Alphabet (Google’s parent company) is significantly expanding its collaboration with Intel for the production of its custom AI chips, Tensor Processing Units (TPUs). This strategic move aims to diversify its supply chain, reducing reliance on TSMC amidst high demand and validating Intel’s foundry capabilities. Google has reportedly placed an order for 3 million TPUs with Intel, underscoring its commitment to specialized AI processors for its cloud and AI initiatives.
A cornerstone of Alphabet's (GOOGL) (GOOG) artificial intelligence (AI) strategy has always centered on its proprietary custom-built semiconductors. More than a decade ago, Google's parent company began developing its Tensor Processing Units (TPUs), processors specifically engineered for AI tasks, which have since become a significant competitive advantage.
These application-specific integrated circuits (ASICs) are, by Google's definition, "a chip designed for a single, specific purpose" – to deliver the immense computational power required for the unique matrix and vector-based mathematics essential to building and running AI models.
Alphabet's latest move has drawn considerable attention, signaling a deep commitment to its existing deal with Intel (INTC).
Image source: The Motley Fool.
Titans Join Forces for AI
Intel and Alphabet have a history of collaboration, working for years on specialized processors designed for AI workloads. However, the manufacturing of these chips traditionally fell to Taiwan Semiconductor Manufacturing (TSMC). As the AI revolution continues to accelerate, TSMC has faced increasing challenges in meeting the escalating global demand, creating a significant opportunity for Intel.
Reports indicate that Google has placed a substantial order with Intel for 3 million TPUs, slated for delivery through 2028. This monumental commitment follows months of rigorous testing by Google to ensure Intel's advanced chip packaging technology could meet its stringent performance standards.
An order of this scale could be transformative for Intel, suggesting that Google is strategically diversifying its advanced chipmaking suppliers and is no longer content to rely solely on TSMC for its needs. This move provides Google with an essential alternative source for AI-centric chips, mitigating potential bottlenecks that arise from a single-provider dependency. Furthermore, it serves as powerful validation for Intel's recently revitalized foundry business, which is strongly positioned to capitalize on the ongoing surge in AI demand.
Google's AI Core Strategy
These custom processors are becoming increasingly integral to Alphabet's overarching cloud and AI strategies. Earlier this year, at its Cloud Next conference, Google unveiled two potent new AI chips that represent a significant evolution from their predecessors. While past efforts focused on general-purpose TPUs, this year introduced distinct architectures: the TPU 8t, optimized for training AI models, and the TPU 8i, designed specifically for inference tasks.
During the unveiling, Amin Vahdat, Google's senior VP and chief technologist for AI and infrastructure, emphasized the strategic rationale behind this specialization. In a world increasingly driven by AI agents, he noted, "we determined the community would benefit from chips individually specialized to the needs of training and serving." Vahdat highlighted that this specialization "unlocks significant efficiencies and gains," enabling Google to execute its most demanding AI workloads "two to four times faster and at a 30% lower cost" compared to its previous generation of TPUs.
In a notable shift from its historical practice of using these chips exclusively in-house, Google recently announced its intention to sell TPUs "to a select group of customers." Executives stated that this decision is expected to significantly "expand our total addressable market."
This strategic expansion and increased demand have contributed to a substantial increase in Google's backlog, which has nearly doubled year over year to a staggering $460 billion.
