3Q FY26: Record Demand, Memory Takes the Margin

Thesis status: No tracked condition weakening. The variable that moved the quarter sits outside the condition set.

Thesis Refresher

This is the first update to the Apple capital-light primer published 16 June 2026. Third quarter results, filed 30 July 2026, are the first quarterly assessment of the primer's seven conditions, and this memo walks all seven.

The question under evaluation is whether Apple ($AAPL)'s capital-light model can deliver the earnings its valuation requires. Hardware generates the installed base, services extract rent from it at gross margins more than double the hardware that created it, and effectively all free cash flow returns to shareholders through a buyback that compounds EPS independent of revenue. The primer set that model against a price embedding 11% annual free cash flow growth for a decade, over a FY2026 to FY2028 horizon with FY2029 as the sustained-model validation point.

3Q FY26 in Brief

Tariff refunds contributed approximately 200 bps to consolidated gross margin, over 250 bps to products gross margin, and $0.11 of diluted EPS. Excluding them, consolidated gross margin was 48.1%. Apple recognises refunds received as a reduction of products cost of sales.

The consolidated margin expanded 360 bps in a quarter where the primer's stated mechanism for margin expansion ran backwards. Services fell to 28.1% of revenue from 29.2%, because products grew faster. Section II of the primer put a 1 point rise in services share at roughly 40 bps of consolidated margin, so mix subtracted this quarter rather than added. What replaced it was a tariff refund management has already guided down to roughly 100 bps next quarter, and a product mix that the filing attributes to Pro models. Beneath both, the ex-refund margin of 48.1% is down from 49.3% in the March quarter.

The Memory Pass-Through

The primer built its margin risk around regulation. Its seven conditions track antitrust remedies, commission mandates, AI capability, China and a leadership transition, and not one of them tracks what Apple pays for components. In the June quarter the component line became the largest single claim on the margin structure.

Management treats the March figure as the comparable base and did not adjust it. September guidance of 47% to 48% includes an expected refund benefit of approximately 100 bps.

Management quantified the attribution twice and landed in the same place both times. On the 120 bps step from March to June excluding refunds, CFO Kevan Parekh stated that "more than 100%" of the change is explained by the memory cost change. On the 160 bps step from June to the September midpoint, he stated that more than that is explained by the same factor. Foreign exchange, the obvious alternative in a quarter carrying a 2.5 point currency headwind, he characterised as "a pretty minimal impact when you look at that versus the memory."

The direction of travel is set by a price Apple does not control. Cook disclosed the sequence: more in March than December, significantly more in June than March, more again expected in September, and past that point "we see the market pricing for memory continuing to increase, which could drive an increasing impact on our business." He characterised the episode as a 100-year flood, with exponential increases.

Three offsets are holding realised compression below its gross rate, and management has flagged that the largest of them decays. Carry-in inventory absorbed part of the June increase and will absorb part of September's, with a decreasing benefit beyond the September quarter. Non-memory component costs are falling. Mix is favourable. None of the three is a durable hedge against an input still repricing upward.

Apple has moved to price rather than absorb. Cook stated the company "reluctantly raised prices" on iPad and Mac, weighing units, revenue and margin over the long term rather than applying a formula. Elasticity is unresolved, and he stated it is "too early to come to a definitive conclusion" while channel inventory delays the consumer response.

DRAM has three suppliers. Cook stated that more of them "would help us on the supply side and perhaps the pricing side," that the pricing benefit is unclear, and that Apple is evaluating all options. The outsourced manufacturing model the primer identifies as the source of Apple's capital efficiency is the same structure that leaves it a price-taker on a three-supplier input. Samsung, the capital-heavy comparison the primer runs in Section IV, owns its memory fabrication.

Sized against the conditions the primer does track, the pass-through is not a second-order item.

The 280 bps is the net ex-refund consolidated margin decline from the March actual to the September midpoint, sized on LTM revenue of $466.8B and taxed at 17.5%, held against the price-implied growth rate. Management attributes more than the whole of that net decline to memory, so the gross memory headwind is larger and other components offset part of it. C1 and C2 carry the primer's own assumptions, refreshed to the LTM free cash flow base of $136.7B.

