The compute substrate on both ends of the agent.
Worked backwards from Arm's real business — the architecture under roughly 99% of smartphones and, increasingly, the data center, where Graviton, Grace, and Axion are all Arm Neoverse. Arm is the compute substrate on both ends of the 🤫 thesis: the devices a private, on-device agent runs on, and the power-efficient CPUs behind the edge and garage grid. Performance-per-watt is your moat — and the literal enabler of a low-cost-per-watt AI factory. This is our proposal to make the private agent an Arm-native workload, device to data center.
Written from public information. A proposal in pursuit — not a claim of partnership, endorsement, or investment by Arm.
Why Arm × 🤫 is obvious — in one look.
Roughly 99% of the devices our agent runs on are Arm, and KleidiAI accelerates on-device AI with no code changes.
We integrate KleidiAI + SME2 into Agent One's on-device inference path.
A private, on-device agent accelerated across every Arm phone and PC — royalties on each.
Graviton, Grace, and Axion are all Neoverse — the energy-efficiency leaders for the AI grid.
Our garage and warehouse grid standardizes its CPU tier on Arm Neoverse.
Per-chip royalties across our device and grid footprint alike.
You just moved up the value chain into your own silicon — the Arm AGI CPU for agentic AI infrastructure.
We're a real, private-agent reference workload for exactly that market.
A design partner for agentic AI on Arm, from device to data center.
How Arm wins, honestly.
Arm designs CPU architectures and IP and earns license fees plus per-chip royalties that recur for the life of a design — a uniquely durable model. Every figure is public (Arm FY2026 filings, year ended March 31, 2026) and approximate.
Up 23% YoY — a third consecutive year of >20% growth, split roughly 53% royalty / 47% license.
Cumulative Arm-based chips — the largest install base of any CPU architecture.
Industry estimate — Apple, Qualcomm, MediaTek, and Samsung SoCs are all Arm-based.
Arm's claim — Neoverse-based CPUs (Graviton, Grace, Axion) are the perf-per-watt leaders in the cloud.
Arm's first in-house production silicon (announced Mar 2026), for agentic-AI data-center infrastructure.
Scalable Matrix Extension 2, delivered via Lumex + KleidiAI — Arm's on-device AI acceleration path.
Ubiquity, efficiency, and one architecture end to end.
Arm's strengths are reach and power efficiency across every tier. These are the ground a private agent runs on.
One ISA, phone to data center
The same architecture family spans smartphones, edge, automotive, and now cloud servers — one software target end to end.
Power efficiency
The structural moat behind mobile dominance and the data-center pitch — performance-per-watt, the literal enabler of a low-cost-per-watt grid.
The largest ecosystem
The biggest developer and software ecosystem of any CPU architecture — software written for Arm runs across the fleet.
Compute Subsystems (CSS)
Pre-integrated, validated subsystems that shorten time-to-silicon and carry higher royalty rates — a growth driver in FY2026.
Neoverse in the cloud
Graviton, Grace, and Axion are all Neoverse — Arm now claims ~50% of top-hyperscaler CPUs and near-100% of data-center DPUs.
Armv9 + SME2
The current architecture generation with on-device matrix acceleration — higher per-chip royalties and real AI performance.
Lumex, KleidiAI, and the Arm AGI CPU.
What Arm shipped or announced most recently — the surfaces a partnership plugs into. Stated honestly, shipped vs. announced.
Lumex CSS (on-device AI)
Announced Sept 2025 — Armv9.3-A C1 CPUs, SME2, and the Mali G1 GPU; up to ~5x faster on-device AI and ~3x better energy efficiency vs prior gen.
KleidiAI
Libraries that bring SME2 acceleration to apps with no code changes — integrated into PyTorch ExecuTorch, Google LiteRT, and Microsoft ONNX Runtime.
Arm AGI CPU
Arm's first in-house production silicon (announced Mar 2026, broader availability H2 2026), aimed at agentic-AI data-center infrastructure — lead partner Meta.
Neoverse V-series
The cores behind Graviton5, Grace, and Axion — the performance-per-watt CPU tier for the AI factory, with NVLink Fusion integration.
What Arm is building toward.
AI everywhere on Arm
Haas: agentic computing is the next phase — Arm's foundation of high-performance, power-efficient computing at global scale.
