Tech Explained: Here’s a simplified explanation of the latest technology update around Tech Explained: Arm’s new CPU represents a major shift for the AI data center market – what does it mean for UK tech? in Simple Termsand what it means for users..
Arm has lifted the lid on its first ever in-house chip, in a move that could cement the semiconductor pioneer’s place in AI data centers and reshape the UK’s role in the AI ecosystem.
The Arm AGI CPU isn’t competing with the raw computational power of graphics processing units (GPUs), provided by the likes of AMD and Nvidia. Instead, the chip acts as an orchestrator, scheduling actions taken by AI agents, managing the operations of AI accelerators, and overseeing memory and storage systems within a data center.
CPUs, it seems, are back in the spotlight.
While 2023 and 2024 were characterized by immense demand for GPUs, subsequent years have seen attention fall on more specialized chips. Application-specific integrated circuits (ASICs) are all the rage: Google has tensor processing units (TPUs), Amazon has Trainium and Inferentia, Microsoft has Maia, and Meta has Meta Training and Inference Accelerator (MTIA) accelerators.
It makes sense for hyperscalers to adopt these chips in addition to (or instead of) third-party GPUs for more efficient AI inference at scale. Google, for example, optimized TPU v7 Ironwood for the best cost per token with its Gemini AI models and to unlock massive throughput improvements.
Arm, however, isn’t wading into this market. Instead, it’s carved out a spot perpendicular to the likes of GPUs and ASICs, with a CPU designed to act as the pace-keeper for AI agent workloads, no matter the hardware they use for raw compute.
Indeed, the firm is leaning into a rising focus on CPUs, which are ideal for the sequential processing necessary for orchestrating workloads. Arm has long acted as the design powerhouse behind the world’s mobile chip architectures, which it licenses to the likes of Apple, Google, Microsoft, Nvidia, Samsung, and Intel.
The potential financial uplift is clear. Arm is forecasting $15 billion in annual revenue in the next five years, according to Reuters. As a UK-headquartered firm, a good portion of this revenue will be reflected in UK tax, payroll, and further investment.
Arm AGI CPU specs
Arm said the AGI CPU has been optimized for high-performance, massively parallel agentic workloads run in dense server racks.
To this end, the chip is equipped with up to 136 of Arm’s Neoverse V3 cores for a frequency of 3.7Ghz, with 2MB L2 cache per core and memory latency below 100ns.
Arm’s reference server configuration for the AGI CPU is a 1OU, 2-node design with two AGI CPUs per blade. Scaled up to a 36kW rack containing 30 blades, this allows for 8,160 Neoverse V3 cores per rack.
The firm has also collaborated with Supermicro on a liquid-cooled, 200kW rack design that can contain 336 AGI CPUs, for a total of more than 45,000 cores. Arm said this configuration delivers an estimated 2x performance uplift in comparison to the latest x86 options.
It will compete directly with Nvidia’s Arm-based Vera CPU, which also runs the orchestration layer for large-scale AI agent deployment. Arm says its chip delivers 6GB/sec of memory bandwidth per chip, while Nvidia’s boasts 14GB/sec. The former is packed with 12-channel, DDR5 memory versus the latter’s LPDDR5X.
Ultimately, however, this is about more than specs on paper. Arm’s advantage is its open design: with compute express link (CXL) 3.0 Type 3 interconnect and DDR5 memory, the AGI CPU can be paired with a far wider range of hardware than the Vera CPU, which is optimized for Nvidia systems.
Arm has also rejected multithreading, in which CPU cores execute multiple threads at the same time. Mohamed Awad, EVP of cloud AI at Arm, told The Register this was to keep performance deterministic.
In other words, Arm is betting that its customers will benefit more from predictable, sustained performance when it comes to orchestrating AI agents rather than from the peaks and troughs of performance unlocked through multithreading. Given the immediate customer interest, we’ll see that theory put to the test soon.
Partners already in place
In advance of the announcement, Arm secured a major commitment with Meta to deploy the new chip. Describing Meta as its “lead partner and customer”, Arm said the company is looking to use the AGI CPU to support its gigawatt-scale data centers alongside MTIA accelerators.
This could be seen as a significant vote of confidence, given Meta is scrambling to catch up with competitors in the AI space. In choosing Arm’s solution, in parallel with its own hardware, Meta has lined up billions in revenue for Arm while strengthening the chip designer’s arguments that its product is up to the task of handling AI agents at the hyperscaler level.
We’ll have to wait and see for more financial details on the deal, in order to assess the price Arm puts on its chip and the rate Meta was willing to pay to shore up its data centers.
Similarly, OpenAI will use the chip for future projects.
“OpenAI runs AI systems at massive scale,” said Sachin Katti, head of Industrial Compute at OpenAI.
“Hundreds of millions use ChatGPT every day, businesses build on our API, and developers rely on tools like Codex. The Arm AGI CPU will play an important role in our infrastructure as we scale, strengthening the orchestration layer that coordinates large scale AI workloads and improving efficiency, performance, and bandwidth across the system.”
It’s unclear at time of writing if the AGI CPU could play a role in Stargate UK, OpenAI’s tie up with data center operator Nscale. Other major customers noted in the announcement include Cloudflare, SAP, and SK Telecom, which represent demand for the chip beyond agentic AI use cases.
“To continue our mission of helping build a better Internet, Cloudflare needs infrastructure that scales efficiently across our global network,” said Stephanie Cohen, chief strategy officer at Cloudflare.
“The Arm AGI CPU provides high-performance, energy-efficient compute designed for the next generation of workloads.”
UK tech on the rise
The extent to which Arm is a UK success story is up for debate. It’s beyond doubt that the nearly 3,500 workers at its Cambridge HQ have played a major role in getting products like the AGI CPU over the finish line, though the firm also has approximately 5,500 workers in other regions.
In her remarks at the 2026 Mais Lecture, chancellor Rachel Reeves paid tribute to Arm, alongside Google DeepMind and the autonomous driving firm Wayve, as examples of British deep tech success.
Only one of these companies – Wayve – remains British owned, however.
In 2016, SoftBank acquired Arm for £24 billion, and while the investment firm promised to keep the company’s HQ in the UK, Arm co-founder Hermann Hauser described the deal as a “a sad day for technology in Britain”.
2023 saw Arm listed itself on the US rather than UK stock market. At the time Russ Shaw, founder of Tech London Advocates and Global Tech Advocates, told the BBC that the decision was “a significant blow to the UK tech sector”.
Nevertheless, there’s no doubt that rising revenue at Arm will be felt in the UK tech ecosystem. To date, Arm has carved its place through design brilliance and an open-ended approach that has made it an invaluable supplier for the world’s mobile market.
With a direct route to increased revenue in the AI market now secured, it’s likely to retain its ‘crown jewel’ status in Cambridge and become more integral still to the world’s hyperscalers.