Broadcom AVGO rises nearly 5%: How can AI custom chips ASIC become the second growth curve outside of NVIDIA?

July 9, 2026 (Beijing time), the three major U.S. stock indexes closed mixed. The Dow fell 1.09% to 52,348.39, the Nasdaq rose 0.20% to 25,870.65, and the S&P 500 fell 0.28% to 7,482.71. The semiconductor sector stood out, with the Philadelphia Semiconductor Index up 2.23%. Among them, Broadcom Inc. (NASDAQ: AVGO) surged 4.83%, closing at $388.69 and hitting an intraday high of $395.09—its highest level since June 22.

This gain is not an isolated market fluctuation. Over the past week, Broadcom has released multiple key signals in succession: it reached a new multi-year partnership agreement with Apple, extending their cooperation to 2031 and expanding into custom ASIC chips; it jointly released the custom AI inference chip Jalapeño with OpenAI, with plans to deploy the first batch of servers by the end of 2026; in Q2 of fiscal year 2026, AI semiconductor revenue reached $10.8 billion, up 143% year over year. With these three positive catalysts stacked together, the market has moved to reprice the AI business value of Broadcom. From three dimensions—ASIC custom chips, AI networking chips, and the customer ecosystem—we break down the logic chain of why Broadcom is becoming a core beneficiary of AI infrastructure.

ASIC Custom Chips: Broadcom’s “Second Ace”

The outside world’s traditional perception of Broadcom often stays at “a networking chip company.” But the financial results for Q2 of fiscal year 2026 clearly reveal another growth curve—custom AI accelerator chips (ASICs).

On June 3, 2026, Broadcom announced its Q2 results for fiscal year 2026: total revenue of $22.19 billion, up 48% year over year, setting a record high. Among this, AI semiconductor revenue reached $10.8 billion, up 143% year over year—exceeding not only the company’s own expectations but also Wall Street analysts’ forecasts. Non-GAAP earnings per share were $2.44, beating analysts’ expectations of $2.40. The company expects full-year fiscal year 2026 AI semiconductor revenue to reach $56 billion, up about 180% from fiscal year 2025.

What is even more noteworthy is order visibility. During the earnings call, Broadcom CEO Hock Tan revealed that in Q2, AI semiconductor orders exceeded $30 billion, while actual shipments were only $10.8 billion. Other data shows that contract backlog orders for AI chips total as high as $73 billion, of which $53 billion comes from custom accelerators. Order visibility has extended to fiscal year 2028. Broadcom also reiterated its goal of achieving AI semiconductor revenue exceeding $100 billion in fiscal year 2027.

In a recent semiconductor industry research report, JPMorgan estimates that the digital AI ASIC market will reach about $60 billion to $70 billion by 2026, and that in the coming years the compound annual growth rate will remain above 40% to 50% . Broadcom currently holds about 80% to 85% share in the high-end ASIC market. Marvell ranks second with about 10% to 12%.

Broadcom’s ASIC model differs from NVIDIA’s general-purpose GPU model in competitive positioning. NVIDIA offers standardized compute products, while Broadcom tailors AI accelerator chips for core customers such as Google, Meta, Anthropic, and OpenAI. The moat of this model lies in time cost—once Broadcom completes the design, verification, and deployment of custom chips, the entire process typically takes more than two years, making the cost for customers to switch suppliers extremely high.

JPMorgan expects Broadcom’s AI revenue to jump significantly from about $20 billion in fiscal year 2025 to over $60 billion in fiscal year 2026, and to reach over $150 billion in fiscal year 2027. Its project pipeline includes Google TPU, Meta MTIA, ByteDance AI video and networking chips, OpenAI XPU, SoftBank/Arm XPU, and Anthropic-related TPU rack-level solutions.

Apple Renewed Through 2031: Strategic Upgrade from RF Components to AI Custom Chips

On July 6, 2026, Broadcom announced that it reached a new multi-year agreement with Apple, extending their long-term technology cooperation relationship to 2031. Under the agreement, Broadcom will develop and supply a series of custom ASIC chips for multiple generations of Apple products. The scope of cooperation has been comprehensively upgraded—from traditional RF and connectivity components to AI-customized silicon chips.

