Google expands partnership with Intel for AI chips

Google committed to multiple generations of use. Intel The central processing units in AI data centers are an expansion of an existing partnership.

Intel shares gained 2% on Thursday, while Alphabet shares fell more than 1%.

The internet giant has long relied on Intel processors since its first server rack targets nearly three decades ago. Intel’s latest Xeon 6 CPUs will now run AI training and inference workloads, potentially giving the chipmaker a stronger position in an AI market that has been dominated by AI so far. Nvidia.

“Their Xeon roadmap gives us confidence that we can continue to meet the increasing performance and efficiency demands of our workloads,” Amin Vahdat, Google’s chief technologist for AI infrastructure, said in a statement Thursday. he said.

No financial terms were disclosed and the companies did not provide a timeline for the deal.

The deal comes as the CPU comes to the fore in the next phase of the AI ​​race. Dion Harris, NvidiaThe head of AI infrastructure told CNBC in March that CPUs are “becoming the bottleneck” as agent workloads expand their computing needs beyond the graphics processing units that have hitherto driven AI.

“Scaling AI requires more than accelerators; it requires balanced systems,” Intel CEO Lip-Bu Tan said in a statement about the Google deal on Thursday. he said.

Intel, which has been struggling to keep up with new trends in technology for years, sold its 10% stake to the US government in August as the Trump administration highlighted the chipmaker’s ability to produce advanced chips on US soil. The following month, Nvidia announced it would purchase $5 billion worth of shares from Intel.

Intel’s shares almost tripled last year with the impact of these investments.

Intel produces its latest Xeon processor with the most advanced 18A technology at its Arizona chip manufacturing facility, which opened last year. Even though billions of dollars have been poured into the foundry side of the business, Intel’s own processors remain the new factory’s biggest customer.

But skin color sent Earlier this week, LinkedIn reported that Elon Musk had enlisted Intel to design, manufacture and package custom chips for SpaceX, xAI and Tesla’s on the ambitious Terafab project in Texas, but no financial details or timelines have been disclosed.

As part of Thursday’s announcement, Google and Intel reiterated that they are collaborating on another type of chip the two companies are working on together, the infrastructure processing unit, or IPU. since 2022. In a press release, Intel said this programmable accelerator is used to “offload networking, storage, and security functions from host CPUs.”

The IPU was the first of its kind when the companies first collaborated four years ago, Google told CNBC in an email. Google said it’s designed to help customers make better use of the main CPU in a traditional data center by taking on “upstream” tasks like routing network traffic, managing storage, encrypting data and running virtualization software.

For more than a decade, Google has also developed its own custom AI accelerator called a tensor processing unit, or TPU. In 2024, Google also began making its own custom CPU, Axion, opting for an Arm-based design over Intel’s leading x86 architecture.

— CNBC’s Kristina Partsinevelos contributed to this report.

WRISTWATCH: How advanced packaging became the next bottleneck in manufacturing AI chips