Evid Invest
EvidInvest
← Back to Blog

ARM after a 224% run: what the SEC filings confirm about AI infrastructure

·EvidInvest Research
ARMArm HoldingsAetherSEC filingsAI infrastructuresemiconductors

Draft for EvidInvest review. This is financial research content, not investment advice. Built from authenticated Aether financial search over ARM SEC filings, direct SEC filing checks, SEC companyfacts, filed 6-K earnings exhibits, and Yahoo Finance one-year price data. Aether transcript search failed in this research run, so this draft does not rely on earnings-call transcript evidence.

Thesis

ARM is one of the cleaner examples of why AI infrastructure research needs evidence labels.

Based on market data checked in the parent research run, ARM moved from $126.055 on 2025-06-02 to $408.850 on 2026-06-01, a roughly 224.34% one-year close-to-close gain. The market has already rerated ARM as an AI infrastructure beneficiary.

The filing-backed question is narrower and more useful: does ARM’s SEC evidence confirm real AI infrastructure exposure, and where does the story become inference?

Aether’s SEC-first evidence stack points to three conclusions:

  1. ARM has filing-confirmed exposure to AI workloads from edge to cloud, including AI data centers. The latest 20-F says Arm CPUs already run AI workloads in billions of devices and are used from smartphones and cameras to cars and AI data centers.
  2. The revenue mechanism is royalty/licensing value capture, not direct GPU-like server sales. FY2026 filings show growth in royalty revenue and licensing revenue, plus repeated references to Armv9, CSS, data-center deployment, and higher royalty rates per chip.
  3. The downside boundary is concrete. Top-five customers represented 57% of FY2026 revenue; Arm China represented 16%; PRC/export policy, Taiwan manufacturing concentration, and newer production-silicon execution risks are all filing-confirmed caveats.

That makes ARM a strong Aether case study: the AI claim is real, but the exact revenue attribution, named customer economics, and valuation upside still need disciplined source boundaries.

What changed

The biggest change is that ARM’s AI infrastructure story is no longer only strategic language in an annual report. It now appears repeatedly in filed quarterly earnings exhibits.

ARM’s FY2026 Q4/FY2026 6-K exhibit reported record quarterly revenue of $1.49B, with licensing revenue up 29% YoY to $819M and royalty revenue up 11% YoY to $671M. The same filed exhibit said royalty growth was driven by smartphones, Edge AI, Physical AI, and Cloud AI, and that data-center royalty more than doubled year over year.

For the full year, the filed Q4 exhibit reported record FY2026 revenue of $4.92B, royalty revenue of $2.61B, up 21%, and licensing revenue of $2.31B, up 25%.

Evidence label: confirmed SEC fact for the reported revenue, royalty, and licensing figures. ARM’s causal wording around Cloud AI, Edge AI, Physical AI, and data-center royalty momentum is SEC management claim / forward-looking context because it is company language filed with the SEC. The inference is that AI infrastructure demand can increase ARM royalty/licensing opportunities; the filings retrieved do not provide an exact AI revenue line.

Financial growth: real numbers, no clean AI revenue percentage

ARM’s FY2026 growth is filing-confirmed:

  • FY2025 revenue: $4.007B.
  • FY2026 revenue: $4.920B.
  • FY2026 net income attributable to parent: $904M.
  • FY2026 R&D expense: $2.776B.
  • Remaining performance obligation at 2026-03-31: $2.071B.
  • FY2026 quarterly revenue sequence: $1.053B in Q1, $1.135B in Q2, $1.242B in Q3, and $1.490B in Q4.

The year-over-year revenue step from $4.007B to $4.920B is meaningful, and the filed 6-K exhibits repeatedly connect royalty momentum to data center, AI, CSS, and Armv9 adoption.

But this is not the same as a filing-confirmed AI revenue percentage. The clean evidence boundary is:

  • Confirmed SEC fact: total revenue, royalty revenue, licensing revenue, R&D, net income, RPO, customer concentration, and filed management language about data-center/AI royalty drivers.
  • Inferred relationship: AI workload growth can increase demand for ARM IP, CSS licenses, Armv9 adoption, and royalties on Arm-based chips.
  • Needs source: exact AI revenue percentage, exact data-center revenue percentage, customer-by-customer AI economics, or a filing-backed fair-value target.

Customer and end-market signals

ARM’s filings support broad AI end-market exposure, not a clean named hyperscaler revenue map.

The latest 20-F says the Arm platform runs software across smartphones, tablets and PCs, data centers and networking equipment, and vehicles. It also says Arm CPUs already run AI workloads from the edge to the cloud, including AI data centers, and that CPUs are vital in AI systems either handling workloads directly or working with co-processors such as GPUs and NPUs.

The FY2026 quarterly exhibits add end-market signals:

  • Q1 FY2026: royalty revenue growth from all target end markets, including data center, automotive, smartphones, and IoT.
  • Q2 FY2026: royalty revenue growth from all target end markets, including smartphones, data center, automotive, and IoT.
  • Q3 FY2026: royalty revenue growth driven by AI and general-purpose data center, smartphones, physical AI, and edge AI; the exhibit also said triple-digit data-center royalty growth illustrated ARM’s growing role in modern data-center architectures.
  • Q4 FY2026: data-center royalty more than doubled year over year.

The CSS adoption data is particularly important. ARM’s filed 6-K exhibits said CSS licenses grew to 16 licenses across 10 companies in Q1 FY2026, 19 licenses across 11 companies in Q2, and 21 licenses across 12 companies by Q3, with data-center CSS licenses specifically cited in Q1 and Q2.

