Arm Enters Chip Manufacturing: How a 136-Core AI CPU is Disrupting Intel and AMD’s Data Center Dominance

The chip industry just got a lot more interesting.

On March 24, 2026, Arm Holdings plc (Arm) made a move that could reshape the entire data center landscape—they announced the start of sales for their first mass-produced, self-designed chip products.

For over 30 years, Arm has been the invisible backbone of computing, licensing their processor architecture to everyone else.
Now they’re stepping into manufacturing.
And they’re going after the trillion-dollar AI market.

—

Meet the Arm AGI CPU: The New Contender in Data Center Chips

At an event in San Francisco, Arm unveiled the “Arm AGI CPU”—a processor specifically engineered for agentic AI workloads inside data centers (shuju zhongxin 数据中心).

Here’s what makes it noteworthy:

• Up to 136 Arm Neoverse V3 cores per chip
• ~300 watts power consumption
• Built for single-rack performance optimization and efficiency at scale
• Arm calls it the “world’s most efficient agentic CPU”

To put this in perspective: data center chips are a completely different beast than smartphone processors.
They’re designed to handle massive computational loads, and that means they command massive price tags—we’re talking ¥140,000 RMB ($20,000 USD) for high-end server chips.

That’s a huge jump from the $50-100 you’d pay for a premium mobile processor.

—

Why Arm is Abandoning Its Licensing-Only Business Model

This is actually a pretty radical shift for Arm.

The company has built its entire empire on Intellectual Property (IP) licensing—designing chip architectures and then selling the blueprints to other companies who manufacture the actual silicon.
Apple (Pingguo 苹果), Qualcomm (Gaotong 高通), Samsung (Sanxing 三星)—they all use Arm designs.

But now, Arm is expanding into Compute Subsystems (CSS) and independently designed chip products.

Why the change?

The AI revolution is forcing their hand.

Their ecosystem partners are demanding new ways to deploy Arm technology at massive scale.
And frankly, the money is better in manufacturing than licensing.

Rene Haas, Arm’s CEO, explained the strategic reasoning:

“AI is fundamentally redefining how computing is built and deployed. Agentic AI is accelerating this transformation, and today marks the next phase of the Arm computing platform—a significant moment in our company’s journey.”

Translation: The old model wasn’t enough anymore.
The market is moving too fast.

—

Meta is the First Major Customer (And That Tells You Everything)

• Task Orchestration: Managing the flow of complex AI agent operations.
• Data Preparation: Cleaning and structuring data before it hits the high-power GPUs.
• Query Response: Handling the final output and interaction logic for AI users.
• Infrastructure Symbiosis: Working in tandem with Meta’s custom MTIA hardware.

Here’s the real validation: Meta (Meita 美塔) is the primary development partner and first major customer for the Arm AGI CPU.

That’s not some small startup betting on Arm.
That’s one of the world’s largest cloud infrastructure operators with massive AI ambitions.

What does Meta plan to do with it?

• Optimize infrastructure across their applications
• Work alongside Meta’s custom chip, MTIA (Meta Training and Inference Accelerator)
• Handle task orchestration, data preparation, and AI query responses at scale

Think of it this way: Nvidia (Yingweida 英伟达) GPUs do the heavy lifting on AI inference.
The Arm AGI CPU handles the orchestration and preparation work.
Together, they create a more efficient pipeline.

And Meta isn’t alone.
Other partners already lined up include:

• Cerebras
• Cloudflare
• F5
• OpenAI
• Positron
• Rebellions
• SAP
• SK Telecom

That’s an impressive roster for a brand-new product line.

—

The Real Winner: Much Higher Profit Margins

Let’s talk about the money, because that’s why this matters.

Smartphone chips? They sell for dozens of dollars, even the premium ones.
Data center chips? Tens of thousands of dollars.

Rene Haas sees the opportunity clearly: by moving from IP licensing to chip manufacturing, Arm can capture significantly higher margins on each unit sold.

It’s not about volume anymore.
It’s about selling fewer chips at massively higher prices.

—

The Competitive Threat: Intel and AMD Should Be Worried

Arm’s move is a direct threat to Intel (Yingte’er 英特尔) and AMD’s data center business.

Both companies have built their empires on the x86 architecture—the instruction set that’s dominated servers and data centers for decades.
But x86 isn’t optimized for AI.
It’s power-hungry and complex.

Arm’s approach is different: simpler, leaner, more efficient.

Haas made it clear that Arm views this as a market expansion play, not a zero-sum game:

“This market is large enough to accommodate multiple players.”

Translation: There’s room for Arm, Intel, AMD, and others.
But each taking share from traditional x86 players.

—

One Complication: Arm’s Existing Clients Are Self-Designing Chips Too

Here’s where things get tricky.

Many of Arm’s biggest customers—including Meta—have their own chip development programs.
Almost all of them license Arm’s architecture to build their custom silicon.

Now Arm is selling directly competing products.

It’s not necessarily a dealbreaker (Meta is still working with them), but it creates some awkward dynamics.

These companies now have three options:

• License Arm’s IP and design their own chips (what most do now)
• Buy Arm’s pre-designed AGI CPU
• Some combination of both (most likely)

From Arm’s perspective, offering all three options is actually a feature, not a bug.
They get revenue either way.

—

The Nvidia Factor: They’re Entering This Market Too

Nvidia has also launched a new CPU line, entering the same space Arm is targeting.

Haas clarified that Arm’s chips are aimed at different market segments than Nvidia’s latest offerings—but let’s be real, that’s what you’d say publicly.

The reality is: the AI infrastructure market is heating up, and everyone wants a piece.

—

What This Means for the Industry

Arm entering chip manufacturing is a watershed moment.
It signals a fundamental shift in how computing infrastructure is being built for AI.

The takeaways:

• AI is accelerating architecture shifts.
  Traditional x86 isn’t built for modern AI workloads.
• Efficiency matters more than raw power.
  The Arm AGI CPU’s focus on power consumption at scale reflects real customer priorities.
• Vertical integration is the trend.
  Every major player (Meta, Cerebras, OpenAI) wants control over their own silicon.
• The data center chip market is fragmenting.
  Gone are the days when Intel and x86 had a monopoly.
• IP licensing alone isn’t enough anymore.
  Companies want options—and they want them fast.

For investors and founders, this is a reminder: markets don’t stay static.
Thirty years of dominance doesn’t guarantee thirty more.
And when the architecture of computing fundamentally changes, the winners and losers can shift pretty quickly.

Arm entering chip manufacturing isn’t just business news.
It’s a sign of how serious the AI transformation has become.

—

References

• Arm Leads the Charge! 136-Core AI CPU Pierces the Heartland of Intel and AMD – Cailian Press (Cailian She 财联社)
• News and Press Releases – Arm
• Meta Training and Inference Accelerator (MTIA) – Meta
• Intel Xeon Data Center Processors – Intel