Arm Holdings (ARM) Q4 FY26 earnings call: guidance and Q&A

Where did the incremental AGI CPU demand come from over six weeks, and how will Arm secure the supply while keeping guidance unchanged?

Demand is a mix of launch customers raising forecasts and new customers ready to deploy, aided by finished-rack availability from Supermicro/Lenovo/ASRock and completed software work. Supply for $1B (memory, wafers, packaging, test) is in place; teams are working to secure supply for the $2B. Jason held the ~$90M Q4 FY27 revenue target unchanged pending supply-chain progress.

“for the $2 billion, we are now in the process of securing supply to support that. And the teams are working around the clock to make sure we can find the right answers for our customers.”

asked by Andrew Gardiner · answered by Rene Haas

How should royalty growth trend across FY27 given data-center strength versus consumer/smartphone weakness?

Q4 slowed on a tough MediaTek comp; guidance returns royalty to ~20% by Q1. Mobile units may turn slightly negative, concentrated at the low end, so limited impact; any weakness is more than offset by data center strength across NVIDIA (Vera/Grace), Google (TPU + Axion 2) and AWS (Graviton). Automotive continues double-digit growth with share gains.

“any of the weakness that we see in memory and mobile, we do expect that to be largely offset plus, I would say, on the cloud side.”

asked by Joseph Quatrochi · answered by Jason Child

With AMD, Intel and captive programs all claiming share, where does Arm actually take share to grow?

Rene noted Arm was first to cite a $100B+ TAM and the market is now catching up ($120B possible). Growth comes from rising cores-per-CPU (AGI CPU at 136 cores) where efficiency-per-core favors Arm, from accelerator head nodes (TPU, Trainium, Rubin) trending to 100% Arm, and from customers (Cloudflare, Meta, SAP, SK Telecom, OpenAI) who won't design their own CPU. He expects Arm to hold the largest CPU-type share by end of decade.

“I'm actually confident that by the end of the decade, I believe the largest market share by CPU type will be Arm.”

asked by Vivek Arya · answered by Rene Haas

Is a ~100% Arm attach to accelerators expected, what support OpEx does the merchant model add, and how do royalty and OpEx flow through FY27?

Rene expects the vast majority of accelerator CPU share to be Arm over time (NVIDIA already, Graviton/TPU trending there), citing Google's 80% performance improvement. Support (firmware, BootROM, hardware issues) is Arm's responsibility but OpEx is already baked into long-term guidance. Jason guided FY27 royalty and license both to ~20%, license back-end weighted (~60% H2), and OpEx growing a few percent sequentially with incremental margin returning by year-end.

“the OpEx for support, that was baked in the numbers that we shared at Arm Everywhere and embedded within our long-term guidance and so already accounted for.”

asked by Timm Schulze-Melander · answered by Jason Child

When does the CPU:GPU ratio cross 1:1, and how does the head-node opportunity differ from the host node?

Rene reframed from chip count to core count: GPUs are reticle-limited so chip ratios may not shift much, but CPU cores (136 today, potentially 500) will scale sharply. Growth is less in the fixed head node and more in dedicated CPU racks for agentic orchestration — citing NVIDIA's 256-Vera, 200kW liquid-cooled rack — and management believes it has undercalled the CPU demand transition beyond the stated 4x.

“we probably have undercalled the CPU demand in terms of the transition here. We talked about a 4x increase. We could get our heads around a bigger number than that.”

asked by Sreekrishnan Sankarnarayanan · answered by Rene Haas

How have large IP customers who also sell Arm CPUs reacted to Arm selling its own silicon?

Rene said Arm briefed the ecosystem early — chip builders, EDA, software community and licensees (AWS, Microsoft, Google, NVIDIA) — and every partner asked for support at Arm Everywhere said yes, with 50+ providing quotes, references or videos. The strategy is framed as customer-driven and additive, since more Arm-optimized software strengthens everyone; Arm is 'sold out' with demand exceeding supply.

“we're sold out and we've got people looking for more products, there's demand for these products. And at the end of the day, customer demand speaks volumes.”

asked by Sebastien Cyrus Naji · answered by Rene Haas

How fast are data center royalties growing into FY27, and is licensing still high-single-digit growth?

Rene confirmed Neoverse-based royalties doubled YoY and are expected to double again in FY27, running in tandem with (not cannibalizing) the AGI CPU business toward $15B AGI + $10B IP by FY31. Jason guided FY27 license to the 20s percent range, with a long-term floor of at least 10% and high-single/low-double-digit as the through-cycle target.

“we have 2 extremely strong sources of revenue that I think the best way to think about it is don't cannibalize each other. They're going to run in tandem with each other.”

asked by Vijay Rakesh · answered by Rene Haas

What OpEx supports the chip business this year and next, and when does it become accretive to earnings?

Jason said the AGI CPU chip business leverages the CSS compute die already funded by the IP business, so incremental OpEx is small — a team in the dozens, not hundreds — making it operating-profit positive next year. By FY31 he expects the IP business at ~65% operating/EBITDA margin and the chip business at ~35%, with timing dependent on revenue growth. The analyst referenced ~30%+ first-gen gross margin and a ~$3B chip-business OpEx endpoint by calendar 2030.

“the incremental costs and OpEx that we're having to add to the chip business is not that significant. It's a team that's in the -- probably in the dozens of people, not hundreds.”

asked by Harlan Sur · answered by Jason Child

Does each agentic flow map to one core, or should it be modeled as Arm instructions per token generated?

Rene favored the simpler single-core-per-agent framing: an asynchronous agent job (schedule, wait, pause) runs more power-efficiently on a single core than across multiple, so more cores means more concurrent batches. This drives increasingly large core counts and higher ASPs, underpinning the $100B-$125B TAM — a per-core-best-job model rather than multi-instruction-across-cores.

“our viewpoint is very much one of more cores is better. And that's why I think you're going to see increasingly larger core counts in these CPU chips.”

asked by Lee Simpson · answered by Rene Haas