Google Just Went All in on Its Deal With Intel
Alphabet has reportedly ordered 3 million Tensor Processing Units (TPUs) from Intel for delivery through 2028, according to The Motley Fool. This strategic shift reduces Google’s reliance on Taiwan Semiconductor Manufacturing (TSMC) and leverages Intel’s foundry services to secure the hardware needed for its expanding AI infrastructure.
Why is Alphabet diversifying its AI chip production?
Alphabet is moving away from a single-source manufacturing model to avoid supply chain bottlenecks. While the company has collaborated with Intel for years, those specialized processors were previously manufactured by TSMC, according to The Motley Fool.
As AI demand surges, TSMC has struggled to keep pace. By ordering 3 million TPUs from Intel, Google creates a secondary pipeline for its AI-centric chips. This move follows months of testing by Google to ensure Intel’s chip packaging technology meets the company’s rigid internal standards.
How do the TPU 8t and TPU 8i chips differ?
Google has shifted from all-purpose chips to specialized architectures. Amin Vahdat, Google’s senior VP and chief technologist for AI and infrastructure, stated at the Cloud Next conference that the community benefits from chips specialized for either training or serving.
The company now utilizes two distinct paths:
- TPU 8t: Dedicated specifically to training workloads.
- TPU 8i: Designed specifically for inference (running the trained model).
Vahdat noted that this specialization allows Google to run demanding AI workloads two to four times faster than previous generations. Additionally, the company has reduced costs by 30%, according to official company statements reported by The Motley Fool.
Comparison: Training vs. Inference Hardware
| Feature | TPU 8t (Training) | TPU 8i (Inference) |
|---|---|---|
| Primary Goal | Building the AI model | Serving the AI model |
| Key Benefit | Computational horsepower | Operational efficiency |
Will Google sell its AI chips to other companies?
Yes. In a departure from its historical practice of keeping TPUs for internal use, Google recently announced it will sell these chips to a select group of customers. Executives stated this shift is intended to expand the company’s total addressable market.
This commercialization strategy is already impacting the company’s financials. According to The Motley Fool, Google’s backlog nearly doubled year over year, reaching $460 billion.
What happens next for Intel’s foundry business?
The 3-million-chip order serves as a major validation for Intel’s reinvigorated foundry business. Because Google spent months auditing Intel’s packaging technology before signing the deal, the partnership suggests Intel can now compete with TSMC on high-end AI specifications.
This development creates a competitive tension in the semiconductor market. By splitting orders between Intel and TSMC, Alphabet gains leverage over pricing and ensures that a geopolitical or technical failure at one foundry doesn’t freeze its AI development.
Frequently Asked Questions
What is a TPU?
A Tensor Processing Unit (TPU) is a custom-developed AI accelerator created by Google to speed up machine learning workloads using matrix mathematics.
Why did Google choose Intel over TSMC for this order?
Google isn’t replacing TSMC but is adding Intel to its supply chain to avoid bottlenecks and reduce reliance on a single provider.
How much faster are the new TPUs?
According to Amin Vahdat, the specialized TPU 8t and 8i architectures run demanding workloads two to four times faster than previous generations.
What do you think about Google’s move to sell its internal chips to other companies? Let us know in the comments below or subscribe to our newsletter for more deep dives into AI infrastructure.