• March 26, 2026

This week CNBC published a feature story and video by journalist-producer Katie Tarasov about Arm’s new AGI CPU for datacenters. Meta is the flagship customer for this device, which is the first-ever in-house (though still fabless) chip produced by Arm. My analysis appears in both the article and the video. You can click the link below for the story and embedded video, or read on for an AI-generated, human-edited overview and summary.

Link to video.

Overview

Arm has launched its first in-house chip — the AGI CPU — which is designed for power-efficient datacenter applications in AI. Its debut customer is Meta, with support from other major tech companies including AWS, Google, Microsoft, and NVIDIA. The chip, optimized to eventually support artificial general intelligence (AGI), promises twice the performance per watt compared to traditional x86 architectures. To help develop and test the chip, Arm has built a new $71 million lab in Austin, Texas, where it has operated for more than 20 years and now employs over 1,000 people. Production for the chip, including manufacturing on TSMC’s 3nm node in Taiwan, is expected later this year. Arm’s move into physical silicon is seen as a significant shift in the market, with potential implications for competitors like Qualcomm and AMD.

Outline

Arm’s New In-House Chip and Its Implications

• Katie Tarasov introduces Arm’s new in-house CPU designed for power-efficient inference, with Meta as its initial customer.
• Mohamed Awad, Arm EVP for cloud AI, explains the CPU’s design for the AGI era, emphasizing its low power consumption and high performance.
• Patrick Moorhead expresses surprise at Meta’s embrace of Arm’s first chip, noting the difficulty of getting the first chip right.
• Tarasov describes the new chip lab in Austin, Texas, where Arm is optimizing the chip for customers like Meta, OpenAI, and SAP.

Arm’s Historical Context and Market Position

• Tarasov highlights Arm’s 35-year history as a chip design firm, selling its architecture to major chip makers and collecting royalties.
• Moorhead discusses the challenges of powering datacenters in the future and Arm’s expertise in low-power performance.
• Tarasov explores the shift in Arm’s business model from licensing designs to making physical silicon — and competing with some of its top customers.
• Moorhead contrasts Arm’s instruction set with the traditional x86 architecture, noting Arm’s approach to royalties and licenses.

Arm’s Rise and Market Impact

• Tarasov recounts Arm’s rapid growth from a small firm to a leading chip architecture, thanks to early deals with Apple and Texas Instruments.
• Awad mentions the success of Arm’s Neoverse platform, which has seen over 1.25 billion shipments in datacenter chips.
• Tarasov discusses Arm’s public offering in 2023, the largest in nearly two years, and its customizable instruction sets.
• Moorhead emphasizes the importance of Arm’s building blocks for chip makers like Amazon, Google, and Microsoft.

Arm’s Market Entry and Customer Reactions

• Tarasov explains Arm’s decision to enter the physical silicon market, hinted at in previous earnings calls and in the works for 18 months.
• Awad cites strong customer demand and successful silicon testing as reasons for launching now.
• Jensen Huang, CEO of NVIDIA, congratulates Arm on its first datacenter chip, and dozens of customers publicly endorse Arm’s new venture.
• Awad highlights the trillion-dollar market potential and the positive industry impact of Arm’s new chips.

Customer Support and Competitor Reactions

• Tarasov notes that at least eight customers have committed to purchasing Arm’s new GPU, with 50 existing customers showing support.
• Tarasov adds that not all customers have been vocally supportive, mentioning Qualcomm’s absence and ongoing legal disputes between Qualcomm and Arm.
• Tarasov inquires about AMD’s reaction, with Awad emphasizing the importance of both strong partnerships and competition.
• Forrest Norrod of AMD discusses the impact of competition on driving innovation and meeting customer requirements.

Arm’s Partnerships and Manufacturing Plans

• Tarasov inquires about Arm’s relationship with Intel, and Awad focuses on Intel’s IPU products based on Arm’s architecture.
• Awad mentions the potential for future partnerships with Intel.
• Tarasov asks about any plans to manufacture with TSMC in Arizona; Awad emphasizes customer needs as the primary consideration.

Arm’s Chip Testing and Optimization Process

• Tarasov describes Arm’s new $71 million lab in Austin, Texas, built in under two years to support chip testing and development.
• Awad explains the validation board in the Austin lab, which tests the signals of the new chips.
• Tarasov details the next stage of testing, where chips are placed in server racks for further evaluation.
• Awad describes the final stage of testing, involving automated tester equipment and failure analysis.

Power Efficiency and Market Potential

• Tarasov highlights the dense configuration of Arm’s new CPUs, appealing to datacenter customers worldwide.
• Awad discusses the power efficiency of Arm’s AGI chips, optimized for artificial general intelligence.
• Paul Saab of Meta emphasizes the importance of wattage in today’s infrastructure, making best-in-class CPUs crucial.
• Tarasov notes the shortage of memory and capacity in the AI chip market, with Awad also speaking about these challenges.
• Awad expresses confidence in securing supply chains and meeting demand through Arm’s forward planning.

Meta’s AI Infrastructure and Arm’s Role

• Tarasov explains Meta’s significant investment in AI infrastructure, making it a logical first customer for Arm’s new CPU.
• Saab of Meta discusses the benefits of having a diverse set of chip options, adding Arm to the ecosystem.
• Tarasov highlights Meta’s commitment to open hardware designs, reducing energy consumption and costs.
• Saab recounts the early conversations with Arm about making their new CPU available to the global market.

Meta’s Datacenter Plans and Arm’s Future

• Tarasov outlines Meta’s extensive datacenter plans, including sites in Louisiana, Ohio, and Indiana.
• Moorhead discusses Meta’s increased spending on AI and the need for more chips to power its data centers.
• Tarasov notes Meta’s new deals with Nvidia and AMD for GPUs and its own line of training and inference accelerators.
• Moorhead predicts significant revenue potential for Arm if it can capture even a small percentage of Meta’s AI spending.

Arm’s Global Market and Future Prospects

• Awad confirms that Arm’s new CPU is not export-controlled, allowing sales in China and other markets.
• Tarasov speculates about potential U.S. customers for Arm’s new CPU, with NVIDIA being a particularly interesting candidate.
• Awad emphasizes the strong partnership between Arm and NVIDIA, seeing potential for Arm’s new CPU in datacenter racks alongside NVIDIA’s products.
• Moorhead suggests that Huang of NVIDIA recognizes the need for more CPUs to sell more GPUs, making Arm’s new CPU a strategic fit.

Companies Mentioned

• Arm
• Meta
• Apple
• NVIDIA
• Amazon
• Google
• Microsoft
• OpenAI
• SAP
• Texas Instruments
• Qualcomm
• AMD
• Intel
• TSMC
• Samsung
• Broadcom
• Micron
• Oracle

Keyword Summary

Arm, Meta, Nvidia, AMD, Intel, TSMC, Neoverse, ai, agentic ai, artificial general intelligence, agi cpu, cpu, gpu, ai chips, data center, data center chips, x86, hyperscalers, export controls, data center power

Recent Posts