Report: Google may collaborate with MediaTek to develop AI chips, Broadcom's exclusive supply may be impacted

According to reports, Broadcom's stock fell more than 3.9% during trading after news emerged that Google might collaborate with MediaTek to develop AI chips.

On March 17th, Eastern Time, The Information cited two individuals involved in the project, revealing that Google is preparing to collaborate with MediaTek to develop a new generation of AI processing chips, TPU, which is planned to go into production next year.

The timing of this announcement coincides with Broadcom's strong promotion of its potential revenue of hundreds of billions of dollars from TPU and other custom AI chip businesses in the coming years.

However, this does not mean that Google has completely severed ties with Broadcom. For the past nine years, Google has exclusively collaborated with Broadcom to develop AI chips. According to Broadcom employees, both parties are still negotiating to continue jointly designing some of Google's AI chips.

But if Google ultimately implements this new plan, Broadcom may have to share the TPU business with its competitor MediaTek. Analysts point out that this move could mean Broadcom will lose exclusive control over Google's TPU business, which may impact Broadcom's AI revenue expectations.

Google Seeks Cost Reduction and Efficiency Improvement

Google has differences with its long-term partner Broadcom over TPU pricing, prompting Google to start looking for alternative partners like MediaTek.

The Information cited insiders from TSMC and Broadcom, revealing that part of the reason Google chose MediaTek is that MediaTek's chip quotes are lower than Broadcom's, helping to optimize supply chain costs.

According to one individual involved in the project, Google will lead most of the work in the design of the next-generation TPU, including the design of the processor, while MediaTek's main responsibility will be to manage the input/output (I/O) module, which handles communication between the main processor and peripheral components. This collaboration model differs from Google's partnership with Broadcom, as Broadcom previously participated in developing the core chip of the TPU.

However, Google is also working to internalize more chip development work, further reducing its reliance on Broadcom or MediaTek. According to someone who has discussed this matter with Google's management, Google is also willing to fully develop its chips independently. According to someone who recently interviewed for a chip design position at Google, the company is actively recruiting chip design engineers.

Analysts say that this move by Google could pose a threat to companies like Broadcom. If Google successfully achieves self-developed chips, Broadcom will lose one of its largest customers, which will impact its revenue and profits.

Broadcom Still Benefits from Google Orders but Faces Competition

Even if Google takes on more chip development work, according to another insider who has participated in Google's TPU project, Google may still rely on partners like MediaTek or Broadcom to handle the chip production process, including packaging and quality testing.

Some analysts point out that Broadcom still benefits from Google's orders. Currently, Google remains the largest source of revenue for Broadcom's custom AI chip business. Google invests billions of dollars annually in Broadcom's chips, while Broadcom is responsible for the physical design of the TPU, developing chips based on blueprints provided by Google Broadcom stated last week that its AI chip design and sales revenue grew by 77% year-on-year in the fiscal quarter ending February 2, reaching $4.1 billion. In contrast, NVIDIA's AI chip revenue is about ten times that of Broadcom.

Additionally, Broadcom is co-designing AI chips or chip network technologies with large companies such as Apple, Arm Holdings, Meta, and OpenAI.

Google Attempts to Reduce Dependence on NVIDIA

Google typically produces two different types of TPUs, one for training new models and the other is "inference" chips that support services like Google Search, YouTube, and the Gemini application. It is currently unclear whether Google plans to adjust its strategy to focus on producing only one type of TPU.

In the field of AI chips, TPUs have been a competitive advantage for Google, helping it reduce dependence on NVIDIA. In contrast, competitors like OpenAI and Meta remain more reliant on NVIDIA, especially in the past two years, as demand for NVIDIA's AI chips surged, with major companies racing to develop large language models (LLMs) such as ChatGPT.

According to research firm Omdia, Google's investment in TPUs last year was estimated to be between $6 billion and $9 billion. Nevertheless, Google remains one of NVIDIA's major customers, as it not only rents NVIDIA's chips to cloud computing clients but also uses them internally. Previously, Google had ordered over $10 billion worth of NVIDIA's flagship Blackwell chips.

Other companies are following Google's lead in developing their own AI chips, such as OpenAI planning to launch its self-developed chips next year.

According to sources involved in OpenAI's chip development, some companies, including OpenAI, are emulating Google's approach because they recognize the TPU strategy and wish to reduce dependence on NVIDIA to avoid being constrained by NVIDIA's limited chip supply