Arm Holdings plc ARM

Business Summary

Arm designs and licenses the instruction set architecture and the CPU, GPU, NPU and system intellectual property that sit inside almost every smartphone, and increasingly inside data centre servers, cars, robots and PCs. It does not fabricate chips: it earns revenue two ways, from licensing (upfront fees and multiyear access agreements, including Arm Total Access) and from royalties charged per chip shipped. In the June 2026 quarter, licensing revenue was $574m (+23% YoY) and royalty revenue was $715m (+22% YoY), so roughly 55% of the top line is an annuity that compounds with customer unit volumes rather than with Arm's own sales effort. Scale is $5.16b of trailing twelve month revenue and a $212.9b market capitalisation, on a 97.5% gross margin that is close to the theoretical maximum for a business selling design files rather than atoms.

Competitively, Arm is the de facto standard in mobile CPU IP, present in effectively all smartphone application processors, and it is now the challenger taking share from x86 in the data centre. Management cited IDC data showing spending on Arm based accelerated server platforms has nearly doubled in two quarters and has now surpassed x86, and the customer list is the proof: NVIDIA Vera, Google Axion, AWS Graviton5, Microsoft Cobalt 200 and Qualcomm Dragonfly C1000 are all Arm based. The structural quirk in the equity story is the free float: 13F filers hold just 0.8% of the shares, SoftBank retains the vast majority, and that scarcity helps explain a 52 week range of $100.02 to $452.70.

Investment Summary

We rate Arm a Buy with a 12 month target of $255, implying 26% upside from $202.49. The core of the call is that this has been a multiple event, not an earnings event: NTM consensus EPS has risen 1.2% over the past month to $2.69 while the share price has fallen 34.1%, so the forward P/E has compressed from a two year average of 153.4x to 75.2x, some 1.1 standard deviations below the mean. Estimates going up while the stock goes down is the cleanest setup in the market when the de rating is sentiment driven rather than fundamental, and the 34.1% one month drawdown against a 3.0% six month drawdown tells you the selling is recent and reflexive.

Three things support the earnings side. First, data centre royalties have more than doubled year over year for several consecutive quarters, and Neoverse has now shipped over 1.5bn cores, with 500m of those in the last nine months against 1bn in the first six years. Second, the Armv9 and compute subsystem mix shift is doing what it was designed to do: management expects double digit smartphone royalty growth even though the handset unit market is projected down by double digits, because higher rates per chip more than offset lower volumes. Third, the Arm AGI CPU, launched in March 2026, has $1bn of secured supply across fiscal 2027 and 2028 against demand now north of $2bn, with management stating that confidence in upside to the $1bn has increased over the last 90 days.

What the market is missing is the durability of the royalty annuity; what it is pricing is margin dilution from silicon. The AGI CPU carries gross margin in the high 30s to low 40s against a 97.5% corporate gross margin, which is why the stock sits above its two year average on EV/Revenue (+7%) while sitting 51% below it on forward P/E. The call is wrong if blended gross margin falls below 90% without a commensurate volume ramp, or if smartphone royalties roll over as memory cost inflation forces handset makers to cut units faster than Arm can raise rates.

Financial Performance

The trailing twelve months delivered $5.16b of revenue on a 97.5% gross margin, but only a 7.6% GAAP operating margin and a 20.2% net margin, which tells you the gross profit is still being spent on research and development and on share based compensation rather than dropping to the operating line. The quarterly sequence shows both the growth and the lumpiness: $1.14b, $1.24b, $1.49b and $1.29b across the last four reported periods, with the March 2026 quarter (fiscal Q4 2026) the largest at $1.49b, up 20% year over year, on licensing of $819m (+29%) and royalties of $671m (+11%). Net income across those four quarters totalled roughly $1.04b, and non GAAP earnings run well ahead of GAAP: $0.60 adjusted in the March quarter (a $0.02 beat) and $0.45 in the June 2026 quarter, up 29% year over year and above the high end of guidance.

The quality signal is the mix. Royalties grew 22% in the June quarter to a record first quarter $715m, and within that the data centre line more than doubled again while edge AI grew despite a soft handset market and physical AI contributed on the back of ADAS and autonomous systems. That is a portfolio of royalty streams diversifying away from a single end market, which is precisely what a 97.5% gross margin business needs to justify a premium multiple. The balance sheet is unlevered at 5.6% debt to equity and ROE is a modest 13.4%, dragged by the equity raised at IPO and by the share count; cash and equivalents were not disclosed in the materials we reviewed.

Outlook and Estimates

Consensus looks for NTM revenue of $6.09b, up 36.4%, and NTM EPS of $2.69, with fiscal 2028 revenue of $8.30b and EPS of $3.02 across 19 analysts. Company guidance is qualitative on the swing factor: $1bn of AGI CPU revenue across fiscal 2027 and 2028, gross margin in the high 30s to low 40s for the first generation, moving toward 50% over a couple of years as more of the work currently done by an ASIC partner is brought in house, with a detailed revenue and margin framework promised at the fiscal Q3 print. Licensing is the lumpy line and the one to watch: it fell from $819m in the March 2026 quarter to $574m in the June 2026 quarter, a 30% sequential decline, which is normal for a business where large access agreements land when they land.

