Demand doubles but hesitant to provide guidance—what is Arm’s AGI chip afraid of?

① **AGI Chip Demand Doubled in 6 Weeks Is the Biggest News**: Client demand for Arm’s first self-developed server chip AGI CPU has doubled from $1 billion at the end of March to over $2 billion (FY27-28), but management maintains the $1 billion guidance due to supply chain capacity limitations—demand is not the problem, capacity is. ② **Datacenter Royalties Double for Second Consecutive Year**: Datacenter royalties driven by Neoverse IP grew over 100% year-over-year; management expects another doubling in FY2027. AWS Graviton, Google Axion, NVIDIA Vera all run on Arm architecture, and Arm's share among top supercomputing clients is approaching 50%. ③ **Licensing Revenue Growth Masks Slowdown in Royalties**: Q4 total revenue was $1.49 billion, up 20% year-over-year and above expectations; breaking it down, licensing revenue of $819 million (+29%) was the main engine, with royalty revenue at $671 million (+11%), slowed due to weak smartphone market and high base from MediaTek last year. ④ **Non-GAAP Operating Margin Hits Record 49%**: But GAAP operating margin was only 29.4%—the gap comes from $1.052 billion in equity incentives for the year, accounting for 21.4% of revenue, a cost often overlooked in Arm’s valuation narrative. ⑤ **Business Model Restructuring Has Begun**: Arm is transforming from "the Switzerland of the semiconductor industry"—a pure IP licensor—into an IP+chip dual-track company, aiming for $25 billion revenue in FY2031 ($10 billion IP + $15 billion chips), and EPS above $9. This is a bold bet on corporate identity. ⑥ **Q1 FY2027 Guidance Remains Solid**: Revenue guidance is $1.26 billion ±50 million (midpoint up 20% YoY), Non-GAAP EPS $0.40 ±0.04, maintaining ~20% annual growth pace for both royalties and licensing. From SoftBank's $32 billion acquisition, to the failed NVIDIA $40 billion deal, to the 2023 $54.5 billion relisting, every identity shift for Arm brought skepticism. This time, Arm chose to make chips itself, directly entering its clients' battleground. This quarter's data shows two things: first, the old IP licensing story is still on a >20% growth track; second, the new AGI chip story has demand doubling in 6 weeks, so fast the supply chain can’t keep up. But Wall Street’s reaction was intriguing. Arm shares surged about 13.6% to $237 intraday, but fell back to around $219 after hours, down more than 7%. The average target price among 24 covering analysts is about $180; the share price far exceeds consensus—not so much “unoptimistic” as a stock priced too perfectly searching for its next catalyst. ------------------------------------------------------------------------ ## Royalties: Datacenter Alone Can’t Prop Up the Sluggish Phone Segment Q4 royalty revenue was $671 million, up 10.5% YoY, the lowest growth in the past 5 quarters. This figure would be healthy for most companies, but for Arm—and its recent >20% growth rates—it’s a signal worth unpacking. The core drag comes from the phone market. CFO Jason Child admitted on the call that MediaTek Dimensity 9400’s strong shipment last year created a high base, and worldwide smartphone shipments continued to weaken in the low-end market, resulting in pressure on phone royalties. However, penetration of Armv9 architecture in high-end models is lifting per-chip royalty rates—falling volume offset by rising price. The real highlight is in datacenter. Datacenter royalties driven by Neoverse architecture have doubled YoY for the second consecutive year; management expects another doubling in FY2027. Google announced at Cloud Next that its TPU 8t/8i training/inference chips are fully replaced by Arm-based Axion CPUs, with 80% performance improvement; AWS continues expanding Graviton deployments paired with Trainium; NVIDIA’s Vera CPU also uses Arm architecture. CEO Rene Haas is bold: by decade’s end, by CPU type, Arm will have the largest market share in datacenter. But we must be clear: datacenter royalties’ share of total is still not big enough to fully offset phone and IoT cyclicality. Management’s full-year royalty growth guidance is about 20%, implying datacenter will accelerate after Q1 to make up for Q4’s gap. ## Licensing: ACV Up 22% Reveals Long-Term Health Q4 licensing and other revenues were $819 million, up 29.2% YoY, with SoftBank’s tech licensing and design services contributing $200 million (flat QoQ). Excluding SoftBank, third-party licensing is around $619 million, still robust. More important than a single quarter is ACV (Annual Contract Value), which smooths out big contract timing differences; Q4 reached $1.66 billion, up 22% YoY, consistently above management’s “long-term expectations”. Two new next-gen CSS (compute subsystem) licenses—one for smartphone chips, one for datacenter network chips—and an AI tech strategic partnership with the Indonesian government, all point to greater penetration of Arm tech stack in more endpoints and regions. RPO (Remaining Performance Obligations) fell from $2.226 billion to $2.071 billion, down 7%. This needs context: large licensing contracts recognize revenue up-front at signing, so RPO naturally falls as consumed. This does not mean softening demand. Continued ACV growth is a better forward-looking indicator. ## Margin: Non-GAAP Dazzle vs GAAP Reality Q4 Non-GAAP operating margin was a record 49.1% since listing; Non-GAAP EPS $0.60 (consensus $0.58). Full-year Non-GAAP margin increased from FY2025’s ~47% to ~43%—wait, why lower for full year than Q4? Because Q1-Q3 margins were only 39.1%, 41.1%, 40.7%, with concentrated R&D spending in the first three quarters. But GAAP tells quite