Apple announced that this cooperation will invest more than $30 billion to purchase U.S.-made chips, which is expected to drive production of more than 15 billion U.S.-made chips. This is one of the largest investments to date under Apple’s “U.S. manufacturing plan.” Broadcom will invest $1.5 billion in capital expenditures to expand and upgrade its facilities located in Fort Collins, Colorado.

The core upgrade of this renewal lies in the substantive expansion of the cooperation scope—from traditional RF connectivity chips to AI-dedicated custom ASIC chips. Apple is developing a dedicated AI server chip codenamed “Baltra.” It plans to deploy it in 2027 using Broadcom technology to support cloud-based Apple Intelligence features, including heavy-load AI tasks such as text generation, image generation, and information summarization.

For a long time, Broadcom has supplied Apple with key chip components, including custom RF chips used in iPhones, Wi-Fi and Bluetooth connectivity chips, and other networking semiconductor products. Apple contributes about 20% of Broadcom’s annual revenue, making it one of its largest customers. With this ASIC long-term contract extended through 2031, the market receives two signals: first, Broadcom remains irreplaceable in Apple’s supply chain; second, the cooperation scope has shifted from traditional connectivity chips to higher-value AI and customized compute chips. With the cooperation endpoint in 2031, viewed in the context of AI industry iteration measured in months, it means the technological binding between Broadcom and Apple will span at least three generations of AI chip architecture cycles.

AI Networking Chips: The “Compute Power Pipeline” That Was Underestimated

The investment logic for AI infrastructure is shifting from “standalone compute power” to “system-level infrastructure.” Training a large language model with a trillion parameters not only requires the parallel compute power of tens of thousands of GPUs, but also high-speed, low-latency data transmission between these GPUs. The networking layer—an area previously viewed as a “pipeline”—is becoming a key bottleneck that determines the actual utilization of compute power within AI clusters.

Broadcom holds an irreplaceable position in this area as well. In March 2026, Broadcom’s Tomahawk 6 switching chip entered mass production. Its switching capacity reaches 102.4 Terabit per second, making it the industry’s only mass-produced 102.4T Ethernet switching chip at this scale. Jericho 4 has been shipping since August 2025 with bandwidth of 51.2 Terabit per second, supporting secure, lossless network architecture for more than 1 million XPU clusters. In addition, Tomahawk Ultra provides ultra-low latency as low as 250 nanoseconds.

Analysts predict that Broadcom’s AI networking business revenue may double in fiscal year 2027, exceeding $45 billion. The driving forces include the rapid adoption of 1.6Tbps optical communication speeds and the uptake of next-generation platforms such as Tomahawk 6, Jericho 4, and Tomahawk Ultra.

In the first half of 2026, five hyperscale cloud service providers—Microsoft, Amazon, Google, Meta, and Oracle—collectively raised their capital expenditure guidance. Analysts at Bank of America Securities predict that in 2026, global hyperscale cloud providers’ AI capex will exceed $800 billion, up 67% year over year, and will surpass $1 trillion in 2027. At Cisco’s Fiber Connect 2026 event, Cisco stated that AI is pushing network architecture from the core to the edge. AI traffic currently accounts for 5% of backbone network utilization, whereas two years ago this figure was less than 1%.

This structural change means the investment logic for AI infrastructure is shifting from “whose GPU is the strongest” to “whose data center architecture is the most complete and efficient.” Broadcom simultaneously occupies two irreplaceable positions: ASIC custom chips and AI networking chips.

Customer Ecosystem: Long-Term Lock-in with Six Core Customers

Broadcom’s AI business growth does not rely on a single customer; instead, it is built upon a diversified ecosystem of hyperscale customers. JPMorgan’s report shows that Broadcom’s project pipeline includes Google TPU, Meta MTIA, ByteDance AI video and networking chips, OpenAI XPU, SoftBank/Arm XPU, and Anthropic-related TPU rack-level solutions.

In terms of long-term customer relationships, Broadcom has made multiple recent advances. In April 2026, Alphabet extended its partnership with Broadcom on custom AI chips through 2031. In the same month, Meta expanded its partnership with Broadcom through 2029, jointly developing multiple generations of custom AI processors. In June 2026, OpenAI announced a strategic collaboration with Broadcom to jointly develop systems, including accelerators and Ethernet solutions. The goal is to begin deployment in the second half of 2026 and complete the deployment of approximately 10 gigawatts of OpenAI-designed AI accelerator by the end of 2029.

