for a larger share of overall revenue. Based on a simple calculation using total third-quarter revenue, AI semiconductors account for approximately 56% of revenue.
| Custom Accelerators and Networks Grow Together
Broadcom's AI business isn't structured around selling just one type of chip. Major cloud companies develop chips tailored to their own needs Custom AI AcceleratorsIt helps design these systems and also supplies network semiconductors and equipment that connect numerous accelerators into a single large-scale system.
As AI models grow larger, The ability to quickly exchange data between chips is becoming just as important as the computational performance of individual chips. As a result, the value of network technologies such as Ethernet switches, optical components, and interconnects is also increasing.
Broadcom explained that demand for both custom accelerators and networking solutions is strong. This means that investment in AI infrastructure is expanding beyond just GPU purchases to encompass data movement and system architecture as a whole.
| What the $115 billion Forecast Means
AI Semiconductor Revenue Forecast for Fiscal Year 2027 $115 billionThat's about double the forecast for fiscal year 2026. The company's previous forecast for 2027 was over $100 billion, but this time it has raised that specific figure even higher.
For fiscal year 2028, they projected approximately $230 billion again. If the forecast holds true, That amounts to a very steep growth trajectory, with sales nearly doubling for two consecutive years.
This figure reflects the assumption that infrastructure investments by major AI companies and platform operators—including OpenAI, Anthropic, and Meta—will continue to expand. At the same time, it also accounts for the trend among customers to opt for custom chips tailored to their own models and data center architectures rather than purchasing only general-purpose chips.
| More Options Are Emerging Even in the NVIDIA-Dominated Market
NVIDIA's influence in the AI accelerator market remains overwhelming. However, enterprises with very high usage have a strong incentive to use their own chips or custom chips alongside NVIDIA's to improve power efficiency, supply stability, and the performance and cost of specific tasks.
Broadcom supports these companies by facilitating design collaboration and connecting them to mass production. From the customers' perspective, rather than completely replacing NVIDIA GPUs, Depending on the nature of the task, you'll have the option to use a combination of general-purpose GPUs and custom accelerators.
Therefore, it is difficult to view Broadcom's growth as merely a shift in market share among suppliers. As AI data centers grow in scale, we are getting closer to a point where the types of chips and system configurations are becoming more diverse.
| Supply Capacity Is Key to Achieving Forecasts
Even if there is sufficient demand, if we can't actually produce the chips and install them in data centers, it won't translate into revenue. Multiple supply chains—including advanced process manufacturing capabilities, high-bandwidth memory, packaging, substrates, network components, and power and cooling equipment—must all work together simultaneously.
Although Broadcom has stated that it is securing the necessary supply, maintaining such a high growth rate will require stable production capacity over the long term. Even if a specific customer’s investment schedule is delayed or data center power connections are delayed, the timing of revenue recognition could change.
Custom chips also tend to be highly dependent on customers. We also need to consider that design changes or order adjustments by a small number of large customers can have a greater impact on overall performance than those involving generic products.
| We need to distinguish between long-term outlooks and actual results
While the third-quarter AI semiconductor revenue of $16.7 billion is already reported, the figures of $115 billion for 2027 and $230 billion for 2028 are management's projections. Actual results may vary depending on changes in customer demand, supply contracts, production schedules, and market conditions.
Furthermore, even if long-term sales increase significantly, it cannot be assumed that profitability will improve at the same rate. This is because costs related to securing production capacity, contract terms with individual customers, intensifying competition, and the development of next-generation chips may all rise simultaneously.
The key points to look for going forward are Will fourth-quarter AI semiconductor revenue reach the $21.7 billion figure the company projected?...whether supply capacity will be secured as planned by 2027 and how much more diverse the customer base for custom chips will become. This announcement can be seen as a strong signal that the AI semiconductor market is rapidly expanding beyond general-purpose GPUs to include custom accelerators and networks.
Tags
#Broadcom #Broadcom #AI Semiconductors #AI Chip #Custom AI Accelerator #Data Center #Semiconductor Infrastructure #Network Semiconductors #AI Infrastructure #Earnings Release #ASIC #Cloud Computing