a different story: Q4 GAAP operating margin 29.4%, full year only 18.3%. The main gap is stock-based compensation (SBC): $261 million for Q4, $1.052 billion for the year, 21.4% of annual revenue. For a company valued over $250 billion, this SBC intensity is rare in the semiconductor industry. R&D spending is also rapidly expanding. Non-GAAP R&D expense for Q4 was $493 million, up 33% YoY; full year $1.911 billion, up 43%. Total employees grew to 9,584 (+15%), engineers 8,058 (+16%), reflecting ongoing investment in AGI CPU product line and next-gen architecture. Management promises expense growth will be below revenue growth by year-end, returning to positive margin expansion. ## Cash Flow: FY2025's “Abnormality” Finally Corrected FY2026 operating cash flow was $1.524 billion, up sharply from FY2025’s $397 million. Last year’s abnormally low level was mainly due to ballooning contract assets and accounts receivable draining cash; this year these normalized. Non-GAAP free cash flow was $882 million (FY2025 only $99 million), but capex soared from $219 million to $545 million—this is the real price of investing in AGI CPU and datacenter infrastructure. Balance sheet remains clean: cash + short-term investments $3.6 billion, zero interest-bearing debt. ## AGI CPU: Leap from “Blueprint Seller” to “Builder” This is Arm’s most significant business model change in its 35-year history. Previously, Arm was the “Switzerland of semiconductors”—selling design blueprints to everyone, not taking sides, not competing. Now, AGI CPU makes Arm sell finished chips directly to datacenter clients, competing on the same track as its licensing clients AWS, Google, NVIDIA. On the call, analysts asked directly about this “sensitive topic”: how do existing IP customers feel about Arm making chips? Rene Haas answered: every partner was notified in advance, all expressed support—as a larger Arm software ecosystem benefits everyone. Over 50 partners publicly endorsed Arm at the Arm Everywhere event. In numbers, client demand doubled from $1 billion at end of March to over $2 billion in 6 weeks. Demand comes from two types: one, announced clients like Meta increasing orders, and two, new customers—not willing to self-develop chips but needing Arm computing power (SAP, Cloudflare, SK Telecom, OpenAI, etc.), which can directly buy Arm racks from ODM partners like Lenovo, Supermicro. But management cautiously keeps guidance at $1 billion (Q4 FY2027 contribution ~$90 million, FY2028 ~$910 million), because wafer, memory, and packaging/test capacity are not yet locked in. More concrete supply chain updates will come in the Q3 earnings call. Long-term blueprint: FY2031 chip revenue $15 billion + IP revenue $10 billion = $25 billion total, EPS above $9. Chip business aims for ~35% margin long-term, IP business ~65%. Most chip development costs can be shared with CSS IP R&D (same core design), so incremental team size is “in the dozens not hundreds”—management expects chip business to turn profit in FY2028. ## Outlook: Can Two Growth Curves Run in Parallel? Arm faces two key variables determining its valuation path over the next three years. First is the AGI CPU ramp-up speed. The bottleneck for $2 billion demand translating to revenue is in the supply chain, not the market. TSMC advanced process capacity allocation, HBM/DDR5 memory availability, and CoWoS advanced packaging scheduling could all be limiting factors. AMD faced similar GPU supply bottlenecks last year, only relieved this year. Arm’s AGI CPU has 136 cores, requiring high-end packaging. Second is client attitude evolution. The current support from 50+ partners is more at a "welcoming" stage. When Arm chips start to materially grab their datacenter revenues, will this harmony persist? Rene Haas’s argument is partly persuasive: AWS has started selling Graviton compute externally, showing Arm ecosystem demand far exceeds what any single supplier can meet. 24 analysts give an average target price of ~$180, highest $255. Shares surged to $237 on earnings day, traded above most analysts’ bull scenarios. This means market pricing already partly reflects AGI CPU success; from now, every supply chain update will be a stock catalyst or risk. AMD recently raised its 2030 datacenter CPU TAM from $100 billion to $120 billion, in line with Arm's previous $100+ billion forecasts. In this fast-expanding market, Intel’s share dropped from >90% to about 62%, AMD’s is 29% and rising, Arm’s share among supercomputing clients is near 50%. All three companies claim they can get 50%—adding up to 150%, obviously someone will be disappointed. Arm’s unique advantage lies in: whether Graviton, Axion, Vera, or AGI CPU, all these Arm-architecture chips generate royalties for Arm. Even if Arm loses the chip competition to clients, royalties remain its moat. But AGI CPU’s ambition is clearly more—$15 billion FY2031 target means Arm aims to establish standalone chip-driven growth, not just “collect tolls”. This financial report is Arm’s last full annual answer as a pure IP company. Next time we assess Arm, it’ll wear a new label: chipmaker. For the first time in 35 years, the blueprint designer decides to build the house themselves—the question is, will tenants keep paying rent, or hire another architect? ------------------------------------------------------------------------ Risk Disclaimer The market is risky, and investment requires caution. This article does not constitute individual investment advice and has not considered the special investment objectives, financial situations, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their own circumstances. Investment based on this, at your own risk.