Arm AGI CPU wins China customers, targets 43% server share

Arm's first production silicon is entering mainland China data centers through Volcengine, Lenovo, and Unigroup, as JPMorgan projects Arm server CPU share reaching 43 percent by 2028 — a direct challenge to Intel and AMD's decades-long x86 dominance.

"Customers kept asking us to go one step further and provide products closer to finished chips," Ami Badani, chief marketing officer at Arm, said at the Sept. 8 Arm Everywhere China event in Shanghai. "This is a main reason we launched the AGI CPU."

The AGI CPU packs 136 Neoverse V3 cores across two 3nm dies at 300 watts, with more than 800 GB/s of memory bandwidth and CXL 3.0 support. Arm claims more than 2x rack-level throughput versus x86 platforms, though independent benchmarks have not been published. The company also unveiled Neoverse CSS N4, its most configurable cloud CSS generation, supporting up to 128 cores per die with LPDDR6 and PCIe Gen 7.

Arm reported FY2027 Q1 revenue of $1.289 billion, up 22.4 percent year over year, with a 41.2 percent non-GAAP operating margin. The company targets roughly $1 billion in AGI CPU chip revenue by 2028, scaling to $15 billion by fiscal 2031, against a server CPU TAM projected at $100 billion by decade's end.

Three product tiers, one competitive front

Arm now offers three product depths: IP licenses for customers with full chip design capability, CSS for those seeking to reduce integration work, and the AGI CPU for rapid deployment. This structure widens Arm's addressable market but also creates direct competition with its own licensees — AWS's Graviton, Google's Axion, Microsoft's Cobalt, and Nvidia's Vera are all Arm-based processors that now face a first-party rival.

Badani framed the AGI CPU as market expansion rather than encroachment, noting that IP licensing reaches AWS but does not naturally extend to all of SAP's on-premises deployments. The AGI CPU provides a direct path into those environments.

The mobile business that built Arm — smartphone application processors still contribute roughly 45 percent of royalty revenue at 99 percent market share — has limited headroom. That constraint pushed Arm toward the data center, where the combined TAM for Edge AI, Physical AI, and Cloud AI is projected to grow from about $535 billion today to more than $1.5 trillion by fiscal 2031.

China's data center pipeline opens

Volcengine plans to integrate the AGI CPU into its data centers for next-generation AI services, including an agentic sandbox solution. Eddie Ramirez, Arm's vice president of cloud AI business extension, said the two companies are working closely on deployment, with timing to be disclosed by Volcengine. Lenovo is building AI infrastructure around the AGI CPU, while the partnership with New Unigroup covers both AGI CPU and CSS, extending to agent orchestration, CXL memory pooling, and rack-level system architecture.

The Chinese adoption comes as agentic AI workloads reshape CPU demand. Arm management projects CPU-to-GPU ratios moving toward 1:1 as autonomous agents require sustained CPU-side processing. Research from Georgia Tech and Intel shows CPU-side tool processing accounts for up to 90.6 percent of total latency in representative agentic workloads. JPMorgan estimates agentic AI CPU shipments growing from 1.6 million units in 2025 to 27 million by 2028, a 155 percent compound annual growth rate.

The competitive pressure on x86 is mounting from multiple directions. AMD held roughly 44 percent of server CPU share in Q1 2026, according to JPMorgan, while Arm's share is projected to nearly double from 22 percent in 2025 to 43 percent by 2028. Even with agentic AI expanding the overall market, Intel and AMD face structural share erosion as Arm's reach — spanning hyperscaler custom silicon, CSS, and now first-party AGI CPUs — continues to grow.

Arm shares surged more than 17 percent on March 25 following the AGI CPU launch, adding roughly $20 billion to market capitalization. The stock's valuation now reflects expectations that the company can convert its architecture dominance into a direct silicon business. Whether that conversion succeeds depends on execution: Arm has no track record shipping production silicon at scale, and its 2x rack-density claims remain unverified by independent testing. The next 12 to 18 months, as Meta, OpenAI, and Chinese customers move from commitments to production telemetry, will determine whether the AGI CPU becomes a $15 billion business or a well-marketed reference design.

This article is for informational purposes only and does not constitute investment advice.