NextFin News - Arm Holdings lifted its confidence in a revenue stream that did not exist a year ago, telling investors it now sees up to $2 billion from its own Arm AGI CPU silicon across fiscal 2027 and 2028, even as it guided third-quarter revenue to $1.54 billion. The signal, upgraded from the "more than $1 billion" line of sight the company offered at its March Arm Everywhere event, arrived alongside a second-quarter beat that pushed revenue 27% higher to $1.44 billion and licensing 50% higher to a record $773 million. Yet the stock fell 5.3% to $191.80 on November 4, extending a decline of roughly 16% from the October 29 close of $228.77. The divergence frames the question this earnings cycle forces investors to answer: is Arm's move into selling its own silicon the next leg of a structural data-center share shift, or the start of a lower-quality revenue mix that deserves a cheaper multiple?
The Quarter: A Beat Built on Licensing Timing, Not Royalty Acceleration
Arm's second quarter of fiscal 2027, which ended September 30, delivered revenue of $1.443 billion against guidance of $1.38 billion plus or minus $50 million. It was the fifth consecutive quarter of year-over-year growth exceeding 20%, a streak the shareholder letter led with. But the composition of that beat matters more than the headline, because Arm earns money in two fundamentally different ways: chipmakers pay up front to license its designs, and then pay a small fee on every chip they ship. This quarter, the up-front money did the heavy lifting.
License and other revenue rose 50% to $773 million, a record for a second quarter and roughly $100 million above what management had anticipated, according to the earnings call. Royalty revenue, by contrast, grew 8% to $670 million. Data-center royalties doubled year over year, and royalties from what Arm calls physical AI - chips in cars and robots - rose 50%, but the smartphone book, which still anchors the royalty stream, was weighed down by memory-chip prices that have pressured handset demand. Annualized contract value, the yearly value of signed license deals, rose 12% to $1.79 billion from $1.60 billion a year earlier. Partners shipped more than 10 billion Arm-based chips in the quarter, the first time that threshold has been crossed, 1.5 billion more than a year ago.
Profitability told two different stories depending on the accounting lens. Non-GAAP earnings per share rose 33% to $0.52, with non-GAAP operating income up 37% to $638 million on operating expenses of $774 million. Under standard GAAP rules, net income fell 45% to $130 million, with diluted earnings of $0.12 compared with $0.22 a year earlier. The gap between the two measures is the cost of Arm's expansion: stock-based compensation, acquisition-related items, and the mounting R&D required to build a silicon business from scratch.
For the current quarter, Arm guided revenue to $1.54 billion plus or minus $50 million, up 24% year over year. License and other revenue is expected to rise a little over 50%, while royalty revenue is guided to grow low teens percent sequentially and only low single digits year over year. Non-GAAP operating expenses are seen at about $825 million, with non-GAAP EPS of $0.56 plus or minus $0.04. The guidance confirms the pattern: licensing carries the near-term growth, while the annuity-like royalty stream grows modestly.
The $2 Billion Signal: What It Buys, and What It Does Not
The AGI CPU is Arm's first move into selling its own silicon rather than licensing designs. Launched in March 2026 as a processor built specifically for agentic AI workloads, it initially came with a "line of sight" to more than $1 billion in revenue across fiscal 2027 and 2028. Nine months later, that has become a confidence statement.
Our confidence in achieving up to $2 billion in Arm AGI CPU revenue over FY27-28 has increased over the last 90 days, supported by additional supply and customer mix.
Purchase orders from multiple customers now extend through calendar 2027, and first commercial deliveries remain on track for fiscal fourth quarter 2027, the January-to-March 2027 quarter.
Two numbers put the signal in scale. First, it is a two-year figure, not a run rate. Evenly split, it would imply about $1 billion a year against a company that already does $1.44 billion per quarter - meaningful, but not transformative on its own. Second, the margin profile is different from the business investors bought. Chief Financial Officer Jason Child put first-generation gross margins in the 30% to 40% range, against a company-level non-GAAP operating margin of 44.2% this quarter. Silicon carries supply-chain, manufacturing, and inventory cost that a license dollar never sees. When management says margins should reach the market standard of 50% or more on future versions, it is acknowledging that the first version is a foothold product priced to win design share.
