Written by: Rita

Trends Guideline

On July 7, JPMorgan held an investor meeting with Broadcom's management, led by CEO Chen Fuyang. The core information revealed at the meeting was very dense: Google's TPU v9 roadmap is proceeding as planned, Apple's ASIC partnership has been extended to 2031 with new AI accelerator projects, OpenAI's next-generation chip is about to go into production, and demand for the Tomahawk 6 switching chip is exceeding supply. JPMorgan maintains an overweight rating on Broadcom, defining it as the world's second-largest AI semiconductor supplier, the largest custom chip ASIC supplier, and the largest networking semiconductor supplier.

Custom Chips: Demand for Inference is Changing the Market Landscape

Broadcom's management provided an assessment: the rapid growth of inference workloads is accelerating the adoption speed of custom XPUs. Among the leading large model builders, the shipping ratio of XPUs to GPUs may approach 50 to 50 next year. This means that custom chips are transitioning from a "supporting role" to being "on par with GPUs."

Customers are demanding an accelerated timeline for the next generation of ASIC production and pursuing more chip design variants. The evolution of inference workloads from single inference to inference chains and then to AI entities has significantly increased the strategic value of a customized chip strategy focused on workload optimization. The competition over the cost per token is becoming a new battlefield for AI computing power.

Broadcom and OpenAI's first-generation XPU "Jalapeno" took only 9 months from design to production, validating Broadcom's execution capability in complex multi-chip design, advanced packaging, and high-speed SerDes. The next generation of OpenAI chips will "soon" go into production. Management emphasized that these cutting-edge XPUs are "pushing the limits" in multiple dimensions such as bandwidth, SerDes, power consumption, and packaging, with complexity often being underestimated by the market.

Regarding the competitive landscape of custom chips, management provided a clear assessment of the threat posed by mega-scale customers developing their own chips (COTs): modern XPUs are no longer simple single-chip designs, but increasingly utilize multi-chip configurations, advanced packaging, and high-speed SerDes, facing significant technical barriers in mass production. Configurations of 6 chips and 8 chips have become industry standard configurations, which is precisely where Broadcom's core barrier lies.

Google TPU and Apple ASIC: Two Long-term Growth Lines

There had been rumors in the market that Google's next-generation TPU project could be delayed or canceled. Broadcom's management clearly responded: the roadmap is "proceeding completely as planned," and 400G SerDes chips have been developed and are operational. Broadcom also confirmed that the TPU revenue in its five-year agreement with Google is growing year-on-year and maintains a majority share, which was a key premise for Broadcom entering into this agreement and a direct response to the previous market concerns over "share loss."

On the Apple side, Broadcom's partnership with Apple extends over ten years, involving multiple ASIC categories including touch controllers and wireless charging. The latest agreement extends their design collaboration to 2031, covering new design sockets (which JPMorgan believes includes AI/data center ASICs), and also includes a radio frequency supply agreement. Broadcom's expansion of its wafer fabrication in Fort Collins, Colorado further confirms the company’s confidence in long-term opportunities. The significance of this agreement lies not only in the long-term visibility of revenue but also in the fact that Apple, as one of the strictest chip customers globally, continues to choose Broadcom as its ASIC partner, which itself is a testament to Broadcom's technical strength.

AI Computing and Network Dual-Engine Drive

Broadcom's management does not believe there will be an oversupply of AI computing power in 2027 or 2028. The Tomahawk 6 switching chip is sold out, with major customers being leading large model builders. When allocating Tomahawk and Jericho switching products, Broadcom will prioritize meeting the needs of XPU customers.

Broadcom's competitive barriers in the AI networking field are deeper than the market perceives. The pricing power of Tomahawk 6 comes from two aspects: first, the technological lead of the switching chip itself, and second, the deep binding relationship with custom chip customers. This "computing + networking" combination strategy is forming a positive flywheel, where the success of custom computing drives network share, and the scarcity of network resources in turn deepens collaborations with XPU customers. After purchasing Broadcom's custom chips, customers tend to naturally prefer using Broadcom's networking solutions to form a complete AI computing power stack, with very high switching costs.

On the cost side, Broadcom stated that it will maintain stable ASIC gross margins through technological leadership and will not allow rising costs of wafers, substrates, HBM, etc., to erode its profit margins. Management pointed out that "in any case," they will keep ASIC gross margins within the target range. Technological leadership is the fundamental source of pricing power. Management also explicitly refuted the claim that customers could independently procure HBM, emphasizing that HBM is an inseparable part of XPU design, which must be closely connected with core logic chips to ensure performance; customers cannot separate HBM from Broadcom's designs for independent procurement.

Trends Perspective

The signals released from Broadcom's management meeting are very focused: the demand for custom AI chips is expanding from "one company (Google)" to "multiple top-tier customers." The three major customers, Google, Apple, and OpenAI, are accelerating their developments simultaneously, with each line representing years of long-term partnerships. Apple's involvement is particularly noteworthy; Apple has primarily relied on self-development in the AI chip field, and this time, by procuring AI accelerator ASICs from Broadcom, it indicates that Apple's AI computing strategy may be shifting from "fully in-house development" to "partially outsourced." This is an important structural expansion signal for the total addressable market of the entire custom chip industry.

Broadcom's current valuation logic is switching from "semiconductor cyclical stock" to "core supplier of AI infrastructure." The sellout of Tomahawk 6, the queue for ASIC customers, and long-term agreements extended to 2031 all point to a situation where demand visibility far exceeds any previous cycle.

The validation points will be in the second half of the year. The production rhythm of TPU v9, the production progress of OpenAI's next-generation chip, and the specific rollout time of Apple’s AI accelerator project will be key variables in determining Broadcom's growth curve beyond 2027.

Disclaimer

This article is a compilation and interpretation of the research report by a third-party brokerage (JPMorgan, July 7, 2026) by Trend Research. The ratings, target prices, earnings forecasts, and related judgments cited in the article represent the views of the brokerage analysts and only reflect the position of their respective organizations, not the views of Trend Research, and do not constitute any investment advice.

The market carries risks, and decisions should be made independently. This article should not be used as a basis for buying or selling any securities.