The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Short answer: The February 2025 report that Arm would make its own data-center chip was broadly borne out, but the product did not receive its formal public announcement until March 24, 2026. It is the Arm AGI CPU: a processor for AI data centers, designed to work alongside GPUs and other accelerators—not a direct replacement for Nvidia’s AI GPUs. The shift matters to investors because Arm is moving beyond its traditional licensing model and into a business that could bring more revenue per system, but could also complicate relationships with customers that design their own competing chips.
What the 2025 report said—and what was confirmed later
On February 13, 2025, reports based on information attributed to people familiar with the plans said that Arm, which is majority-owned by SoftBank Group, was preparing its own data-center chip and that Meta was an early customer. The product was described as a server CPU for cloud and AI computing. The reports said Arm might unveil it as early as summer 2025 and that manufacturing would be outsourced rather than carried out in Arm-owned factories. These were reported plans, not an Arm product announcement. Reuters reporting carried by Yahoo Finance and TechCrunch’s account of the report described the proposed change; The Information’s summary characterized the product as a server CPU rather than an AI accelerator aimed directly at Nvidia or AMD.
The public confirmation came later than the tentative 2025 timetable. On March 24, 2026, Arm announced the Arm AGI CPU as its first production silicon product and first Arm-designed data-center CPU. Meta was identified not just as an early customer but as Arm’s lead partner and co-developer. The companies said they intended the platform for broader AI infrastructure use and planned to release board and rack designs through the Open Compute Project later in 2026. Those announcements establish a product and a partnership; they do not by themselves establish broad commercial shipments, order volumes, or widespread adoption. Arm’s launch announcement and Meta’s announcement describe the partnership and plans.
| Date | What happened | What it means |
|---|---|---|
| 2016 | SoftBank acquired Arm. | Arm became a central part of SoftBank’s semiconductor strategy. |
| 2024 | SoftBank acquired Graphcore. | The acquisition added AI-chip design expertise to the group; it does not establish that Graphcore technology is in the Arm AGI CPU. |
| February 13, 2025 | Media reports described plans for an Arm server chip, with Meta as an early customer. | A reported plan, not a confirmed product launch. |
| March 2025 | SoftBank announced an agreement to acquire Ampere Computing for $6.5 billion. | A significant server-chip investment, but not proof that Ampere technology is part of Arm’s CPU. |
| March 24, 2026 | Arm announced the Arm AGI CPU; Meta announced a multi-generation partnership. | The first-party CPU strategy became public, with Meta as lead partner and co-developer. |
| Later in 2026 | Arm and Meta said they planned to release board and rack designs through the Open Compute Project. | A systems and ecosystem plan; it is not confirmation of general retail availability. |
Why Arm selling its own silicon is a strategic change
Arm’s traditional business has centered on licensing processor architectures and designs to other companies. Those customers develop or customize chips, arrange manufacturing, and sell processors or systems under their own brands. Arm earns revenue through licensing and royalties rather than generally selling its own branded production CPUs directly. The AGI CPU adds a different route: Arm can sell a designed product and potentially capture more value from a complete data-center platform.
#1 Best Overall
- High-performance foundation line, ARM Cortex-M4 core with DSP and FPU, 512 Kbytes Flash, 180 MHz CPU, ART Accelerator, Dual QSPI
- On-board ST-LINK/V2-1 debugger/programmer with SWD connector
- Can be powered from USB
- Three LEDs, Two Push-buttons
- Support of wide choice of Integrated Development Environments (IDEs) including IAR, ARM Keil, GCC-based IDEs
That opportunity comes with a conflict. Some companies that license Arm technology also develop their own processors. Arm’s 2025 Form 20-F discusses customers’ competing products and the potential implications of Arm’s own silicon strategy for its competitive position. Arm’s Form 20-F makes clear why neutrality is a business issue: customers may be less comfortable sharing roadmaps or relying on a supplier that competes for their system sales.
- Potential upside: Arm can participate more directly in demand for AI data-center infrastructure, offer a ready-designed platform, and help accelerate adoption of Arm-based systems.
- Potential cost: Cloud and chip customers could see Arm as a competitor as well as an IP supplier, potentially encouraging them to diversify or negotiate harder.
