On July 9, 2026 (Beijing time), the three major U.S. stock indexes closed mixed. The Dow Jones Industrial Average fell 1.09% to 52,348.39 points, the Nasdaq rose 0.20% to 25,870.65 points, and the S&P 500 dropped 0.28% to 7,482.71 points. The semiconductor sector stood out, with the Philadelphia Semiconductor Index up 2.23%. Broadcom Inc. (NASDAQ: AVGO) surged 4.83% to close at $388.69, hitting an intraday high of $395.09—the highest since June 22.
This rally is not an isolated market fluctuation. Over the past week, Broadcom has issued several key signals: it signed a new multi-year partnership with Apple, extending their collaboration to 2031 and expanding into custom ASIC chips; it jointly launched the Jalapeño custom AI inference chip with OpenAI, with the first servers scheduled for deployment by the end of 2026; and its AI semiconductor revenue for fiscal Q2 2026 reached $10.8 billion, up 143% year-over-year. These three major catalysts have prompted the market to reprice Broadcom’s AI business. Let’s break down how Broadcom is becoming a core beneficiary of AI infrastructure across three dimensions: custom ASIC chips, AI networking chips, and its customer ecosystem.
ASIC Custom Chips: Broadcom’s "Second Trump Card"
Broadcom is often traditionally viewed as a "network chip company." However, its fiscal Q2 2026 earnings report clearly reveals another growth trajectory—custom AI accelerator chips (ASICs).
On June 3, 2026, Broadcom reported its fiscal Q2 2026 results: total revenue was $22.19 billion, up 48% year-over-year, setting a new record. AI semiconductor revenue soared to $10.8 billion, up 143% year-over-year, beating both company guidance and Wall Street expectations. Non-GAAP EPS was $2.44, above analysts’ forecast of $2.40. The company projects full-year AI semiconductor revenue for fiscal 2026 to reach $56 billion, up roughly 180% from fiscal 2025.
Order visibility is even more noteworthy. On the earnings call, Broadcom CEO Hock Tan revealed that Q2 AI semiconductor orders exceeded $30 billion, while actual shipments were $10.8 billion. Additional data shows that AI chip contract backlog stands at $73 billion, with $53 billion from custom accelerators. Order visibility now extends through fiscal 2028. Broadcom also reaffirmed its target of surpassing $100 billion in AI semiconductor revenue for fiscal 2027.
J.P. Morgan’s recent semiconductor industry report estimates the digital AI ASIC market will reach $60–70 billion by 2026, with a compound annual growth rate of over 40–50% in the coming years. Broadcom currently holds about 80–85% share of the high-end ASIC market, with Marvell in second place at about 10–12%.
Broadcom’s ASIC model competes with NVIDIA’s general-purpose GPU approach. NVIDIA offers standardized compute products, while Broadcom custom-designs AI accelerator chips for core clients such as Google, Meta, Anthropic, and OpenAI. The moat of this model is time—designing, verifying, and deploying custom chips with Broadcom typically takes over two years, making it extremely costly for clients to switch suppliers.
J.P. Morgan projects Broadcom’s AI revenue will jump from about $20 billion in fiscal 2025 to over $60 billion in 2026, and could surpass $150 billion in fiscal 2027. Its project pipeline covers Google TPU, Meta MTIA, ByteDance AI video and networking chips, OpenAI XPU, SoftBank/Arm XPU, and Anthropic’s rack-scale TPU solutions.
Apple Partnership Extended to 2031: Strategic Upgrade from RF Components to AI Custom Chips
On July 6, 2026, Broadcom announced a new multi-year agreement with Apple, extending their long-term technology partnership through 2031. Under the deal, Broadcom will develop and supply a range of custom ASIC chips for multiple generations of Apple products, upgrading the collaboration from traditional RF and connectivity components to AI-customized silicon.
