Broadcom and Apple extend collaboration to 2031: How ASIC custom chips reshape the AI infrastructure competitive landscape

On July 6, 2026, a filing submitted by Broadcom (AVGO.US) to the U.S. Securities and Exchange Commission stirred up the entire semiconductor investment community. The filing disclosed that Broadcom and Apple (AAPL.US) have signed a new multi-year agreement extending their long-term technology partnership through 2031. Under the agreement, Broadcom will develop and supply a range of custom ASIC (Application-Specific Integrated Circuit) chips for multiple generations of Apple products.

The day after the announcement (July 7 Beijing time), Broadcom's stock closed up 3.73% at $373.90, having surged as much as 6.3% to $383.16 during the session. This gain not only reflects the market's optimistic interpretation of the agreement itself but also highlights a reassessment of the long-term value of the custom chip sector by capital markets.

The simultaneous resonance of strategic positioning by traditional tech giants and abnormal price movements in crypto assets on the same day—could there be a deeper industrial logic behind it? This article will start from the Broadcom-Apple partnership extension, break down the competitive landscape of the custom chip sector, and explore the second curve of AI infrastructure investment.

Broadcom-Apple Partnership Extended to 2031: Three Layers of Signals from One Agreement

The collaboration between Broadcom and Apple dates back nearly a decade. In January 2020, the two companies announced a wireless component supply agreement worth approximately $15 billion. In May 2023, they signed another multi-year agreement worth billions of dollars for Broadcom to develop and manufacture 5G RF components. This renewal through 2031 is not just an extension of time but also a leap in the depth of their cooperation.

First layer signal: Capability upgrade from RF components to custom ASICs. Broadcom has long supplied Apple with custom RF chips for iPhones, Wi-Fi and Bluetooth connectivity chips, and other networking semiconductors. The core change in the new agreement is the expansion from traditional RF components to custom ASIC chips. ASICs are custom chips designed for specific applications, seeing surging demand in AI inference and high-performance computing. This shift means Broadcom's role in Apple's supply chain has upgraded from a "connectivity component supplier" to a "computing chip co-design partner."

Second layer signal: Official endorsement of Apple's "Baltra" AI chip. Earlier market reports indicated that Apple is collaborating with Broadcom on an AI server processor codenamed "Baltra," expected to be manufactured using TSMC's N3P process and slated for mass production in 2026. Bank of America Securities previously noted in a report that Broadcom's fifth AI ASIC customer "could be Apple." The signing of the new agreement confirms that assessment. With Apple joining, Broadcom's AI ASIC customer portfolio now includes five giants: Google, Meta, ByteDance, OpenAI, and Apple.

Third layer signal: Locked-in certainty for 20% of annual revenue base. According to analyst estimates, Apple contributes approximately 20% of Broadcom's annual revenue, making it its most important customer. There had been market concerns that Apple's ongoing push for in-house chip development (including the C1 modem) could gradually reduce its reliance on Broadcom. In June, Broadcom's stock had plunged about 20% over such fears. The new agreement converts this potential risk into long-term stable orders, significantly boosting Broadcom's revenue visibility for years to come.

Why Can't Apple Do Without Broadcom? A Moat Analysis of Custom Chips

Given Apple's continued push to strengthen its in-house chip strategy—from M-series Mac processors to A-series iPhone chips and the self-developed C1 cellular modem—why does it still heavily rely on Broadcom in wireless communications and custom ASICs?

Technology barriers: RF and connectivity chip expertise is hard to replicate in the short term. Broadcom has decades of accumulated technology and patent portfolios in RF front-end, Wi-Fi, and Bluetooth connectivity chips. These chips involve complex analog circuit design, signal processing, and power optimization, representing a different technology stack from Apple's in-house digital logic processors (e.g., A-series, M-series). Even though Apple has introduced its self-developed N1 chip (integrating Wi-Fi and Bluetooth) for the latest iPhones, iPads, and Macs, it remains difficult in the short term to break away from Broadcom in analog chip areas like RF front-end.

