Key Takeaways
- Google is negotiating with Samsung to manufacture an I/O component for its upcoming Icefish TPU using 2nm process technology
- TSMC would still produce the main computing die, but Google is actively diversifying its chip supply chain
- Mass production could begin as early as 2028, with MediaTek contributing to chip design
Google is in talks with Samsung Electronics to manufacture part of its next-generation artificial intelligence processor, according to a report from The Information. The discussions center on Samsung producing a component that connects the chip to memory, using its 2-nanometer production technology.
The processor, codenamed 'Icefish,' would still rely on TSMC for its main computing die. But the memory input-output component could go to Samsung. Mass production could start as early as 2028.
This split-manufacturing approach shows Google's effort to reduce its dependence on a single chip foundry. With AI demand straining TSMC's capacity, Big Tech companies are scrambling to secure alternative production sources.
What Google Is Building
Google's in-house Tensor Processing Units have become a credible alternative to Nvidia's dominant graphics processors. The company unveiled two new custom chips in April 2025, designed for training AI models and running inference workloads.
TPU sales have become a growth driver for Google Cloud. By building its own silicon, Google can optimize chips specifically for its AI workloads rather than relying on general-purpose hardware.
The Icefish project involves MediaTek, the Taiwanese chip design firm, contributing to the chip's architecture. This marks another departure from Google's traditional partnerships and suggests the company is casting a wider net for chip expertise.
Why Samsung Matters
Landing a Google contract would be a significant win for Samsung's foundry business. The Korean company has struggled to match TSMC's yield rates and customer roster, but it has been aggressively investing in advanced manufacturing.
In April, Samsung said it expects to win more customers for chips made with its advanced process technology. The company is also considering a second Texas plant to increase production capacity.
Samsung already scored a major AI chip deal this year. In July 2025, Tesla agreed to a $16.5 billion contract with Samsung to manufacture AI chips using its 2nm process. A Google partnership would further validate Samsung's capabilities in cutting-edge chip production.
Google's Foundry Diversification
The Samsung talks are part of a broader strategy. The Information reported earlier this week that Google is also in discussions with Intel to manufacture more than three million TPUs by 2028.
TSMC dominates advanced chip manufacturing, but that dominance creates risk. If demand exceeds capacity, or geopolitical tensions disrupt supply, companies relying solely on TSMC face production bottlenecks.
By splitting the Icefish chip between foundries, Google hedges its bets. TSMC handles the most complex computing logic, where its process leadership matters most. Samsung takes the I/O die, which connects the processor to high-bandwidth memory.
The 2nm Race
The 2-nanometer process represents the next frontier in chip manufacturing. Smaller transistors mean more computing power in less space, with better energy efficiency.
TSMC, Samsung, and Intel are all racing to bring 2nm chips to market. TSMC is expected to begin volume production in 2025. Samsung has been less specific about timelines but has committed to the technology for major customers like Tesla and potentially Google.
For AI applications, efficiency gains matter enormously. Training large language models requires massive computing clusters that consume significant power. Even small improvements in chip efficiency translate to lower operating costs and reduced energy consumption at scale.
What We Don't Know
Neither Samsung nor Google has confirmed the negotiations. Samsung declined to comment. Alphabet did not respond to a request for comment. Reuters could not independently verify The Information's report.
The 2028 timeline is tentative. Chip development can slip by years, and manufacturing yields at new process nodes often take time to stabilize. Samsung's 2nm yields remain a question mark.
It's also unclear how much of the Icefish chip's value would reside in the Samsung-manufactured component. The I/O die handles memory connections, not the core AI computations. This suggests Google may be testing Samsung's capabilities on a less critical piece before committing to more.
Logicity's Take
Frequently Asked Questions
What is Google's Icefish chip?
Icefish is the codename for Google's next-generation Tensor Processing Unit, an AI chip designed for training and running machine learning models. Mass production could begin as early as 2028.
Why is Google working with Samsung instead of just TSMC?
Google wants to reduce its dependence on a single chip manufacturer. TSMC faces capacity constraints from surging AI demand, and diversifying suppliers reduces risk if bottlenecks occur.
What is a 2nm chip?
The 2-nanometer process refers to the size of transistors on a chip. Smaller transistors allow more computing power in less space, with better energy efficiency. It represents the cutting edge of semiconductor manufacturing.
Is Samsung's chip manufacturing as good as TSMC?
Samsung has historically lagged TSMC in yield rates and process technology. However, it has been investing heavily in advanced manufacturing and recently won a $16.5 billion AI chip contract from Tesla.
When will Google's Icefish chip be available?
According to reports, mass production could begin as early as 2028. However, chip development timelines often slip, and manufacturing at new process nodes takes time to stabilize.
Another look at how companies are building alternatives to dominant platforms
Need Help Implementing This?
Source: Tech-Economic Times / ET
Manaal Khan
Tech & Innovation Writer
Produced with AI assistance and reviewed by the Logicity editorial team. Learn more in our Editorial Policy.