Data-center Neoverse growth
Neoverse royalties doubling year-on-year, with ~50% hyperscaler CPU share and near-100% of data-center networking.
Up the value chain into silicon
The AGI CPU — expanding from IP and subsystems to Arm-designed production chips for agentic AI.
On-device AI acceleration
Lumex + SME2 + KleidiAI — real-time assistants, translation, and personalization on-device, with no per-app porting.
Efficiency leadership
Performance-per-watt as the through-line — from the phone edge to the server rack.
The people carrying it.
Our respectful reading of public roles and priorities — so any partnership makes each of their jobs easier, not harder.
Rene Haas
CEO
AI everywhere on Arm and the move into production silicon — power-efficient computing for the agentic era.
Jason Child
EVP & CFO
The royalty + license model and the economics of the new AGI CPU business.
Richard Grisenthwaite
EVP & Chief Architect
The architecture roadmap — Armv9, SME2, and the compute foundation an agent runs on.
Chris Bergey
EVP, Edge AI
Mobile and edge — Lumex, KleidiAI, and on-device AI: where the private agent actually runs.
Mohamed Awad
EVP, Cloud AI
Data center, Neoverse, and the AGI CPU — the grid tier of the partnership.
Buy together. Build together. Sell together. Take risk together.
Arm is the compute substrate on both ends of the 🤫 thesis — the devices the on-device agent runs on, and the efficient CPUs behind the grid. Every party is better off, and no one extracts from the human.
Build together — KleidiAI in the agent
SME2 + KleidiAI accelerate on-device AI across Arm phones and Copilot+ PCs, with no code changes.
Integrate KleidiAI + SME2 into Agent One's on-device inference so the private agent runs accelerated, on-device, with no per-device porting.
A provably on-device, consent-first agent; Arm earns royalties on every device it runs on.
Build together — the grid on Neoverse
Graviton, Grace, and Axion are all Neoverse — the efficiency leaders.
Standardize the 🤫 Puppy / Warehouse One CPU tier on Arm Neoverse — one software target across device and data center.
Arm takes load off the sold-out cloud grid cheaply; we get the most efficient CPU tier.
Buy together — a royalty-bearing Arm footprint
Arm earns a per-chip royalty for the life of every design that ships on its architecture.
Commit 🤫's device and grid procurement to Arm-based silicon — Lumex-class SoCs on-device, Neoverse CPUs in the grid — so every 🤫 node ships as a royalty-bearing Arm design.
Arm books recurring royalties across the whole 🤫 footprint; we buy one architecture, device to data center.
Sell together — power-efficient private AI on Arm
Arm's narrative is edge AI and cloud AI on one efficient architecture.
Co-market 'private, power-efficient edge AI on Arm' — Puppy One nodes on Neoverse, Agent One accelerated on Lumex-class silicon.
A joint story that serves both companies' edge and cloud AI narratives.
Take risk together — the agentic-AI reference
The AGI CPU targets agentic-AI infrastructure — Arm's stated new market.
Be an early design partner with a real private-agent reference workload across device and grid.
Arm gets a genuine agentic workload and royalties across our whole footprint; we get first-class acceleration.
Performance-per-watt — the spine of the AI factory.
Arm's power efficiency is the literal enabler of a low-cost-per-watt edge grid — the spine of the AI-factory thesis, from the phone in the person's hand to the server in the garage. The whole stack runs on one architecture, owner-operated. And 🤫 Agent One is free for life to every US federal and state government employee and every past or present service member who helps build it.
Buy together. Build together. Sell together.
The best business is fun when there is real money to be made and real value created for the people we both serve. That is the only partnership we will build — and we would be proud to build it with Arm.
Sources.
What this is, and isn't.
This is our proposal, assembled from public information — not a claim of partnership, endorsement, sponsorship, or investment by Arm. Arm is not currently a partner of HushOne or Hushh Technologies. The '~99% of smartphones' figure is a widely-reported industry estimate, not an Arm-published number; the '~50% hyperscaler CPU share' is Arm's own claim; the AGI CPU is announced, not shipping; AGI-CPU revenue figures are management forecasts, not booked revenue. Arm, Neoverse, Cortex, Lumex, and related marks are trademarks of Arm Limited. Our own certifications are 'in pursuit' until a 3PAO assessment or agency ATO says otherwise.