This diversified customer structure reduces the risk of dependence on a single customer, while also providing Broadcom with technical accumulation and scale benefits across customers. The design experience from each generation of custom chips can be reused for the next generation of products or new customer projects, forming a positive feedback loop.

Conclusion

Broadcom is undergoing a valuation reconstruction—from being seen as a “networking chip supplier” to becoming a “core platform for AI infrastructure.” This reconstruction rests on three verifiable pillars: order visibility for ASIC custom chips has extended to 2028; its technical leadership in AI networking chips has been further consolidated after the mass production of Tomahawk 6; and long-term agreements with six core customers—including Apple, Google, Meta, and OpenAI—provide revenue visibility.

JPMorgan expects that by 2027, unit shipments of AI ASIC/XPU will exceed those of GPUs. At that time, total shipments of AI accelerators are expected to reach 23.3 million units, of which GPUs will be 10.9 million units, accounting for 47%, while ASIC/XPU will be 12.5 million units, accounting for 53%. This assessment does not mean that NVIDIA’s demand will quickly weaken. Instead, it points to a divergence in AI infrastructure investment: GPUs will continue to serve general training and inference needs, while cloud providers’ self-developed ASICs will achieve higher penetration in large-scale, stable, and predictable internal workloads.

For investors, Broadcom’s case reveals the long-term logic of the AI hardware chain expanding from GPUs to ASICs, advanced packaging, HBM interfaces, SerDes, optical interconnects, and CPO. When the focus of compute power investment expands from “a single chip” to “a complete data center architecture,” Broadcom—thanks to its dual layout of ASIC custom chips and AI networking chips—has been becoming the most important AI infrastructure beneficiary outside of NVIDIA.

FAQ

Q1: What are the main segments of Broadcom’s AI business?

Broadcom’s AI business is mainly divided into two segments: one is custom AI accelerator chips (ASICs), which design dedicated AI processors for customers such as Google, Meta, and OpenAI; the other is AI networking chips, including Tomahawk 6 switching chips and Jericho 4 routing chips, used for high-speed interconnection in AI data centers. In Q2 of fiscal year 2026, these two segments together contributed AI semiconductor revenue of $10.8 billion, up 143% year over year.

Q2: How does Broadcom compete with NVIDIA in the AI chip market?

Broadcom and NVIDIA are not in direct competition; they are positioned differently. NVIDIA provides standardized general-purpose GPUs that apply to a wide range of training and inference scenarios. Broadcom focuses on customizing ASIC chips for hyperscale customers, optimizing performance and power consumption for specific workloads. JPMorgan expects that by 2027 ASIC shipments will exceed GPU shipments, but this reflects the divergence in AI compute demand rather than replacement.

Q3: What is the significance of the new cooperation agreement between Broadcom and Apple?

In July 2026, Broadcom and Apple extended their cooperation through 2031. The scope expanded from traditional RF connectivity chips to AI custom ASIC chips. Apple will invest more than $30 billion to purchase U.S.-made chips. This agreement provides Broadcom with up to a decade of revenue visibility, and also marks that Apple has officially introduced Broadcom technology into its AI server chip (codenamed Baltra).

Q4: What is Broadcom’s expected AI revenue for fiscal year 2026?

Broadcom expects full-year fiscal year 2026 AI semiconductor revenue to reach $56 billion, up about 180% from fiscal year 2025. The company also reiterates its goal that AI semiconductor revenue will exceed $100 billion in fiscal year 2027. In Q2 of fiscal year 2026, AI semiconductor revenue has already reached $10.8 billion, up 143% year over year.

Q5: Is Broadcom’s AI business growth sustainable?

Broadcom’s AI business growth is based on verifiable order demand. The company disclosed that backlog orders for AI chip contracts total as high as $73 billion, with order visibility extended to fiscal year 2028. At the same time, Broadcom has long-term agreements with six core customers—including Google, Meta, Apple, and OpenAI—whose cooperation terms generally extend to 2029-2031. The design cycle for custom chips typically exceeds two years, and customer switching costs are extremely high, forming a structural moat.