- Investor question: Can Arm build a valuable silicon business without weakening the broad licensing ecosystem that supports its existing revenue model?
The AGI CPU is not a GPU replacement
“AI chip” is a broad label. The confirmed Arm product is a CPU, designed for general-purpose processing in AI data centers. A CPU handles operating-system work, control flow, networking, storage tasks, orchestration, and other processing that supports AI jobs. GPUs and specialized accelerators are built to handle large volumes of parallel mathematical operations central to many model-training and inference workloads. AI servers commonly use CPUs and accelerators together.
Rank #2
- Ultra-low-power with FPU ARM Cortex-M4 MCU 80 MHz with 1 Mbyte Flash, LCD, USB OTG, DFSDM
- On-board ST-LINK/V2-1 debugger/programmer with SWD connector
- Can be powered from USB
- Three LEDs, Two Push-buttons
- Support of wide choice of Integrated Development Environments (IDEs) including IAR, ARM Keil, GCC-based IDEs
Arm describes the AGI CPU for tasks including accelerator management, agentic-AI orchestration, networking, and data-plane computing. Meta has said it plans to use the CPU alongside its own Meta Training and Inference Accelerator, or MTIA. This makes the product an attempt to strengthen the CPU and system side of AI infrastructure—not evidence that Arm has launched a competing Nvidia-style GPU. Arm’s technical announcement outlines its intended workload areas.
| Layer | Typical role | How the Arm AGI CPU fits |
|---|---|---|
| CPU | General-purpose processing, system control, orchestration, and data handling. | This is the Arm AGI CPU’s primary category. |
| Accelerator | Parallel computation for AI training and inference; may be a GPU or a purpose-built chip. | The Arm CPU is intended to work with accelerators, not replace them by definition. |
| Memory and interconnect | Move and store data among processors, accelerators, servers, and racks. | These components affect system performance; the CPU alone does not determine it. |
| Software | Operating systems, compilers, kernels, runtimes, and deployment tools. | Compatibility and optimization will matter to real deployments and workload results. |
| Rack and data center | Power, cooling, networking, density, and system-level operating cost. | Arm’s performance claim is framed at rack level, so configuration and measurement details matter. |
What Meta and SoftBank contribute
Meta is a development partner, not just a reported buyer
The 2025 coverage described Meta as an early customer. In 2026, Meta and Arm publicly described a partnership in which Meta is the lead partner and co-developer, with plans spanning multiple CPU generations. Meta says it expects to use the CPU alongside MTIA. That is stronger evidence of a concrete design relationship than the earlier customer report, but it does not disclose purchase volumes or prove adoption by other operators.
Rank #3
SoftBank is building a broader AI strategy
SoftBank’s stated strategy links Arm and semiconductor development with AI, data centers, and power infrastructure. Its annual-report materials describe semiconductor development anchored by Arm as a priority. SoftBank also announced an agreement to acquire Ampere Computing for $6.5 billion in 2025, and had acquired Graphcore in 2024. These moves show strategic interest in chip capabilities; they do not establish that Ampere or Graphcore designs are incorporated into the AGI CPU. See the SoftBank CFO message, SoftBank CEO message, and SoftBank’s 2025 annual report.
What Arm says about performance—and what remains unanswered
Arm claims the AGI CPU can deliver more than twice the performance per rack of the latest x86 platforms. This is Arm’s comparison, not an independently established result. It does not mean twice the single-thread speed, twice the AI-model throughput, twice the chip-level performance, or twice the performance of Nvidia GPUs. Arm’s SEC-filed exhibit contains the company’s claim.
Rank #4
- Mainstream Mixed signals MCUs ARM Cortex-M4 core with DSP and FPU, 512 Kbytes Flash, 72 MHz CPU, MPU, CCM, 12-bit ADC 5 MSPS, PGA, comparators
- On-board ST-LINK/V2-1 debugger/programmer with SWD connector
- Can be powered from USB.
