Google Deepens Marvell Partnership in Major AI Chip Move
Google is set to bolster its partnership with Marvell Technology to design custom artificial intelligence chips, in another effort to diversify and strengthen its AI computing ecosystem.
The deal includes an option for Google to buy $12.2 billion worth of Marvell shares, which would make it one of Marvell’s largest investors. It also gives Marvell a bigger foothold in the market for custom chips.
The deal comes as Big Tech starts to seek alternatives to Nvidia’s GPUs for its AI needs.
Google has its own portfolio of Tensor Processing Units, while Amazon and Microsoft are also investing in custom silicon.
Google and Marvell to Develop Custom AI Silicon
As part of the expanded partnership, Google and Marvell will collaborate on products incorporating semiconductors for Google’s AI infrastructure.
The companies have inked a pact to develop a range of AI inference accelerators, memory controllers, and other specialized semiconductors. These are likely to complement Google’s Tensor Processing Units (TPUs) and reduce its reliance on silicon designers such as Nvidia and Broadcom.
Google has received a warrant that will allow it to buy nearly 59 million shares from Marvell at $206.58 per share, with the potential transaction valued at approximately $12.2 billion. The terms are laced with milestones relating to the development and revenues from customized semiconductors through 2033. The investment underscores the extent to which Google is willing to go to insulate its AI ecosystem from its cloud infrastructure rivals.
Why Google Wants Custom AI Chips
The rise of generative AI has created an insatiable demand for semiconductors. Nvidia dominates the market for graphics processing units (GPUs), which are currently the preferred choice for training and operating cutting-edge large-scale models.
Still, Nvidia’s GPUs can be prohibitively expensive and less efficient for specific applications, and other tech giants have started to develop alternatives.
Google has had the option to rely on its TPUs, but partnering with Marvell would allow it to design customized accelerators.
Customized chips can provide benefits in terms of performance, power consumption, and cost depending on the workloads they are intended for.
For instance, AI inferencing is set to consume more computational capacity worldwide, and purpose-built silicon can help reduce expenses significantly.
Marvell has been positioning itself as a critical enabler of AI infrastructure, with its broad portfolio of semiconductors for accelerators, networking, memory, and storage. The company has also recognized the importance of memory bandwidth, particularly graphic random access memory (GRAM), in enabling large-scale AI models with intensive workloads.
Nvidia Faces a More Diversified AI Chip Market
The deal between Google and Marvell does not signal the demise of Nvidia. The GPU giant retains a considerable edge over its rivals, including its expanding software ecosystem and robust networking solutions for AI infrastructure. Nvidia also has a broader presence in the market for AI accelerators. For instance, in March 2026, it announced a partnership with Marvell through its NVLink Fusion platform.
This collaboration is designed to allow Marvell to develop specialized chips and connectivity solutions for Nvidia-based AI infrastructure. Furthermore, Nvidia invested $2 billion in Marvell to support its AI initiatives. By doing so, Nvidia hopes to benefit from Marvell’s expansive portfolio of semiconductors for networking, memory, storage, and accelerators.
Meanwhile, Marvell can serve as a critical supplier of customized silicon to Google while also supporting Nvidia’s efforts in AI.
In other words, Marvell can benefit from Nvidia’s ecosystem while also helping Google reduce its reliance on the GPU maker.
Broadcom Could Also Feel the Heat
The partnership with Marvell has implications for Broadcom, which has traditionally supplied Google with application-specific integrated circuits (ASICs) for its TPUs. Announcements of the Google-Marvell deal sent Broadcom’s shares plummeting, after Marvell’s stock climbed sharply following the news. Analysts, however, are quick to note that the deal does not spell the end of Broadcom’s relationship with Google.
In essence, Google would continue to source TPUs from Broadcom while also tapping into Marvell’s capabilities to design accelerators meant for different workloads. The same would apply to other cloud computing firms, including Amazon, Microsoft, and Meta.
Instead of relying on a single supplier of customized silicon to satisfy their diverse needs, these firms can diversify their chipset suppliers. In turn, Google can mitigate the risk of a supply chain disruption and gain more leverage in its dealings with semiconductor manufacturers.
The AI Chip Race is Heating Up
The Google-Marvell deal is only the latest development in the broader competition for dominance in the AI silicon space. The market for AI accelerators is becoming fragmented as more firms invest in different approaches to AI computing. Although GPUs remain essential, TPUs and other specialized processors are also garnering interest from cloud infrastructure providers.
Firms such as Google, Amazon, Microsoft, and Meta, continue to push the envelope in their efforts to optimize AI processing, training, and inferencing. The same players are also investing in other complementary components, including networking switches, storage, and memory technologies. One of the main reasons why these firms are pouring resources into AI infrastructure is that it has become one of the most expensive IT components.
