One of the key tenets of Alphabet's (GOOGL -1.44%) (GOOG -1.20%) artificial intelligence (AI) strategy has long been the company's custom-built semiconductors. The Google parent first developed its Tensor Processing Units (TPUs) more than a decade ago, and these custom processors were "purpose-built specifically for AI," according to the company, becoming a key part of its strategic advantage.
These application-specific integrated circuits (ASICs) are "a chip designed for a single, specific purpose," according to Google: providing the computational horsepower for "running the unique matrix and vector-based mathematics that's needed for building and running AI models."
Alphabet's latest move turned heads, and the company appears to be going all-in on its deal with Intel (INTC -2.47%).
Image source: The Motley Fool.
A pairing of titans
Intel and Alphabet have been collaborating for years, developing specialized processors for AI, but those chips have been manufactured by Taiwan Semiconductor Manufacturing (TSMC). However, as the AI boom continues to gain steam, TSMC has been struggling to keep up with the relentless demand. That's where Intel comes in.
Google has reportedly placed an order for 3 million TPUs with Intel to be delivered through 2028. This newfound confidence comes after Google spent months testing Intel's chip packaging technology to ensure the company could meet its rigid standards.
An order of this magnitude could be a game changer for Intel, suggesting Google is no longer willing to rely solely on TSMC for its advanced chipmaking needs. This would give the company an additional source of AI-centric chips, helping it avoid bottlenecks that result from relying on a single provider. It could also serve as further validation for Intel's recently reinvigorated foundry business, driven by the ongoing demand for AI.
NASDAQ: GOOGL
Key Data Points
At the heart of Google's AI strategy
These custom processors have become an increasingly important part of Alphabet's cloud and AI strategy. Earlier this year, at the company's Cloud Next conference, Google introduced two powerful new AI chips that differed from their predecessors in one key way. Whereas the company had previously focused its efforts on all-purpose TPUs, this year saw the release of two distinct architectures -- the TPU 8t and TPU 8i. As you may have guessed, the TPU 8t is dedicated to training workloads, while the TPU 8i was designed for inference.
At the unveiling, Amin Vahdat -- Google's senior VP and chief technologist for AI and infrastructure -- noted that in a world of AI agents, "we determined the community would benefit from chips individually specialized to the needs of training and serving," he said. The Google exec said that this "specialization unlocks significant efficiencies and gains." Specifically, the company has been able to run its most demanding AI workloads "two to four times faster and at a 30% lower cost" than with its previous-generation TPUs.
Google recently announced that it will sell TPUs "to a select group of customers," a notable shift from its historical practice of only using these chips internally. Executives said this decision will significantly "expand our total addressable market."
This has helped fuel a sizeable increase in Google's backlog, which nearly doubled year over year to $460 billion.
For all that opportunity, Alphabet stock is selling for a song at just 28 times earnings. Time to buy before the market comes to its senses.