Apple stated that it will invest over $30 billion in U.S.-made chips through this partnership, which is expected to drive the production of over 15 billion U.S.-manufactured chips. This is one of the largest investments to date under Apple’s "Made in America" initiative. Broadcom will invest $1.5 billion in capital expenditures to expand and upgrade its Fort Collins, Colorado facility.
The key upgrade in this renewal is the substantial expansion of the partnership—from traditional RF connectivity chips to AI-specific custom ASICs. Apple is developing a dedicated AI server chip, codenamed "Baltra," slated for deployment in 2027 using Broadcom technology. This chip will support cloud-based Apple Intelligence features, including text generation, image generation, and information summarization for heavy AI workloads.
Broadcom has long supplied Apple with critical chip components, including custom RF chips for iPhones, Wi-Fi and Bluetooth connectivity chips, and other networking semiconductors. Apple accounts for about 20% of Broadcom’s annual revenue, making it one of its largest customers. Extending the ASIC agreement to 2031 sends two clear signals to the market: first, Broadcom remains irreplaceable in Apple’s supply chain; second, the partnership is shifting from traditional connectivity chips to higher-value AI and custom compute chips. In the context of the AI industry’s rapid iteration cycles, this 2031 endpoint means Broadcom and Apple’s technology partnership will span at least three generations of AI chip architectures.
AI Networking Chips: The Underestimated "Compute Pipeline"
Investment logic for AI infrastructure is shifting from "single-point compute" to "system-level infrastructure." Training a trillion-parameter large language model requires not just tens of thousands of GPUs working in parallel, but also high-speed, low-latency data transfer between those GPUs. The networking layer—historically seen as just a "pipeline"—is now a key bottleneck determining the actual compute utilization of AI clusters.
Broadcom also holds an irreplaceable position in this area. In March 2026, Broadcom’s Tomahawk 6 switch chip entered mass production, delivering 102.4 terabits per second of switching capacity—the industry’s only mass-produced 102.4T Ethernet switch chip. Jericho 4 began shipping in August 2025, offering 51.2 terabits per second of bandwidth and supporting secure, lossless networking for clusters of over one million XPUs. Tomahawk Ultra, meanwhile, achieves ultra-low latency down to 250 nanoseconds.
Analysts predict Broadcom’s AI networking business revenue could more than double by fiscal 2027, surpassing $45 billion, driven by the rapid adoption of 1.6Tbps optical communications and next-gen platforms like Tomahawk 6, Jericho 4, and Tomahawk Ultra.
In the first half of 2026, the five hyperscale cloud providers—Microsoft, Amazon, Google, Meta, and Oracle—all raised their capital expenditure guidance. BofA Securities analysts forecast global hyperscaler AI capex will exceed $800 billion in 2026, up 67% year-over-year, and top $1 trillion in 2027. At Fiber Connect 2026, Cisco noted that AI is pushing network architecture from the core to the edge, with AI traffic now accounting for 5% of backbone network utilization—up from less than 1% two years ago.
This structural shift means that investment logic for AI infrastructure is moving from "who has the most powerful GPU" to "whose data center architecture is most complete and efficient." Broadcom holds irreplaceable positions in both custom ASIC chips and AI networking chips.
Customer Ecosystem: Long-Term Partnerships with Six Core Clients
Broadcom’s AI business growth does not rely on a single client, but is built on a diversified ecosystem of hyperscale customers. J.P. Morgan’s report shows Broadcom’s project pipeline includes Google TPU, Meta MTIA, ByteDance AI video and networking chips, OpenAI XPU, SoftBank/Arm XPU, and Anthropic’s rack-scale TPU solutions.
Broadcom has made significant progress in deepening these long-term relationships. In April 2026, Alphabet extended its custom AI chip partnership with Broadcom through 2031. That same month, Meta expanded its partnership with Broadcom through 2029 to jointly develop multiple generations of custom AI processors. In June 2026, OpenAI announced a strategic partnership with Broadcom to co-develop systems—including accelerators and Ethernet solutions—aiming for initial deployment in the second half of 2026 and completion of around 10 GW of OpenAI-designed AI accelerator deployments by the end of 2029.