The strategic value lies elsewhere. Oracle Cloud Infrastructure plans to offer Arm AGI CPU instances in 2027, placing Arm-designed silicon on the same platform as partner-designed Arm CPUs from Google, Amazon, Microsoft, and NVIDIA. That gives hyperscalers an internal benchmark: if Arm's own chip delivers roughly twice the work per server rack versus AMD's EPYC 9755, as independent Signal65 testing at the Open Compute Project summit showed, then the price hyperscalers pay for partner Arm designs has a performance anchor. The AGI CPU is not just a new revenue line; it is a pricing and positioning tool for the core licensing business.
The Margin Question the Market Priced in Real Time
The 5.3% decline on a beat is the market asking whether Arm's next dollar of revenue is worth less than its last one. The concern is not that the AGI CPU will fail. It is that a company valued as a high-margin platform licensor is adding a lower-margin silicon vendor inside itself, and that mix shift is dilutive before it is accretive. In the same quarter management called "great," GAAP net income fell nearly in half. Royalty growth - the line investors watch for evidence that AI demand is compounding through shipped units - came in at 8%, and the Q3 guide calls for only low-single-digit year-over-year royalty growth. Acceleration came from licensing timing and a new silicon line, not from the annuity.
There is also a customer-relationship tension that the bullish case has to navigate. Arm's biggest data-center customers are also its licensees: Google builds Axion, Amazon builds Graviton, Microsoft builds Cobalt, and NVIDIA builds Vera on Arm architecture. Arm-based server market share at the top hyperscalers more than tripled, from 15% in calendar 2024 to more than 50% in calendar 2025, and the company expects it to exceed 60% by the end of calendar 2026. Selling a competing CPU into the same data centers where those licensees are deploying their own custom silicon is a delicate act. Management's answer, embedded in the Oracle announcement, is that the AGI CPU complements rather than displaces partner designs - a common platform and software ecosystem rather than a zero-sum fight. Whether hyperscalers treat it that way will show up in whether Oracle is joined by other instance announcements in 2027.
Supply is the other hinge. When Bank of America Securities analyst Vivek Arya asked on the call how a first-time chip vendor secures capacity against AMD's long-standing TSMC relationship, CEO Rene Haas pointed to decades-old ties with TSMC, Micron, Samsung, and SK hynix.
We have been working on this for over a year. It is a 5 to 10-year endeavor.
That timeline is honest - and it is the strongest argument against expecting the $2 billion to arrive smoothly. A 5-to-10-year build means quarters where supply, not demand, sets the revenue ceiling.
The Structural Case: Agentic AI Is CPU-Bound
Beneath the multiple compression sits a sturdier argument, and it is the one that makes the $2 billion signal directional rather than merely additive. The data-center share shift is already happening at a pace that looks structural, not cyclical. Neoverse shipments have passed 1.5 billion cumulative cores, with the most recent 500 million shipping in nine months; the first billion took six years. That is not a demand spike that mean-reverts when inventory normalizes - it is a design-win curve compounding across generations of custom silicon.
The mechanism is the shift from prompt-driven AI to persistent agents. Agents execute software, call tools, and coordinate work between accelerator invocations. That orchestration runs on CPUs, and it scales with the number of agents, not the size of the model.
Agentic demands on AI jobs are CPU-bound, and the more agentic workloads that run, that is going to need more CPUs.
He expects core counts to rise from 128 to 256 to 500-plus, which drives higher average selling prices per socket. Haas sees the data-center CPU market reaching $100 billion by 2030, possibly higher.
Power availability sharpens the argument. Morgan Stanley has projected a 38-gigawatt global power shortfall through 2028, and the Signal65 testing showing roughly 2x performance per rack versus AMD's EPYC 9755 implies Arm-based racks could cut that shortfall roughly in half. In a power-constrained buildout, performance per watt becomes the binding constraint, and that is Arm's historical advantage. The AGI CPU lets Arm monetize that advantage directly instead of only through licensing fees.