For the estimates to be met, three things have to happen. Supply has to materialise for the AGI CPU across wafers, substrates, test capacity and memory, and management says confidence on that has improved over the last 90 days. Smartphone royalties have to keep growing double digits against a handset market projected down by double digits, which requires Armv9 and compute subsystem penetration to keep lifting the rate per chip. And the licensing pipeline has to convert, because a single slipped access agreement can move a quarter by 10% or more. The risk to consensus is not demand, it is the margin arithmetic of a silicon business sitting inside a 97.5% gross margin IP company.

Valuation

On forward numbers the stock trades on 75.2x NTM P/E using the $2.69 consensus figure, 35.5x NTM sales of $6.09b, and 63.4x NTM EV/EBIT. We would flag once that the $2.21 forward EPS in the vendor data set is inconsistent with the revision series and with the 75.2x multiple, and appears stale; we use $2.69 throughout. On trailing numbers the picture is extreme: TTM net income of about $1.04b is roughly $0.99 per share, so the trailing P/E is about 205x and the trailing EV/EBITDA and EV/EBIT multiples are not useful anchors. Against its own two year history the forward earnings multiples are deeply discounted: forward P/E is 51% below the 153.4x average, EV/EBITDA is 50% below at 58.4x, and EV/EBIT is 61% below at 63.4x. But the revenue and book multiples are not cheap at all, with EV/Revenue at 28.7x versus a 26.9x average and price to book at 25.1x versus 21.1x, and that divergence is the whole story: the market has marked down Arm's margins, not its growth. Our DCF, at a 9.5% WACC and 3% terminal growth, gives roughly $210 per share on 25% free cash flow growth for a decade, about $265 on 30%, and about $165 on 20%. The current price is discounting a decade of mid 20s compounding, which is demanding but not unreasonable against 36% NTM revenue growth.

Among peers, Qualcomm trades on 19.0x trailing earnings because it is a mature handset silicon vendor with cyclical exposure and no royalty annuity on other companies' chips, while Synopsys at 81.8x and Cadence at 64.4x are high quality EDA franchises with recurring revenue but high single digit growth and no volume leverage. Marvell at 56.6x is the closest read on custom AI silicon, with lower gross margins and more direct exposure to hyperscaler capex cycles. Arm deserves a premium to Qualcomm and Marvell on 97.5% gross margins and the royalty annuity, and a premium to Synopsys and Cadence on growth, but it should trade below its own two year average until the AGI CPU margin trajectory is proven. Our $255 target is 95x NTM EPS of $2.69 and about 42x NTM sales, a partial re rating that still sits 38% below the two year average forward P/E and well below the stale consensus mean of $375.92, where the lowest target on the street ($250) is above the current share price.

Catalysts

The next hard datapoint is the fiscal Q2 2027 print, expected around late October or early November 2026, where the market will look for confirmation that data centre royalties doubled again and that licensing did not slip. Management has promised a detailed AGI CPU revenue and margin framework at the fiscal Q3 print, reported around February 2027, and that is the single most important event for the margin debate. In between, watch for named AGI CPU customers beyond the multiple customers in the U.S. and China already disclosed, for further hyperscaler design wins on Neoverse CSS, and for any update on Armv9 and compute subsystem penetration rates in smartphones. Any Arm China governance development, and any change in the SoftBank stake or the free float, would also move the shares given how thin the float is.

Key Risks

The most important risk is that the AGI CPU turns Arm into a lower quality business. First generation gross margin of high 30s to low 40s against a 97.5% corporate gross margin means every dollar of AGI CPU revenue dilutes the blend, and if the 50% target slips beyond a couple of years the earnings power implied by consensus is not there. A blended gross margin below 90% with no offsetting volume surprise would invalidate the thesis, and on our numbers that is the difference between a $255 stock and something closer to the bear DCF of $165.

Second is China and the Arm China joint venture, which holds the licence for the China market; the revenue split is not disclosed in the materials we reviewed, but China has historically been a mid 20s percentage of group revenue and the governance structure has been a recurring overhang. Third is licensing lumpiness, illustrated by the 30% sequential decline in licensing from the March to the June 2026 quarter, where a single delayed access agreement can reset a year of expectations. Fourth is the architecture threat: RISC-V in IoT and microcontrollers, and in house cores at Apple, Qualcomm and the hyperscalers, which cap Arm's royalty rate over time even if they do not displace the architecture. Finally, valuation and float: 35.5x NTM sales, a small free float, insider selling of $5.8m by the CFO in September 2026 at $300, and a 52 week range of $100.02 to $452.70 mean position sizing matters as much as the call. A break below $190, the level implied by the two year average EV/Revenue, would signal the market is repricing growth rather than margins.