- Three LEDs, Two Push-buttons
- Support of wide choice of Integrated Development Environments (IDEs) including IAR, ARM Keil, GCC-based IDEs
To assess the claim, infrastructure buyers and investors would need the comparison platform and workload, software versions, rack power and space assumptions, server counts, accelerator inclusion, memory and networking configuration, and latency or throughput results. Cost matters too: a performance-per-rack comparison does not by itself establish lower total cost of ownership. Until comparable independent results are available, treat the figure as a vendor claim rather than a universal performance measure.
Technical details reported by Tom’s Hardware include 136 cores, Arm Neoverse V3 cores, TSMC’s 3-nanometer process, and a 10U dual-node reference server compatible with the Open Compute Project’s DC-MHS standard. These are secondary-source specifications; Arm’s official launch and technical materials should be treated as the authority for its positioning and confirmed product claims. Tom’s Hardware’s coverage reports those details.
Best Value
- STM32F103C8T6 ARM STM32 minimum system development module.
- ST-Link V2 support the full range of STM32 SWD interface debugging, simple interface (including power supply), 4 line speed, stable work.
- Use the current smart phones of Mirco USB interface, easy to use, USB communication and power supply can be done.
- The board lead to all the I/O resources.Download with SWD debug interface, which requires a minimum of 3 wires to complete debug a download task
Who could use it, and what would determine adoption?
The likely audience is large infrastructure operators: hyperscalers, cloud providers, AI developers, server integrators, and enterprises with substantial data-center needs. It is not a consumer processor that an individual can install in a desktop. Arm’s announcement of availability to the broader AI ecosystem and planned reference designs signals an intended route beyond Meta, but the announcements cited here do not establish a public price, general ordering channel, shipment scale, or broad cloud availability.
Potential buyers would weigh the whole platform rather than the processor name alone:
- Performance per watt and per rack for their own workloads.
- Total system cost, including memory, networking, cooling, and support.
- Compatibility with existing Arm software, compilers, and operating systems.
- Memory capacity and bandwidth, interconnects, and accelerator compatibility.
- Availability through server manufacturers or cloud providers and confidence in supply.
- Long-term product support and whether the platform roadmap fits their deployments.
Alternatives occupy different parts of the stack. AWS Graviton, Google Axion, and Microsoft Cobalt are custom Arm-based cloud CPU offerings within their respective cloud environments. AMD combines EPYC CPUs and Instinct accelerators; Nvidia sells accelerated-computing platforms; Meta’s MTIA is an accelerator designed for Meta’s workloads. Ampere sells Arm server processors. These are not interchangeable products: some are cloud instances, some are processors, and some are accelerator or full-system offerings. A buyer should compare the workload and deployment model, not just the word “AI.”
The risks that matter to investors and infrastructure buyers
- Customer trust: Arm’s licensing customers may object to competing with their supplier, especially if the new business targets markets they serve.
- Execution: Designing a CPU is not the same as building a scaled product business. Validation, systems support, manufacturing coordination, and supply assurance all matter.
- Software: Real performance depends on optimized software and compatibility, not architecture claims alone.
- Competition: Hyperscalers may prefer in-house processors for control, while established server CPUs and accelerator platforms already have ecosystems.
- Evidence of adoption: A launch partner and ecosystem interest are not the same as large shipments, recurring revenue, or broad customer deployment.
- Manufacturing: Arm-designed does not mean Arm-manufactured. Outsourced production leaves supply dependent on manufacturing partners and capacity.
For Arm’s investors, the pivot offers a possible new growth avenue but adds execution and relationship risk to a business whose value has long depended on licensing broadly across the industry. For infrastructure buyers, the more immediate question is practical: whether a complete Arm platform is available, supported, and measurably better for a particular workload and budget.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Clear out junk files and repair common Windows errorsFree Scan →What the 2025 headline got right—and what it did not
The 2025 report correctly pointed toward Arm’s move into its own data-center silicon and identified Meta’s importance early. Its projected summer 2025 unveiling was not the formal public launch: Arm announced the AGI CPU in March 2026. And “AI chips” can mislead if read as a GPU claim. The confirmed product is best understood as an AI-oriented data-center CPU and systems effort, with commercial scale, independent performance, and the longer-term effect on Arm’s customer relationships still to be demonstrated.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
