JPMorgan Chase: Custom chips will surpass GPUs in shipments by 2027; Broadcom's TPU "supply chain disappearance" does not indicate doubts about orders.

JPMorgan Chase: Custom chips will surpass GPUs in shipments by 2027; Broadcom's TPU "supply chain disappearance" does not indicate doubts about orders.

The demand for AI computing power is driving the chip industry from a "GPU monopoly" to a parallel expansion of GPUs and custom chips. JPMorgan Chase's latest Fall 2026 US Semiconductor and Semiconductor Equipment Industry Update report predicts that ASICs/XPUs will account for 54% of AI accelerator unit shipments in 2027, surpassing GPUs; and will further rise to 55% in 2028.

JPMorgan Chase projects that the custom AI ASIC market will reach $60 billion to $70 billion by 2026, with a compound annual growth rate (CAGR) of over 40% to 50% in the coming years. Currently, Broadcom and Marvell hold approximately 90% of the market share, with Broadcom accounting for about 80% to 85%, indicating a high degree of market concentration.

The report also points out that the relatively opaque supply chain information for Google's TPU and other custom chip projects does not necessarily mean that orders lack certainty. Broadcom's five-year TPU supply agreement with Google covers 2026 to 2031, involving 3nm, 2nm, and advanced packaging, and includes an annually increasing TPU revenue arrangement, providing strong revenue visibility for related AI businesses.

More broadly, AI capital expenditure remains the most important demand pillar for the semiconductor industry. JPMorgan Chase projects that global semiconductor revenue will grow by 118% year-on-year in 2026, or 32% excluding memory; and by 35% overall in 2027, or 18% excluding memory. During the same period, wafer equipment spending is expected to grow by 31% and 38% respectively, with demand from AI, memory, and traditional chips collectively driving the continuation of the industry cycle.

ASICs are gaining rapid penetration, and custom chips will become an important source of incremental AI computing power.

Hyperscale cloud providers are accelerating the development of ASICs/XPUs, with the core purpose not simply to replace GPUs, but to optimize performance, power consumption, and cost per token for specific workloads, while reducing their dependence on general-purpose GPUs.

This trend is changing the product structure of AI accelerators. JPMorgan Chase predicts that ASICs/XPUs will account for approximately 41% of AI accelerator unit shipments in 2026, rising to 54% in 2027, and reaching 55% in 2028. Custom chips are moving from supplementary solutions to large-scale production, becoming an important incremental factor in the expansion of AI infrastructure.

The custom AI ASIC market is currently highly concentrated, with Broadcom and Marvell together accounting for approximately 90% of the market share. As cloud providers such as Google, Amazon, and Microsoft advance their self-developed AI chips, the demand for ASIC design and related supporting technologies still has significant room for growth.

The lack of transparency regarding Broadcom TPU information does not imply a lack of order certainty.

One concern regarding Broadcom's AI business is the limited supply chain information for projects like Google's TPU, making it difficult for external parties to fully track orders and shipments. However, JPMorgan Chase believes that limited supply chain information does not necessarily indicate weakening demand; rather, attention should be paid to indicators such as customer agreements, product iterations, and capacity planning.

Taking Google's TPU as an example, Broadcom's supply agreement with Google is for a period of 5 years, covering 2026 to 2031, and spanning 3nm, 2nm, and advanced packaging architectures. The agreement also includes an annually increasing TPU revenue arrangement, so even though the details of the supply chain are not entirely transparent, orders and revenue still have a high degree of visibility.

Marvell also benefits from cloud vendors' self-developed chips entering mass production, with its custom chip business covering projects such as Amazon Trainium, Microsoft Maia, and Google XPU. As self-developed accelerators transition from early deployment to large-scale production, ASIC design, interconnect, and advanced packaging will all benefit simultaneously.

AI capital expenditure continues, with WFE and storage cycles further strengthened.

JPMorgan Chase projects that global cloud computing capital expenditures for CY26, CY27, and CY28 will reach $953 billion, $1.41 trillion, and $1.54 trillion, respectively. As returns on AI investments gradually become apparent, cloud service providers are increasingly motivated to maintain high levels of infrastructure investment.

The equipment sector is also maintaining strong growth. Global WFE spending is projected to increase by 31% year-on-year to approximately US$225 billion in 2026, and then by another 38% to approximately US$263 billion in 2027. Expansion of advanced process technology and new DRAM capacity will jointly drive equipment demand, while the shortage of cleanroom space is also prompting some equipment purchases to be brought forward.

Storage has become another important driver. JPMorgan Chase predicts that the average price of both DRAM and NAND will increase by about 250% in 2026, and then by about 30% and 25% respectively in 2027. AI servers are driving DRAM demand, while enterprise-grade SSDs are becoming an important source of NAND demand. Long-term procurement agreements and relatively restrained capital expenditures also help to prevent supply from spiraling out of control quickly.

Overall, JPMorgan Chase believes that the current semiconductor upswing is not driven solely by demand for AI chips, but rather by the simultaneous expansion of multiple supply chains, including custom chips, cloud capital expenditure, storage, and wafer equipment. Within this framework, the sustainability of investment in AI infrastructure remains the core variable determining whether the semiconductor boom cycle can continue.

Risk warning and disclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.