Economies of scale: Cost-sharing logic of custom chips. The development and tape-out costs of ASIC chips are extremely high, requiring sufficient shipment volumes to amortize. While Apple's hundreds of millions of iPhones shipped annually provide scale support, Broadcom also serves multiple hyperscale customers like Google, Meta, and Microsoft, allowing it to spread R&D costs across a broader customer base. This scale effect gives Broadcom a cost and technology iteration speed advantage that single-point suppliers find hard to match.

Co-design capability: Closed loop from spec definition to physical implementation. Custom ASICs are not simply "fabrication by blueprint"; they require deep collaboration between the chip supplier and the customer from the architecture design stage. Broadcom, together with Marvell, holds about 95% of the hyperscale cloud custom AI accelerator co-design market. This co-design capability is built on mutual trust and understanding accumulated over long-term partnerships, forming a moat that new entrants find difficult to cross in the short term.

Supply chain resilience: Apple's diversified procurement strategy. Apple has been pursuing a supply chain diversification strategy in recent years, but signing long-term agreements with core suppliers like Broadcom in key chip areas is itself part of supply chain resilience. In early 2026, memory prices surged nearly 98%, and Apple raised prices on MacBooks and iPads in June to reflect cost pressures. In this context, locking in long-term supply prices and capacity for key chips is a rational strategic choice for Apple.

Broadcom vs. Marvell: A Duopoly Showdown in the Custom Chip Arena

In the hyperscale cloud custom AI accelerator co-design market, only two companies globally have true competitiveness: Broadcom and Marvell. Together, they hold about 95% of this market. However, their strategic paths and market positioning differ significantly.

Market share and scale: Broadcom's absolute lead. Broadcom holds over 70% of the custom AI chip market. In the second quarter of fiscal 2026, Broadcom's AI semiconductor revenue reached $10.8 billion, up 143% year-over-year. The company expects full-year fiscal 2026 AI semiconductor revenue to reach $56 billion and over $100 billion in fiscal 2027. Broadcom's AI ASIC customers now include Google (TPU series), Meta, ByteDance, OpenAI, and newly confirmed Apple.

Marvell's catch-up: Smaller but faster. Marvell's full-year fiscal 2026 revenue was $8.2 billion, up 42% year-over-year, with data center revenue of $6.1 billion, up 46%. JPMorgan expects Marvell's data center revenue to grow from about $6.1 billion in 2025 to about $9.3 billion in 2026 and reach about $14.6 billion in 2027. Marvell's customers include Amazon (Trainium series), Microsoft (Maia series), and others.

Competitive landscape: Differentiation, not zero-sum. The competition between Broadcom and Marvell is not a simple zero-sum game. Broadcom's advantage lies in broader customer coverage and greater scale effects; Marvell seeks breakthroughs through deep engagement with specific clients. JPMorgan expects custom AI chip shipments could surpass GPUs by 2027, meaning the overall market pie is still expanding rapidly, providing ample growth space for both players.

Potential risks for Broadcom. Broadcom is not without worries. In early June 2026, MediaTek, by successfully integrating Google's 336G SerDes solution, secured Google TPU orders. This shows that even in areas where Broadcom dominates, the competitive landscape is dynamically changing. Additionally, Broadcom's stock is still about 24% below its 52-week high of $494.35, reflecting market disagreement over its valuation and growth sustainability.

Viewing the Second Curve of AI Infrastructure Investment through Broadcom

The extension of the Broadcom-Apple agreement is not just good news for one chip supplier; it also reflects a deep shift in the logic of AI infrastructure investment.

First curve: GPU computing arms race. Over the past two years, computing investment centered on NVIDIA GPUs has dominated the AI infrastructure narrative. But the generality of GPUs means suboptimal power and cost configurations in specific scenarios. As AI workloads shift from training to inference, demand for customized, scenario-specific computing power is rising.

Second curve: Custom ASIC computing revolution. ASICs are designed for specific purposes, offering higher performance and energy efficiency than general-purpose chips under particular workloads. Counterpoint Research predicts that custom ASIC shipments will triple between 2024 and 2027. A Morgan Stanley report on June 23 shows that global CoWoS advanced packaging demand will reach 2.69M wafers in 2027, up 93% from 1.39M in 2026, with cloud vendors' in-house ASICs becoming another growth curve for the CoWoS market.

