ARM Holdings

ARM Holdings in 2026 is perhaps the most unusual mega-cap in the market: a chip designer that makes no chips, only sells licenses — and yet sits inside 99% of all smartphones, 70% of all datacenter CPUs (via AWS Graviton, Microsoft Cobalt, Nvidia Grace), and now 100% of Apple Macs. The business model is brilliantly simple: a one-time license fee for architecture access, plus a royalty per chip produced — typically 1.5-3% of chip ASP for ARMv9, about double the ARMv8 rate.

The real 2026 AI-boom lever is not the number of ARM chips but the royalty mix shift. A smartphone SoC like the Snapdragon 8 Gen 4 pays maybe $0.60 ARM royalty per unit; an Nvidia Grace Hopper Superchip pays an estimated $35-60 (ARM cores scale with cluster size and IP complexity). That is a 60-100× leverage per chip unit — and the datacenter mix is growing against mobile at roughly 28% YoY.

The balance-sheet problem: SoftBank still holds 89.9% of shares post the September-2023 IPO. Every quarter brings speculation about another float increase — and each secondary sale typically pushes the stock 8-12% down. The stock trades at a forward P/E of 73 and a PEG of 1.82 — that is a growth story already pricing in enormous expectations.

In Q1/2026 13F reporting, ARM is conspicuously underweight at top-tier smart-money managers — no Berkshire position, no Bridgewater position, no Pershing Square position. This mostly reflects valuation concerns: at 73x forward P/E with a SoftBank 89.9% float overhang, the setup is unattractive for classic value or GARP investors.

It is different at AI-focused growth hedge funds: Coatue Management holds 4.2M shares (~$930M), Tiger Global holds 3.1M (position built in Q3/2025). Both justify the valuation via the ARMv9 royalty-rate uplift argument and the Nvidia Grace pipeline.

Notable: not a single 5% insider sale filing in the last 12 months from CEO Rene Haas or CFO Jason Child — despite the stock running 2.4× since IPO. This is read as bullish by the smart-money community: management believes the stock is not yet at target.

ARM's valuation in 2026 has no peer in the US market: forward P/E 73, PEG 1.82, P/S 32 — all in the 95+ percentile of all S&P 500 tech stocks. Classically that cannot be justified. The bull-case model rests on two assumptions: (1) ARM royalty revenue grows 30%+ annually for 5+ years (possible but not guaranteed — depends on the AI capex cycle), (2) operating margin expands from current 53% to 65%+ (via software leverage). If both hold, a DCF model (15% discount, 5% terminal) implies fair value of ~$240-260 — barely above the current $221. At only 20% growth and 58% margin, fair value drops to $130. The asymmetry is bounded on the downside only if you fully believe the AI story.

1. August 5, 2026: Q1 FY27 earnings — first concrete commentary on ARMv9 royalty mix and AI datacenter pipeline; threshold >35% royalty growth
2. September 2026: Apple event with M5 Pro/Max — ARM royalty per M5 chip estimated $1.20; 35-40M Macs annually
3. Q4 2026: Expected SoftBank secondary tranche to fund Vision Fund 3 — float increase of 10% to 15% likely, historically -10% price impact

ARM is the most honest AI story in the market — unlike Nvidia, which is capex-cyclical, ARM collects royalties on EVERY new chip design, regardless of foundry, regardless of end customer. That is a 'picks-and-shovels' business model in its purest form. But: 73× forward P/E is too rich for me personally. I would only seriously buy ARM for a long-term portfolio below $150 — corresponding to a forward P/E of 50. Currently on my watchlist, not a buy. For anyone seeking AI exposure beyond Nvidia/AMD/AVGO and able to tolerate volatility, this is a unique pure-play. Important note for international investors: ARM trades in USD on NASDAQ, despite Cambridge HQ — no UK withholding tax applies, just standard capital gains in your home jurisdiction.