MW These stocks may benefit from a shift towards Google's highly specialized AI microchips
By Jules Rimmer
Citrini Research sees an array of companies potentially benefiting from a shift away from Nvidia's GPUs
Google's AI microchips have increasingly been in the spotlight.
The apparent success of Google's Gemini 3 launch, a program running on its tensor processing units (TPU) rather than Nvidia's general processing units (GPU), means some analysts are focusing on the potential beneficiaries of this trend becoming more widespread.
And Citrini Research is asking this question: What if the AI hardware narrative shifts from Nvidia's (NVDA) dominance to other challengers, Alphabet $(GOOG)$ foremost among them, expanding the TPU ecosystem?
Citrini has identified a cohort of companies for whom that expansion is driving demand. Those include listed companies such as TSMC $(TSM)$, ASE (TW:3711), Amkor $(AMKR)$, Micron $(MU)$, Lumentum $(LITE)$, TTM Technologies $(TTMI)$ , SK Hynix (KR:000660), UmiMicron (TW:3037), Apple $(AAPL)$, Si Time $(SITM)$ and Macom $(MTSI)$.
In recent months, Alphabet's share price has outperformed Nvidia's as the market's perception of its relative positions in the AI boom has been dramatically revised.
In a Substack post, Citrini illustrates how Google is now placed in the vanguard of frontier models (general-purpose AI systems) and is doing it using its own TPUs. That's instead of the GPUs in which Nvidia has roughly a 90% market share and on which OpenAI depends so heavily for its ChatGPT large language model.
While it was reported in The Information earlier this week that Meta Platforms (META) is in negotiations with Alphabet to buy or rent billions of dollars worth of these TPUs, for Citrini Research this potential has existed for a while:
"While the writing has been on the wall for some time, the market's perception of Google has dramatically reversed over the past several months - transforming from an AI loser, bleeding its search dominance, to a stalking horse destined to undercut the most consensus AI winners.."
In Citrini's view, Meta is interested because it now seems possible to train frontier models without Nvidia's chips. Citrini now perceives an incentive to diversify away from Nvidia's CUDA (compute unified device architecture) ecosystem.
This is the relative performance of the Google complex compared with the OpenAI complex in the past year
Citrini explains that the cost of switching architecture/ecosystem is very high for AI accelerators (components that speed up AI and machine learning tasks), but potential deals struck between Alphabet and the likes of Anthropic, Fluidstack and Meta demonstrate that the moat around Nvidia's business could now be threatened. Citrini's note predicts Alphabet's TPUs represent a long-term risk to Nvidia's high margins.
Scanning for those specific suppliers who can exploit this new dynamic, Citrini said companies like TSMC, ASE and Amkor are providing advanced packaging for TPU's such CoWoS (chip-on-wafer-on-substrate).
Citrini points out that high bandwidth memory remains a bottleneck for both GPUs and TPUs, fueling demand for the services of SK Hynix, Samsung (KR:005930) and Micron .
The printed circuit boards required for TPUs are denser with a proprietary routing and these are produced by companies like Isu Petasys, TTM Technologies and Unimicron. The optical circuit switches $(OCS)$ used by TPUs are manufactured by Lumentum.
Apple may be able to leverageTPUs for cloud-based inference (running a trained AI model to make predictions using new cloud data) on its Siri intelligence system, said Citrini.
SiTime and MACOM could take advantage of an accelerated shift towards 1.6T optical modules (a transceiver that transmits data at 1.6 terabits per second).
-Jules Rimmer
This content was created by MarketWatch, which is operated by Dow Jones & Co. MarketWatch is published independently from Dow Jones Newswires and The Wall Street Journal.
(END) Dow Jones Newswires
November 28, 2025 07:44 ET (12:44 GMT)
Copyright (c) 2025 Dow Jones & Company, Inc.
Comments