In other words, owning a customized and optimized chipset can provide considerable advantages over cloud infrastructure rivals.
Implications of the Google-Marvell Deal
The Google-Marvell deal could have far-reaching implications for the broader technology sector. If Google continues to invest in custom silicon, it would force Nvidia to contend with a broader array of rivals in the AI space. Although the GPU maker has a slight edge in terms of performance, it cannot ignore the possibility that other firms might develop compelling alternatives to its GPUs. After all, TPUs have already proven to be a viable option for AI inferencing, and other accelerators can be optimized to perform similar functions.
Still, there is no doubt that Nvidia has a considerable head start in the race to dominate the AI chipset market. Nvidia’s in-house developed chips, cutting-edge manufacturing processes, and robust ecosystem of software and networking solutions give it a significant edge over the competition. The same applies to its $5 billion investment in the AI infrastructure fund in Q2 2026. This move was designed to enable Nvidia to accelerate its efforts to secure a greater share of the $50 billion+ market for data center infrastructure.
In particular, the partnerships with financial institutions such as BlackRock, JPMorgan Chase, and Goldman Sachs will help Nvidia raise more than $500 billion in capital. These measures will allow Nvidia to fund its R&D efforts in AI accelerators, GPUs, networking switches, and software while also helping it to scale its operations.
Google is Pursuing a More Diversified AI Hardware Strategy
The deal with Marvell highlights Google’s desire to develop a more diverse AI hardware ecosystem. It can continue to invest in its TPUs while also relying on Marvell to develop accelerators and other specialized semiconductors. Doing so would reduce supply chain risks while also giving it more flexibility in terms of hardware selection. In other words, Google can leverage Marvell’s expertise in accelerators, networking, memory, and storage to build more optimized AI infrastructure. It will be able to employ a combination of TPUs and different accelerators for processing, memory, storage, and networking.
This approach will be especially useful as Google’s AI workloads intensify in the coming years.
AI inference is set to become significantly more important in the near term, requiring more investment in accelerators, memory bandwidth, and storage. Marvell’s emphasis on these components makes it well-positioned to support Google’s AI needs in the years to come.
The deal with Marvell also indicates that Google is not satisfied with its options in terms of AI accelerators. Although it has been relying on Broadcom to supply its TPUs, it also wants other alternatives to ensure sustained supply and optimal performance. The same applies to Amazon, Microsoft, and Meta, which are also considering partnerships with other silicon manufacturers.
The Outlook for Google and Marvell
One of the most crucial questions about the Google-Marvell deal is how quickly the two companies will be able to bring their joint efforts to market. The terms of the deal suggest that both parties can scale their operations if they can deliver compelling semiconductors. In other words, the long-term success of the partnership will be determined by its ability to satisfy Google’s AI infrastructure needs.
For Nvidia, the rising competition in the AI silicon space is yet another challenge it must overcome.
It should be noted that the GPU maker has already sustained considerable pressure in recent months, as Amazon, Meta, and Google have invested in TPUs and other accelerators. The competition will only intensify in the near term as more firms are likely to enter the fray.
Nvidia, however, should benefit from its first-mover advantage and its broader ecosystem of networking and software solutions. The company will also be able to tap into its considerable capital reserves to fund its R&D and attract key talent. In the long run, the competition will lead to better and more affordable AI infrastructure, and Nvidia is well-positioned to benefit from these dynamics while also profiting from the overall industry boom.
Conclusion
Google’s expanded partnership with Marvell highlights the technology giant’s determination to diversify its options in the highly competitive and lucrative AI chipset market. The two companies have announced their intention to collaborate on custom-designed AI chips, with Google set to benefit from Marvell’s extensive expertise in a broad range of semiconductors.
The deal includes an option for Google to buy $12.2 billion worth of Marvell shares, further illustrating its commitment to the endeavor. While Nvidia retains its leading position in the market for AI accelerators, the competition from Google, Amazon, Meta, and Microsoft could intensify in the years to come.
In particular, Marvell’s expanded role in AI infrastructure presents a significant challenge to Nvidia’s GPUs. However, the latter is poised to continue to dominate the market for AI accelerators for the foreseeable future due to its robust ecosystem and extensive supply chain. On the other hand, Google’s partnership with Marvell will help it to mitigate its reliance on Broadcom. The company can also leverage Marvell’s experience in accelerators, networking, memory, and storage to augment its TPUs and build its own optimized AI infrastructure. Ultimately, the competition in the AI semiconductor space will lead to more diversified options and benefit the industry as a whole.