This is the structural call: the data-center transition is a regime shift in CPU architecture share, and agentic AI makes CPUs, not just accelerators, the scarce resource. The cyclical leg - smartphone unit softness, memory-driven handset headwinds, the timing of license signings - will revert. The structural leg - Arm's embedded position inside hyperscaler custom silicon and the power-efficiency imperative - will not revert on its own. The two are moving in opposite directions right now, which is exactly why the quarter reads as both strong and disappointing.
The Counter-Thesis: Premium Valuation, Execution Risk, and Legal Overhang
The strongest case against the bullish read is not technical. It is that Arm already trades as if the structural transition is largely secured. The stock has changed hands as high as $452.70 over the past year and still carries a market capitalization of $204.84 billion, well above its 50-day average of $257.07. At that price, years of flawless execution are already embedded. The counter-thesis says the margin dilution from silicon is permanent rather than transitional: once you become a chip vendor, you carry inventory risk, supply commitments, and customer concentration that a pure licensor never faces, and the market is right to compress the multiple on the blended business.
There is also litigation that strikes at Arm's highest-quality revenue. Last month, a Delaware jury found that Arm breached parts of its licensing agreement with Qualcomm and interfered with Qualcomm's relationship with Meta, awarding only $2 in damages but leaving disputes over royalties and future licensing terms to be decided by a judge. An adverse ruling on future terms would affect the license and royalty stream that currently generates the bulk of Arm's profit - not the new silicon line at all.
And the AGI CPU faces a credible incumbent. AMD's EPYC franchise is deeply embedded in hyperscaler and enterprise racks, with long-standing TSMC capacity relationships. Arm's 2x-per-rack claim rests on independent testing of a specific configuration; real-world qualification cycles at cloud providers run many months, and a slip in fiscal Q4'27 deliveries would hand skeptics their first concrete evidence that the 5-to-10-year timeline is the realistic one, not the $2 billion signal.
What Comes Next: Three Scenarios and the Signal That Would Break the Thesis
The base case is that first commercial AGI CPU deliveries land in the January-to-March 2027 quarter as scheduled, Oracle qualifies its instances, and licensing strength converts into reaccelerating royalty growth through fiscal 2027 as those designs enter production. In that path, the margin question gets answered with real data rather than guidance, and the stock stabilizes as investors separate the cyclical smartphone leg from the structural data-center leg.
The upside case runs through core counts and power. If agentic workloads push sockets toward the 256-to-500-plus range Haas described, data-center share exceeds 60% ahead of the end-of-2026 target, and AGI CPU gross margins reach the 50%-plus target faster than expected, the multiple re-expands toward prior highs. The $2 billion signal would then look like the floor of a much larger silicon franchise.
The downside case is a de-rating that compounds. First deliveries slip, silicon margins land at the low end of the 30%-40% range and stay there, royalty growth remains in low single digits, and the Qualcomm judgment tightens future licensing terms. In that path the market stops treating Arm as a platform licensor and starts pricing it as a lower-margin silicon vendor with a legal overhang.
Four items to watch, in order: the fiscal fourth quarter of 2027, when the first commercial AGI CPU deliveries and any associated revenue should appear; quarterly royalty growth, where a reacceleration above the low-single-digit Q3 guide would confirm that licensing is converting into shipped units; the Qualcomm judgment on future licensing terms; and hyperscaler instance announcements beyond Oracle. The falsifying signal for the structural share-shift thesis is specific and observable: if Arm-based server share at the top hyperscalers fails to exceed 60% by the end of calendar 2026, or if data-center royalty growth does not remain in triple-digit year-over-year territory through fiscal 2027, the "agentic AI is CPU-bound" narrative loses its revenue evidence and the structural call breaks.
Arm's $2 billion signal is less about the dollars than the direction of travel. The company that once made money only when others shipped chips is now shipping its own, and the market must decide whether that makes Arm bigger - or just different.
Explore more exclusive insights at nextfin.ai.