The consensus reading of Arm stock is that a 40% drawdown from June’s record has made a quality compounder cheap again. That framing misses the mechanism that actually sets the floor. Arm Holdings closed at $264.90 on 17 September 2026 after an 8.57% single-session jump, but the number that matters more sits in the company’s own annual report: SoftBank Group beneficially owns 86.4% of Arm, and subsidiaries of SoftBank have pledged 769,029,000 ordinary shares — a 72.0% equity interest in Arm — as collateral for a margin loan facility whose terms can require prepayment “in the event the trading price of our ADSs declines below certain thresholds.” Arm’s free float is roughly 145 million ADSs. This Arm stock price prediction puts the bull case at $425 and the bear case at $150, and the distance between them is mostly a question of who is forced to sell, and when.
Here is the synthesis no other Arm coverage seems to be making. Two facts from separate documents, combined, change the shape of the risk. From the Form 20-F filed 26 May 2026: 72.0% of Arm’s equity is pledged against a loan that has a loan-to-value margin-call trigger, and Arm itself warns that in a margin call SoftBank’s subsidiary “may decide to sell some of the pledged shares.” From the Q1 FYE27 shareholder letter and the earnings call: Arm’s record licensing quarter included roughly $193 million from the SoftBank technology licensing and design services agreement, with CFO Jason Child guiding that relationship to about $200 million per quarter for the rest of the year. So SoftBank is simultaneously Arm’s 86%-owner, a leveraged holder of the stock, and one of its largest customers. Having covered a dozen controlled-company semiconductor listings, I have not seen those three roles stacked this tightly. It does not make the bull case wrong. It makes the tail on the bear case much fatter than a 110x forward multiple would suggest on its own.
Key facts: Arm stock at a glance
- Spot: $264.90 close, 17 September 2026 (+8.57% on the day) — stockanalysis.com, cross-checked against CNBC
- 52-week range: $100.02 (5 Feb 2026) to $452.70 (18 Jun 2026); highest close $439.46, so ARM sits 39.7% below its closing peak
- Year to date: +142.3% from the 31 December 2025 close of $109.31
- Q1 FYE27 revenue: $1,289m, +22% YoY — royalty $715m (+22%), licensing $574m (+23%) — Arm, 29 July 2026
- The margin story: GAAP operating margin fell to 7.1% from 10.8%; non-GAAP operating margin rose to 41.2% from 39.1%. The gap is $440m of a single quarter
- Valuation: market cap ~$282.9bn on 1.07bn shares; 54.9x trailing sales, 111x forward earnings on $2.39 forward EPS (CNBC, 17 Sep 2026)
- Balance sheet: $3,888m cash and short-term investments; trailing-twelve-month free cash flow $1,397m
What is actually happening at Arm, and why the stock still fell
Arm does not make chips, historically. It licenses the instruction set and the CPU designs that other companies build into silicon, then collects a royalty on every unit shipped. That business runs at a 97.2% GAAP gross margin because the marginal cost of an additional licensed core is close to zero. Think of it as a toll booth on a road someone else paved: Arm’s revenue scales with traffic, not with construction spending.
In the June quarter the traffic surged in exactly the lane investors care about. Data-centre royalty revenue more than doubled year over year for a second consecutive quarter. Neoverse core shipments passed 1.5 billion, and Arm disclosed that the most recent 500 million of those shipped in nine months, against six years for the first billion. That is a compounding curve, not a cycle.
And yet the stock fell 8% on the print, to about $225 on 29 July, before recovering into September. The reason is that the quality of the beat was worse than the size of it. Royalty growth guidance for the full year came down from about 20% to the high teens, because higher memory prices are pushing handset makers to raise prices across mid-tier and premium tiers rather than just entry-level — a read-across from the same DRAM squeeze FinanceFeeds tracked through the memory complex earlier this month. Child guided Q2 royalty growth to just 13%. Meanwhile the $193m SoftBank licensing line accounted for roughly a third of the quarter’s $574m of licence revenue — a figure larger, on its own, than the entire $106m year-over-year increase in that line. Both details come from the 29 July call, as reported in TIKR’s account of the transcript.
The second thing that happened is strategic, and it is the one that should reset how the multiple is set. In March 2026 Arm launched the Arm AGI CPU — its own production silicon, not IP. Management told shareholders that demand “now exceeds $2 billion across fiscal 2027 and fiscal 2028,” against roughly $1 billion of manufacturing capacity Arm has secured. In the shareholder letter, CEO Rene Haas put it this way: “Arm delivered a record first quarter, with data center royalties more than doubling year over year as the transition of AI infrastructure to Arm continued to accelerate. Demand for the Arm AGI CPU has continued to exceed our initial expectations, and our continued work to expand manufacturing capacity with our partners gives us increasing confidence that we can deliver at the scale our customers require.”
Who is actually building on Arm — and who is pushing back
The named-customer list is the strongest part of the bull case, because it is no longer speculative. Nvidia has moved its Arm-based Vera CPU into full production; Arm says Vera delivers up to 50% higher CPU performance and twice the energy efficiency of comparable x86 systems, and it is the host CPU for Nvidia’s next-generation AI racks. Google now runs its Arm-based Axion as the host processor for its newest TPU systems. AWS announced a multi-year agreement with Meta to deploy tens of millions of Graviton5 cores for agentic AI workloads. Microsoft expanded Azure Cobalt 200 virtual machines built on Arm Neoverse compute subsystems. And Qualcomm — a company that spent years in litigation with Arm over licensing — has said it will enter the AI data-centre CPU market with the Arm-based Dragonfly C1000.
That last one is the tell. When your most combative licensee commits its data-centre roadmap to your architecture, the architecture question is settled; only the price question remains. IDC reported that spending on Arm-based accelerated server platforms nearly doubled over two quarters and has now surpassed x86 platforms, and raised its 2026 AI infrastructure spending forecast to almost $500 billion.
The pushback is subtler and comes from the sell side rather than from customers. HSBC downgraded Arm to Hold in mid-July on foundry capacity constraints at TSMC’s advanced nodes and on a valuation that had already discounted years of growth. Morgan Stanley’s published target has sat well below spot for much of the summer. The consensus target across roughly 21 covering analysts is about $288 — barely 9% above the 17 September close, which is an unusual place for a stock the market treats as a secular AI winner. Compare that with the setups in our Marvell price prediction and Applied Materials price prediction, where consensus sits materially above spot.
The valuation maths, and the margin that nobody models
Arm trades at roughly 54.9x trailing sales and 111x forward earnings. Nothing in the semiconductor complex trades there on fundamentals alone; the multiple is a claim about what Arm becomes, not what it is. The claim is reasonable while 100% of revenue is 97%-gross-margin IP. It is much harder to defend once a material slice of revenue is production silicon.
Here is the arithmetic the bulls and the bears are actually arguing about. Suppose the AGI CPU business delivers the full $2bn of demand across FY27 and FY28 — call it $1bn a year against a current annualised revenue run-rate of about $5.5bn. Custom silicon sold into hyperscalers does not earn 97% gross margins; a generous assumption is 50-60%. Blend $1bn at 55% into $5.5bn at 97% and group gross margin drops toward 90%. That is still extraordinary. But a 700-basis-point structural gross-margin decline on a stock priced at 111x forward earnings is not a rounding error — it is the difference between a $425 print and a $310 one.
| Scenario | Level | vs $264.90 spot | What has to be true |
|---|---|---|---|
| Bull | $425 | +60.4% | AGI CPU capacity expands past $1bn secured supply; data-centre royalties keep doubling; smartphone royalty drag annualises out by FY28; the June high is retested |
| Base | $310 | +17.0% | Q2 FYE27 lands at the $1.38bn guide, royalty growth stabilises in the high teens, multiple holds near 120x forward earnings on rising estimates |
| Bear | $150 | −43.4% | AI infrastructure capex digests; silicon revenue compresses blended gross margin; multiple resets toward 60x forward earnings — and any forced supply from the SoftBank facility lands into a 13.6% float |
Quick take: the bull case needs no new customers — only capacity. The bear case needs no customer losses — only a multiple. That asymmetry is why the base case sits closer to the bull level than the midpoint.
The governance and policy tension
Arm is a “controlled company” under Nasdaq rules and a foreign private issuer, which means it is permitted to skip several governance requirements that apply to ordinary listings — including the requirement that a majority of the board be independent. That is disclosed plainly in the 20-F. For an equity trading at 111x forward earnings, governance optionality that favours the controlling shareholder is a real, if unquantifiable, discount factor.
The policy overlay is more interesting in 2026 than it has been. Washington has moved from subsidising strategic compute to owning it: the Commerce Department has taken non-controlling equity stakes in quantum computing companies in exchange for CHIPS Act research awards, a structure FinanceFeeds has covered alongside Nvidia’s own portfolio stakes, including in Intel. Arm sits outside that programme — it is UK-headquartered, Japanese-controlled and US-listed — but it is exposed to the second-order effect. If US industrial policy increasingly routes advanced-node capacity through firms in which the government holds equity, a foreign-controlled architecture licensor is competing for the same TSMC wafers without the same political sponsorship. HSBC’s foundry-capacity thesis and the AGI CPU supply constraint are the same problem viewed from two angles.
