ARM: Short-term swings can't mask its edge; AI core underpins a long-term premium?

ARM (ARM.O) released its FY2026 Q4 results (quarter ended Mar 2026) after the U.S. close on May 7 Beijing time. Key takeaways are as follows:

1) Headline numbers:$Arm(ARM.US) posted revenue of $1.49bn (+20% YoY), slightly above consensus ($1.47bn). The ~$250mn QoQ uptick was largely driven by License growth, and GPM improved to 97.9%.

2) Biz. mix: License vs. Royalty roughly 1:1

a) License revenue was $820mn, up ~$320mn QoQ, continuing its upward trajectory. Fully paid-up license customers increased by 6 to 56, taking total customers to 385.

b) Royalty revenue was $670mn, up 10% YoY, led by data center, edge AI (handsets) and auto.

Helped by accelerating shipments of ARM-based server CPUs at mainstream hyperscalers, data center royalty more than doubled YoY and continued to climb. ARM’s share in data center networking chips (DPU, SmartNIC) is near 100%.

3) Core metrics: (i) ACV came in at $1.66bn, up 2.5% QoQ, with new contracts estimated to add roughly $540mn of revenue. (ii) RPO was $2.07bn, down 3.6% QoQ and below the Street’s $2.67bn;

(iii) ACV/RPO was 0.8 this quarter, rising further and indicating a shift toward shorter-duration orders.

4) Opex: R&D was $700mn (+28% YoY). The increase reflects heavier investment in next-gen architecture, CSS (Compute Subsystem), and the newly launched ARM AGI CPU line to address more complex compute scenarios. A relatively high R&D ratio directly caps margin expansion, with net margin at 21%.

5) Next-quarter guide: ARM guides FY2027 Q1 revenue of $1.21–1.31bn (midpoint $1.26bn, +19.7% YoY), broadly in line with consensus ($1.25bn). Non-GAAP EPS is guided at $0.36–0.44, a touch above the Street ($0.37).

Dolphin Research view: Soft print, firm AI ambition; in-house CPUs aim at the data center

ARM’s quarter was broadly in line, with revenue and GPM essentially matching expectations. License maintained >20% growth, while Royalty growth slowed to ~10% on weaker handsets. Supported by topline growth and margin expansion, net income reached $300mn this quarter.

Versus the print, investors are more focused on three indicators: guidance, ACV, and RPO.

1) Guidance: muted. Q1 revenue guidance of $1.21–1.31bn is roughly in line with the $1.25bn consensus. Both License and Royalty are guided to grow ~20% YoY.

2) ACV: a leading indicator of next-quarter revenue. ACV was $1.66bn, up 2.5% QoQ. Based on this quarter’s revenue mix, Dolphin Research estimates ~$400mn came from prior contracts, while 'new contracts recognized this quarter + Royalty' totaled about $1.09bn.

3) RPO: RPO was $2.07bn, below the $2.67bn Street view. Given the higher ACV/RPO, the order book is skewing shorter-term, consistent with urgent AI demand.

Taken together, these operating metrics suggest near-term results will remain subdued. The handset slump is a drag, while the next wave of ARM-based chips from players like Nvidia should add material contribution from 2H.

With hyperscalers lifting capex again, the Street is prioritizing medium- to long-term visibility of ARM’s AI growth: higher royalty rate mix, CSS (Compute Subsystem) penetration, and its own AI CPU initiative among others.

(Note: CSS is a pre-assembled IP module that includes Arm CPU cores plus other IP, enabling customers to bypass integration and speed time-to-market.)

i) Data center and AI: per-chip royalties are higher in data center, so unit growth in DC/AI should lift the blended royalty rate.

Data center royalty has more than doubled YoY for multiple quarters, with Cloud AI the largest driver. ARM’s share in DC networking (DPU, SmartNIC) is near 100%.

Management also indicated that future platforms from Nvidia, Google TPU, and AWS Trainium will be ARM-led, with the CPU stack trending toward a 'full-ARM' design.

ii) ARM’s first entry into AI data center CPUs: The ARM AGI CPU features 136 Neoverse V3 cores, TSMC 3nm, TDP <300W, and a dual-die design. A second-gen ARM AGI CPU is planned for 2027, with ongoing iterations thereafter.

Management positions AGI as a superior alternative to x86 CPUs, with higher performance and lower power vs. x86. Notably, ARM has secured committed customers: META, OpenAI, and three startups as core users for AI Head Node servers. SAP, Cloudflare, and two others plan to deploy it for general ARM CPU workloads.

AI Head Node: functions as the 'commander' and 'logistics chief' of compute clusters. It does not require the highest clocks, but needs very high core counts and massive PCIe lanes.

Amid the AI upcycle, the CPU role is diverging across tiers and tasks, including general-purpose CPUs, AI head nodes, and CPUs within agentic architectures. As with ARM’s new AI CPU, the initial focus is non-mission-critical, hyperscale compute and on-prem deployments. The company does not plan to enter traditional PC markets to compete with Intel or AMD, but will target incremental data center needs in data throughput and job orchestration.

At ARM’s current market cap (~$252bn), the stock implies ~100x PE on FY2028 post-tax core OP, assuming a two-year revenue CAGR of +25%, GPM of 98.3%, and a 12% tax rate. Given high AI growth visibility over the next two years and major customer launches in 2H, valuation should be viewed against next fiscal year.

While near-term results look subdued, the market is still willing to ascribe a near-triple-digit PE. Beyond ARM’s scarcity value, investors expect it to remain a key beneficiary of the AI cycle, with ample mid- to long-term runway.

Notably, as CPUs enter an upcycle, ARM’s AI CPU puts it directly on the track and adds a new growth leg. In yesterday’s AMD management commentary, the server CPU TAM was raised from $60bn to $120bn, which should further lift expectations for the CPU industry.

Separately, as next-gen products from Nvidia and Google ramp in 2H, ARM’s results should see clearer inflection. Near-term softness is likely to be 'forgiven,' as it does not alter the medium- to long-term thesis for ARM.

With the ongoing AI wave and CPU cycle, drivers including a rising royalty rate, CSS penetration, and in-house AI CPUs remain intact. ARM should continue to command a premium multiple.

Below are Dolphin Research’s charts and datapoints on ARM: