Arm at 300 times earnings: opportunity or trap?

Our answer

Arm is one of the best business models in semiconductors, and its share price already assumes the next five years go to plan. Royalties are compounding, the cloud is moving its CPUs onto Arm, and the company has started selling its own chips. All of that is real. But at about $300 a share the stock is priced for the 2031 plan to work, while the earnings under the price are thin, part of the growth comes from Arm's own parent, and the new chip business puts Arm in competition with its best customers. On balance, we think the stock is more trap than opportunity at this price. What would change our mind is set out at the end.

What 300 times earnings is measuring

The headline multiple is roughly right. Arm closed at $302.56 on October 6 ^[1]. Diluted GAAP EPS was $0.85 for the year to March 2026 ^[2], and the June quarter earned $0.25 a share against $0.12 a year earlier ^[3]. That puts trailing earnings at about $0.98 a share and the stock at about 309 times.

The forward multiple is where the confusion starts. Analysts' normalized consensus for the year to March 2027 is $2.23 a share, and $3.06 for the year after; on a GAAP basis the same-year consensus is $1.25 ^[4]. Divide the share price by each and you get anything from about 99 to about 242 times. Any single "forward P/E" quoted for Arm without its basis tells you less than it seems to.

The gap between those numbers is the story. In the June quarter, Arm's GAAP operating income was $91 million. The company's adjusted figure was $531 million, after adding back $343 million of share-based pay, $90 million of payroll taxes on it and $7 million of other items; that lifted the operating margin from 7.1% to 41.2% ^[3]. Share-based pay came to $1,052 million for the full fiscal year ^[2].

We don't think share-based pay should be waved away. It is how Arm pays its engineers, and shareholders bear it through dilution. Count it, and the same year's consensus takes Arm from about 136 times earnings to about 242; on trailing GAAP earnings it is about 309. And the June quarter's GAAP profit leaned on investment gains: operating income fell to $91 million from $114 million, while income from equity stakes rose to $128 million from $4 million ^[5]. The doubling in GAAP EPS was not the core business at work.

The bright side is real

Start with what is genuinely strong. June-quarter revenue rose 22% to $1.29 billion, with royalty revenue up 22% to $715 million and data center royalties more than doubling. Arm attributes the royalty growth to higher royalty rates per chip from its Armv9 architecture and Compute Subsystems (CSS), and to more Arm chips in data centers. Neoverse, its infrastructure line, has passed 1.5 billion cores shipped, with the latest 500 million taking nine months against six years for the first billion ^[3].

The mechanism matters more than the growth rate. Arm's royalty rises as more of its own designs go into a chip ^[2]. Moving customers from a bare instruction-set licence to ready-made cores, and then to whole subsystems, lets Arm earn more per chip without needing more chips. That is a rare kind of pricing power, and it explains why investors pay up.

The cloud is the proof. AWS's Arm-based Graviton platform ^[3] is on its fifth generation: Graviton5 has 192 cores per chip, and more than 120,000 customers build on Graviton ^[8]. Microsoft's Cobalt 200 is built on Arm's CSS V3 ^[9]. Google's Axion runs Arm Neoverse cores ^[10]. The premise that more AI chips are being built on Arm is correct.

Where the bull story is weaker than it sounds

Two popular arguments need correcting.

Not every Arm-based chip pays Arm the same. NVIDIA's Grace CPU Superchip used Arm's own cores, 144 Neoverse V2 cores in one module ^[11], but its next CPU, Vera, uses NVIDIA-designed Olympus cores that run Arm's instruction set ^[12]. Qualcomm's Dragonfly C1000, which Arm itself calls Arm-based ^[3], uses Qualcomm's own Oryon cores and is planned for Meta's server fleet ^[13]. Arm's annual report is explicit that customers who license only the instruction set and design their own cores pay it less ^[2]. So the count of Arm-based AI chips can rise while Arm's take per chip falls, and some of the biggest new designs are moving in exactly that direction.

NVIDIA's Arm PC is coming, not here. NVIDIA said RTX Spark laptops and desktops would arrive this fall from ASUS, Dell, HP, Lenovo, Microsoft Surface and MSI ^[14], and Arm calls it the first agentic PC built on its CSS ^[3]. That is the most promising royalty story of the three, because CSS earns Arm more per chip. But on October 7, NVIDIA's product page still offered only a sign-up for availability notices ^[15]. It is an option on a market that has not started.

The dark side

Growth that runs through the parent

This is the part of the story that gets the least attention. Revenue from related parties reached $1,499 million in the fiscal year to March, 30% of the total, against $823 million and 21% a year earlier, while revenue from external customers grew only from $3,184 million to $3,421 million. The biggest piece of the increase is one arrangement with an affiliate of SoftBank Group, Arm's controlling shareholder: $704.4 million of revenue, against $145.5 million the year before, with $645.8 million held as current contract assets at year end ^[2]. It kept growing in the June quarter, to $192.9 million from $126.1 million ^[5].

Strip that arrangement out of both years and fiscal 2026 revenue grew about 9%, not the 23% headline (total revenue was $4,920 million, against $4,007 million ^[2]). Revenue booked from your own parent, much of it not yet billed, is not worthless. But it is not the same quality as royalties from independent chipmakers, and a valuation built on the headline growth rate is paying for it as if it were.

Selling chips means competing with customers

In March, Arm moved into production silicon for the first time, launching the Arm AGI CPU ^[6]. By July it said demand had passed $2 billion across its next two fiscal years, and that the manufacturing capacity it had secured covered the $1 billion opportunity it first outlined ^[3], with production expected by the end of 2026 ^[5]. Chief executive Rene Haas has set a target of $15 billion of revenue from the chip by 2031, alongside total revenue of $25 billion and earnings of $9 a share, and the finance chief said the chip sells at about a 50% gross profit ^[7]. Demand is not revenue, and the $15 billion is a target, not a result.

We see three problems. The first is economics: a chip business earning about half its revenue as gross profit will dilute a company built on near-pure licensing, and Arm's own annual report warns of margin compression, more volatile revenue and higher capital needs. The second is customers: the same report acknowledges that many of Arm's licensees may now face direct competition from it ^[2]. Customers already have alternatives. Big customers are designing their own Arm-compatible cores, and NVIDIA was among the investors when SiFive, which licenses designs for the rival open RISC-V architecture, raised $400 million in April ^[16]. The third is supply: Arm has secured capacity for half the demand it talks about.

A lawsuit with a 9% customer

Qualcomm accounted for 9% of Arm's revenue last year ^[2]. Arm lost its case over Qualcomm's Nuvia licence at trial and has appealed ^[5], and Qualcomm's own suit against Arm went before a jury on October 5 ^[17]. Whatever the verdict, a major customer that is suing Arm and building its own cores is not one to count on.

A small float and a pledged stake

SoftBank owns about 86.4% of Arm, leaving 145,344,760 shares in public hands. Of its holding, 769,029,000 shares, a 72.0% stake, are pledged as security for a SoftBank loan that can require prepayment if Arm's share price falls below certain thresholds ^[2]. Price discovery happens on a small slice of the company, with a large pledged stake behind it. The stock fell from a June high of $452.70 to an August low of $219.39 ^[18], and we expect moves of that size to recur in both directions.

Against its peers

On Yahoo Finance's key-statistics pages viewed on October 7, Arm's forward P/E read 151.52 ^[19], against 24.88 for NVIDIA ^[20] and 39.53 for AMD ^[21]. Yahoo does not say which earnings year or basis these use, and its Arm row does not match the October 6 close, so read the gap as an order of magnitude rather than a precise ratio. At the end of 2025 the same page had Arm at 48.08 times forward earnings and a market value of $116.09 billion ^[19]; at the October 6 close Arm was worth about $323 billion ^[1]. Nearly all of this year's rise has come from the multiple, not from higher earnings forecasts. NVIDIA, whose systems drive the AI build-out, is priced at about a sixth of Arm's forward multiple.

Verdict: more trap than opportunity at this price

At $302.56, management's $9 EPS target for 2031, on a basis the CNBC report does not specify, values Arm at about 34 times earnings five years out; if that is an adjusted figure, the GAAP multiple would be higher. That is not outrageous for a business in Arm's position, and it is exactly why we think the risk is lopsided. The price already assumes the hardest parts of the plan go right. The AGI CPU has to grow from a $1 billion opportunity spread over two fiscal years into a $15 billion business by 2031. Licensees have to keep buying Arm's own designs while Arm competes with them. And external growth has to replace the revenue Arm now books from its parent.

If all of that happens, today's price will look reasonable in hindsight, not cheap. If one of those slips, there is little to cushion the fall: the GAAP earnings are thin, the float is small and a 72% stake is pledged. That asymmetry is what we mean by a trap. It is not a judgment that Arm is a bad company. It is a judgment that a great company is priced for perfection.

Three things would change our view:

• AGI CPU revenue arriving at scale, with supply secured well beyond the first $1 billion.
• Royalties from external customers, not SoftBank, driving growth, with GAAP operating margins recovering.
• A Qualcomm verdict in Arm's favour.

The nearest test is the Qualcomm jury's verdict. After that, Arm reports the September quarter on November 4 ^[1], against its guidance of $1.38 billion in revenue, plus or minus $50 million ^[3].