Evidence label: confirmed SEC fact / SEC management claim for the filed end-market and CSS-license statements. Naming specific hyperscalers or assigning exact customer revenue remains needs source in this draft.

Supplier-chain and dependency signals

ARM is not a classic fabless chip seller with the same inventory and manufacturing exposure as a GPU vendor. Its historic model is IP licensing and royalties, so much of its supply-chain dependency is indirect: customers design and manufacture Arm-based chips, and ARM earns license fees and royalties.

That distinction matters. It is the reason ARM can participate in AI infrastructure growth without showing a GPU-style revenue line.

But the latest filing also adds a newer execution boundary. ARM’s 2026 20-F says that, unlike its IP licensing model, the company now faces risks around transitioning to advanced process nodes, competing against established production silicon companies, coordinating multi-year roadmaps with lead development partners, managing product transitions and inventory, maintaining software compatibility, and executing on advanced packaging, memory integration, and system-level optimization. The filing explicitly says rapidly evolving AI workload requirements create additional uncertainty around product-market fit and technical requirements.

The broader supply-chain risk is also filing-confirmed. ARM’s 2026 20-F notes semiconductor manufacturing concentration in East Asia, particularly Taiwan, and warns that geopolitical escalation could disrupt chip manufacturing.

Evidence label: confirmed SEC fact for the disclosed production-silicon and East Asia/Taiwan risk language. The conclusion that ARM’s AI upside depends on customers’ manufacturing and deployment capacity is an inferred relationship.

Risks and caveats

The filing-grounded bull case is credible, but the risk stack is not abstract:

  • Valuation/rerating risk: ARM’s retrieved one-year stock move was +224.34%. A lot of AI optimism may already be embedded in expectations.
  • Customer concentration: ARM’s top five customers accounted for 57% of FY2026 revenue, and Arm China accounted for 16%.
  • Arm China / PRC risk: ARM’s filings describe Arm China as a sublicensing route into China, while also warning about PRC semiconductor cycles, export controls, trade policy, restricted lists, tariffs, and other national-security policies.
  • Competition risk: ARM competes against x86, RISC-V, proprietary customer architectures, and the possibility that customers develop their own processors based on ARM ISA.
  • Supply-chain/geopolitical risk: Taiwan and East Asia manufacturing concentration can affect the chips that generate ARM royalty economics.
  • Production-silicon execution risk: ARM’s move beyond pure IP licensing adds advanced-node, packaging, memory integration, inventory, roadmap, and product-market-fit risk.
  • Evidence limitation: Aether transcript search failed with HTTP 500 / Vespa 400 in the parent research run, so this draft does not use management transcript quotes.
  • Forecast boundary: The retrieved filings do not provide exact AI revenue, exact data-center revenue, or a valuation target.

The right framing is not “ARM is a buy” or “ARM is just hype.” The stronger framing is: ARM has SEC-confirmed AI infrastructure exposure through royalty/licensing economics, but the market has already rerated the stock and the filings do not quantify exact AI revenue.

Valuation and upside/downside evidence boundary

This draft should not pretend the filings answer valuation on their own.

The one-year stock performance gives the context: ARM’s close-to-close move from $126.055 to $408.850 was a +224.34% gain. That is evidence of a large market rerating, not proof of future upside.

The filings support a directional bull case: rising FY2026 revenue, growth in royalties and licensing, data-center royalty momentum, CSS adoption, Armv9/higher royalty-per-chip language, and AI workloads from edge to cloud.

The filings also support a disciplined downside checklist: customer concentration, Arm China, PRC/export controls, Taiwan manufacturing concentration, competition, and production-silicon execution risk.

Evidence label: confirmed market data for the stock performance and confirmed SEC fact / SEC management claim for the filing items. Any target price, fair value, multiple, or upside/downside percentage beyond the historical one-year move is needs source or model output and should be explicitly labeled before publication.

How Aether changes the workflow

A normal AI-stock screen can tell investors that ARM is tied to AI. Aether’s SEC-first workflow makes the claim usable:

  • 20-F: business model, AI workload positioning, end-market language, customer concentration, Arm China, trade policy, Taiwan manufacturing concentration, competition, and production-silicon risk.
  • 6-K exhibits: quarterly revenue, royalty/licensing growth, Cloud AI / Edge AI / Physical AI language, data-center royalty momentum, CSS license adoption, and Armv9/higher royalty-rate language.
  • SEC companyfacts: revenue, net income, R&D, RPO, and quarterly revenue values.
  • Market data: one-year stock performance, which changes the research question from “is there AI exposure?” to “how much is already priced in?”

That is the EvidInvest angle: not generic AI narratives, but evidence boundaries that say what is confirmed, what is inferred, and what still needs a source.

Bottom line

ARM’s AI infrastructure story is real, but it is a royalty/licensing story — not a direct GPU revenue story.

The filings confirm that ARM is exposed to AI workloads from edge to cloud, including AI data centers. They also confirm strong FY2026 financial growth, repeated filed statements about data-center/AI royalty momentum, rising CSS adoption, and a value-capture mechanism through Armv9, CSS, licenses, and royalties.

The same filings keep the downside honest: customer concentration, Arm China, PRC/export policy, Taiwan manufacturing concentration, competition, and production-silicon execution risk.

My evidence-grounded classification:

ARM is a high-quality AI infrastructure royalty/licensing exposure with filing-confirmed data-center momentum, but after a 224% one-year stock move, the upside case needs stronger valuation evidence than the filings alone provide.

That is the gap EvidInvest and Aether are built to close: not “AI mentioned,” but “what exactly did the filings prove?”

Source notes

Fair Value Weekly

Get DCF breakdowns, fair value updates, and portfolio ideas for serious investors. No spam, no paywalled teasers.