Broadcom's strategic positioning: From "shovel seller" to "shovel maker." Broadcom's uniqueness lies in not only providing ASIC chips but also deeply participating in the entire process from architecture definition to physical implementation. This "co-design" model makes it a core builder of underlying computing power for AI infrastructure. The Jalapeño inference chip co-developed by Broadcom and OpenAI went from design to tape-out in just 9 months, expected to reduce inference costs by about 50%—this efficiency itself is a testament to its technological moat.

Implications for the crypto industry. The crypto mining industry has also experienced a similar evolution from general-purpose GPUs to custom ASICs. The iteration of Bitcoin miners from CPU to GPU to ASIC is highly analogous in underlying logic to the shift from GPU to ASIC in AI computing—once workloads become sufficiently standardized and scaled, the advantages of custom chips in energy efficiency and cost are irreversible. The transition of Bitcoin miners like TeraWulf into AI computing further confirms the liquidity of computing infrastructure across different application scenarios.

Conclusion

The extension of the Broadcom-Apple partnership to 2031 is not just a renewal of a supply agreement but a confirmation of the strategic value of the custom chip era. It tells us three facts:

First, in the semiconductor industry, "in-house" and "outsourcing" are not binary opposites. Even a powerhouse like Apple still relies on specialized partners like Broadcom in specific areas such as RF, connectivity, and custom ASICs. The moat of custom chips lies not only in the technology itself but also in the systematic capabilities accumulated through long-term co-design.

Second, the ASIC track is emerging from the shadow of the GPU narrative to become an independent main line of AI infrastructure investment. From Google TPU to Apple Baltra, from OpenAI Jalapeño to Amazon Trainium, hyperscale customers are voting with real money for customized computing power.

Third, for investors, understanding the duopoly landscape of Broadcom and Marvell holds more long-term value than chasing short-term stock price fluctuations. Broadcom locking in Apple through 2031 is not only a certainty anchor for AVGO's custom chip business but also a landmark event for the entire custom chip sector transitioning from "concept" to "performance."

Back to the market on July 7, 2026—Bitcoin broke through $64,000, and Broadcom closed up 3.73%. Two seemingly unrelated asset prices moved abnormally on the same day, sharing an underlying narrative: computing power is becoming the scarcest asset in the digital age. Whether in decentralized crypto networks or centralized AI cloud services, the thirst for computing power is reshaping the valuation system of the entire tech industry. And custom chips are the most fundamental blueprint in this reshaping.

FAQ

Q: How is Broadcom's new agreement with Apple different from the 2023 agreement?

The 2023 agreement focused mainly on the development and manufacturing of 5G RF components, with a value of tens of billions of dollars. The new agreement expands the scope from RF components to custom ASIC chips, covering multiple generations of Apple products, with the partnership extended to 2031. This is a substantive upgrade from "component supply" to "computing chip co-design."

Q: What percentage of Broadcom's annual revenue does Apple account for?

According to estimates from multiple institutions, Apple contributes approximately 20% of Broadcom's annual revenue, making it Broadcom's largest single customer. The stability of Apple's orders directly impacts Broadcom's valuation logic.

Q: What is Broadcom's market share in the custom AI chip market?

Broadcom holds over 70% of the custom AI chip market. In the hyperscale cloud custom AI accelerator co-design market, Broadcom and Marvell together hold about 95% of the share.

Q: What is Broadcom's AI business revenue in 2026?

In the second quarter of fiscal 2026, Broadcom's AI semiconductor revenue reached $10.8 billion, up 143% year-over-year. The company expects full-year fiscal 2026 AI semiconductor revenue to reach $56 billion.

Q: What is the main difference between custom ASIC chips and GPUs?

GPUs are general-purpose computing chips capable of handling multiple tasks; ASICs are custom chips designed for specific purposes, offering higher performance and energy efficiency under particular workloads. As AI inference demand grows, ASICs are becoming an important source of computing power for AI infrastructure.