This diversified customer structure reduces dependence on any single client and enables Broadcom to accumulate technology and scale across clients. Each generation of custom chip design experience can be leveraged for the next generation or new customer projects, creating a positive feedback loop.
Conclusion
Broadcom is undergoing a valuation re-rating from "network chip supplier" to "core AI infrastructure platform." This transformation is anchored by three verifiable pillars: order visibility for custom ASIC chips now extends through 2028, Tomahawk 6 mass production has further solidified its technological leadership in AI networking chips, and long-term agreements with six core clients—including Apple, Google, Meta, and OpenAI—provide strong revenue visibility.
J.P. Morgan projects that by 2027, unit shipments of AI ASICs/XPUs will surpass GPUs. Total AI accelerator shipments are expected to reach 23.3 million units, with GPUs accounting for 10.9 million (47%) and ASICs/XPUs for 12.5 million (53%). This does not mean demand for NVIDIA will quickly diminish, but rather highlights a diversification in AI infrastructure investment: GPUs will continue to serve general-purpose training and inference, while cloud providers’ in-house ASICs will gain higher penetration for large-scale, stable, and predictable internal workloads.
For investors, Broadcom’s case illustrates the long-term logic of the AI hardware value chain expanding from GPUs to ASICs, advanced packaging, HBM interfaces, SerDes, optical interconnects, and CPO. As the focus of compute investment shifts from "single chip" to "full data center architecture," Broadcom, with its dual strengths in custom ASICs and AI networking chips, is emerging as 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 has two main segments: first, custom AI accelerator chips (ASICs), which are tailor-made AI processors for clients like Google, Meta, and OpenAI; second, AI networking chips, including Tomahawk 6 switch chips and Jericho 4 routing chips, which enable high-speed interconnects in AI data centers. In fiscal Q2 2026, these two segments together contributed $10.8 billion in AI semiconductor revenue, up 143% year-over-year.
Q2: How does Broadcom compete with NVIDIA in AI chips?
Broadcom and NVIDIA are not direct competitors—they have differentiated positions. NVIDIA provides standardized general-purpose GPUs for a broad range of training and inference scenarios. Broadcom focuses on custom ASICs for hyperscale clients, optimizing for specific workloads in terms of performance and power efficiency. J.P. Morgan expects ASIC shipments to surpass GPUs by 2027, reflecting the diversification—not replacement—of AI compute demand.
Q3: What is the significance of Broadcom’s new partnership agreement with Apple?
In July 2026, Broadcom and Apple extended their partnership through 2031, expanding from traditional RF connectivity chips to custom AI ASICs. Apple will invest over $30 billion in U.S.-made chips through this agreement. The deal provides Broadcom with a decade of revenue visibility and marks Apple’s formal adoption of Broadcom technology for its AI server chip (codename Baltra).
Q4: What is Broadcom’s AI revenue outlook for fiscal 2026?
Broadcom expects full-year AI semiconductor revenue for fiscal 2026 to reach $56 billion, up about 180% from fiscal 2025. The company also reaffirmed its goal of surpassing $100 billion in AI semiconductor revenue for fiscal 2027. In fiscal Q2 2026, AI semiconductor revenue already reached $10.8 billion, up 143% year-over-year.
Q5: Is Broadcom’s AI business growth sustainable?
Broadcom’s AI business growth is underpinned by verifiable orders. The company disclosed an AI chip contract backlog of $73 billion, with order visibility through fiscal 2028. Broadcom also has long-term agreements with six core clients—including Google, Meta, Apple, and OpenAI—with most partnerships extending through 2029–2031. The design cycle for custom chips typically exceeds two years, making it extremely costly for clients to switch suppliers and creating a structural moat.