Export controls are the other live wire. Arm’s architecture is licensed globally, including into China through Arm China, and Arm disclosed new customers in both the US and China for the AGI CPU this quarter. Any tightening of US or UK controls on advanced CPU designs reaching Chinese data centres would hit the licensing line directly — the highest-multiple part of the business.
What happens next: three predictions
1. Q2 FYE27 beats the $1.38bn guide but royalty growth prints in the low teens. Arm has beaten its own revenue guidance in each of the last several quarters and guided Q2 royalty growth to 13% explicitly. The likely shape is a revenue beat driven by licensing — including the ~$200m/quarter SoftBank line — with royalties landing in line. Expect the market to treat a licensing-led beat sceptically, as it did on 29 July. Results are due in early November 2026.
2. The AGI CPU capacity number gets raised again before the bull case gets tested. Management moved from “$1bn opportunity” to “>$2bn of demand” in a single quarter and said confidence in upside “has increased in the past 90 days.” The gating item is foundry allocation, not orders. A disclosed capacity expansion — an additional wafer commitment, a second foundry partner, or a named hyperscaler design win — is the single most likely catalyst to carry ARM back through $350 toward the $425 bull level. Absent it, the stock churns in the $240-$310 band.
3. The SoftBank facility becomes a talking point if ARM breaks $200. This is the prediction with the clearest causal chain and the least coverage. The facility’s loan-to-value trigger is undisclosed, but Arm’s own risk factor spells out the sequence: price decline → LTV breach → margin call → SoftBank deposits funds or “may decide to sell some of the pledged shares.” With a free float of roughly 145 million ADSs against 769 million pledged shares, even a small forced sale is large relative to what trades. That is the mechanism that turns an ordinary de-rating into the $150 bear case. It is also why the bear case is a level, not a fair value: below a certain price, supply stops being a function of fundamentals.
For readers weighing Arm against the rest of the AI compute complex, our Micron versus Nvidia comparison and our neocloud coverage map the same capex cycle from the memory and compute-rental ends of the chain.
FAQ
What is the Arm stock price prediction for 2026?
This analysis sets a bull case of $425, a base case of $310 and a bear case of $150 against the 17 September 2026 close of $264.90. The bull level is a retest of June’s $452.70 record high and requires Arm to expand AGI CPU manufacturing capacity beyond the $1bn it has secured. Wall Street consensus sits at roughly $288 across about 21 covering analysts — closer to the base case than to either extreme.
Why did Arm stock fall after record Q1 FYE27 earnings?
Arm fell about 8% on 29 July 2026 despite record revenue of $1.29bn and non-GAAP EPS of $0.45, above the top of guidance. The market focused on full-year royalty growth guidance falling from roughly 20% to the high teens on smartphone memory-price headwinds, on Q2 royalty growth guided to 13%, and on the fact that around $193m of licensing revenue came from the SoftBank related-party agreement rather than third-party customers.
How much of Arm does SoftBank own?
SoftBank Group beneficially owned approximately 86.4% of Arm’s issued and outstanding share capital as of 21 May 2026, leaving a public float of roughly 145.3 million ADSs. SoftBank subsidiaries have pledged 769,029,000 Arm ordinary shares — a 72.0% equity interest — as collateral under a margin loan facility that can require prepayment or additional collateral if Arm’s ADS price falls below certain thresholds.
Is Arm’s 97% gross margin sustainable?
Not indefinitely at that level. GAAP gross margin was 97.2% in Q1 FYE27 because the business licenses IP rather than selling hardware. The Arm AGI CPU changes that: production silicon sold to hyperscalers carries materially lower gross margins. If the AGI CPU delivers around $1bn a year at a 50-60% gross margin against a $5.5bn base at 97%, blended group gross margin drifts toward 90% — still exceptional, but a structural decline the current 111x forward multiple does not obviously price.
What is the Arm AGI CPU and why does it matter?
Announced in March 2026, it is Arm’s first move into selling its own production silicon rather than licensing designs. Arm told shareholders demand now exceeds $2bn across fiscal 2027 and 2028 against roughly $1bn of secured manufacturing capacity. It matters in both directions: it is the largest new revenue pool Arm has opened in a decade, and it is the reason the company’s margin structure will look different in three years.
Which companies use Arm in the data centre?
Nvidia (Vera CPU, in full production), Google (Axion, host CPU for its newest TPU systems), AWS (Graviton, including a multi-year Meta agreement covering tens of millions of Graviton5 cores), Microsoft (Azure Cobalt 200 on Neoverse CSS) and Qualcomm (the announced Dragonfly C1000). Arm reported Neoverse core shipments passing 1.5 billion, with the most recent 500 million shipping in nine months.
This article is editorial analysis and market commentary, not investment advice. Scenario levels are FinanceFeeds estimates and are not price targets. Figures are as of the 17 September 2026 close unless stated otherwise.