Broadcom’s next-generation TPU project delayed? JPMorgan refutes: Google has secured orders through 2031, and the company’s AI revenue is expected to triple next year.

```

In response to recent market rumors about delays or even cancellation of Google’s next-generation TPU v9 project, JP Morgan’s latest report clearly refutes this, stating that the project developed by Broadcom is still progressing as planned and is expected to enter mass production in 2028, with no unexpected variables as feared by outsiders. The report further notes that Google’s internal chip team is still at least 18 months behind Broadcom in terms of technical capability for a true challenge.

More importantly, JP Morgan believes the market has overlooked the five-year cooperation agreement signed between both parties in March this year. This agreement not only secures Broadcom’s leading role in the design of the next four generations of TPU (v8 to v11), but also includes commitment to yearly increasing purchases, extending the visibility of Broadcom’s AI business revenue directly to 2031.

From JP Morgan’s perspective, investors are excessively focused on the potential competition brought by Google’s in-house ASIC, while underestimating Broadcom’s deep-rooted moat in advanced chip design, packaging technology, and IP accumulation. As global major model companies accelerate their shift to custom AI chips, Broadcom is expected to be one of the biggest beneficiaries of this infrastructure wave. The report forecasts that the company’s AI business revenue will grow 2 to 2.5 times year-over-year in 2027, and will double again in 2028.

TPU v9 project “has not been delayed nor canceled”

JP Morgan’s latest industry chain investigation shows that Google’s next-generation AI training chip, TPU v9, remains unchanged, is still being developed by Broadcom, and is planned for mass production in 2028. Rumors about delays or cancellations are not true. The chip uses 2nm process, integrates 4 compute chips, 16 HBM groups, and 400Gbps SerDes interfaces, making it a key part of Google’s AI infrastructure.

In fact, Broadcom started IP design in the first half of 2025 and will shift to SoC development in the second half of the year. This project has always been the top internal priority for the company. Meanwhile, early research and development of the next-generation TPU v10 has begun. From product planning to R&D pace, Google and Broadcom’s partnership is solid, and concerns about cracks in their collaboration are unfounded.

Five-year agreement locks in next four generations of TPU

The report points out the market is overlooking an even more critical message: In March this year, Google and Broadcom signed a new five-year cooperation agreement, locking in the roadmap for the next four generations of TPU: v8, v9, v10, and v11. The agreement covers not only long-term supply arrangements but also includes Google’s revenue commitments, providing Broadcom’s AI business with strong certainty.

JP Morgan expects this means Broadcom’s TPU-related revenue will continue to grow year by year until 2031. In the current AI industry chain, very few companies have secured long-term purchasing commitments from major model providers, and Google’s ongoing investment undoubtedly opens up more clarity for Broadcom’s growth in the coming years.

Google’s in-house chip cannot disrupt Broadcom in the short term

Over the past year, Google has strengthened its internal chip team (COT) and introduced more suppliers for TPU development, which many investors see as Broadcom’s biggest future risk. But JP Morgan believes these concerns are exaggerated.

The report notes that Google’s internal team is currently optimizing the existing TPU v8t (Zebrafish) project, while Broadcom’s TPU v8i (Sunfish), completed certification in mid-2025 and is about to enter mass production. There’s still at least an 18-month technology gap between the two in advanced AI chip design.

The report emphasizes that ASIC competition is far more than a single technology indicator comparison—it involves computing architecture, network design, storage systems, advanced packaging, and IP accumulation, multidimensional comprehensive competition. These are exactly the core moats Broadcom has built over decades of investment. The report specifically notes that in the past 12 years, Broadcom has helped Google complete the design of 14 cutting-edge chips; its experience and execution capability cannot be copied in the short term.

Outstanding advantages in AI order reserves

Broadcom’s client layout advantages in the AI ASIC market should not be ignored.

Currently, Broadcom owns the largest global reserve of AI ASIC projects, with clients including top AI players like Google, Anthropic, OpenAI, Meta, and also participates in SoftBank/Arm’s XPU project, SambaNova’s RDU project, and Apple’s AI CPU project.

As more large AI model companies seek to reduce dependence on general GPUs and turn to custom ASICs, Broadcom is becoming one of the biggest beneficiaries of this trend. Based on this, JP Morgan expects Broadcom’s AI business revenue will increase 2 to 2.5 times year-over-year in 2027 and double again in 2028.

Advantages may be underestimated: advanced packaging and mass production barriers

Currently, the market discussion is overly focused on whether Google will support its internal chip team and ignores Broadcom’s real competitive advantage.

Whether it is advanced chip design, packaging technology, IP accumulation, or large-scale mass production experience, Broadcom still maintains a clear lead. As Singapore’s advanced substrate plant is launched and advanced packaging production capacity ramps up in 2028, the company’s lead in the AI ASIC market could be further expanded in the coming years.

From JP Morgan’s perspective, the market sees Google trying to reduce its reliance on Broadcom, but overlooks that Google’s next four generations of TPU are actually deeply tied to Broadcom. In the custom chip competition in the AI era, Broadcom is still one of the hardest players to replace.

Risk Warning and DisclaimerThe market has risks, and investment needs caution. This article does not constitute personal investment advice, nor does it take into account the special investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, viewpoints, or conclusions in this article are suitable for their own specific circumstances. Those who invest on this basis shall bear their own responsibility. ```