The comparison is not like for like, since the regulatory conditions remove revenue permanently while memory pricing is cyclical and has mean-reverted in every prior episode. What it establishes is order of magnitude, and on that basis an unmeasured cost line is running at more than twice a modelled load-bearing failure.

The second leg is availability rather than price. Cook located the June constraint on Mac primarily and attributed it to the availability of the advanced nodes Apple's SoCs are produced on, putting the cause on demand outrunning Apple's own forecast rather than on any supplier failure. That is a better problem than weak demand. It still caps revenue. September guidance of 9% to 11% against 16.4% delivered carries roughly 2.5 points of currency headwind, leaving the balance to supply.

The balance sheet corroborates the commentary. Inventories nearly doubled to $11.1B from $5.7B at the FY2025 close. Manufacturing purchase obligations rose to $57.0B from the $44.6B the primer recorded, with $56.2B of that payable within twelve months. Apple is buying forward into a rising market, and Cook confirmed the limit on it: "We've been pulling supply ahead. At some point, there's a limit to that."

What the Price Now Embeds

Apple fell 8.7% on 31 July, the session after results. What follows characterises what the price implies about the free cash flow trajectory rather than producing a price target or a formal valuation.

The primer row carries its inputs as published. The two July rows run on the refreshed base: LTM free cash flow of $136.7B, net cash of $62.2B, 14.75B diluted shares, WACC of 8.85%, terminal growth 3.0%.

Apple rose 14.2% between publication and the print. That run-up, rather than the drop that followed it, is where the valuation actually moved. It carried the implied growth requirement from 11.29% to 12.32%, which is $61B of additional year-10 free cash flow demanded without any corresponding increase in the disclosed free cash flow base. The print handed that requirement straight back.

Against the primer the post-print requirement is 0.18 points lower, two effects running opposite ways. The higher price adds 0.55 points, and free cash flow 5.8% higher at $136.7B subtracts 0.72. Apple trades 4.2% above the primer's observation price and asks slightly less of the future than it did at publication.

Holding the pre-print growth rate constant at 12.32% and varying only the margin isolates what the drawdown is consistent with.

Under an unchanged 12.32% growth requirement and fixed discount rate assumptions, the closing price is numerically consistent with roughly 310 bps of permanent gross margin compression. That is a consistency result rather than a decomposition. The same move could carry lower expected revenue growth, the services deceleration, the supply constraints, a different risk premium, or positioning after the summer run-up. It does not carry the market: the Nasdaq closed 31 July at roughly 25,372 against 25,340 at the open, up 0.13% on the session, so Apple underperformed by close to nine points and the drawdown is company-specific. What the exercise establishes beyond that is that no downgrade to the growth trajectory is required to reach the closing price, and that a market expecting memory to mean-revert on its historical pattern would sit nearer the 160 bps case.

Management's own framework is the cross-check either side of that. Ten percent revenue growth for five years decelerating to 7%, with the free cash flow margin reaching 32% on services mix, produces $272.57. The 11.8% gap to market compares with 11.2% at the primer's observation price, so the ecosystem durability premium came through the drawdown intact.

The longer cross-check is less comfortable.

Apple's free cash flow compounded at 3.50% across the decade to FY2025 and declined in five of those ten years. Both endpoints flatter that number. FY2015 was the iPhone 6 peak with revenue up 28%, and FY2025 absorbed a $17.3B increase in cash taxes on the Ireland State Aid settlement, larger than the year's $10.0B decline. On three-year averages either side, which strips both distortions, the realised rate is 7.34%. The price asks 11.11%. The gap survives every basis, and its magnitude is basis-dependent, at 3.2x point to point and 1.5x on averages, across a decade that contained the iPhone X supercycle, the services build-out and the largest buyback programme in corporate history.

Watch Condition Assessment

C1 — Google TAC preservation or offset

⏸ Dormant → Neither the filing nor the call mentions the agreement.

Neither the 10-Q nor the earnings call references the Google agreement, the DOJ remedy or the search default. That is the expected state before a remedy hearing expected in 2H 2027. The only on-condition evidence sits on the offset leg, where the 10-Q attributes services growth "primarily to higher net sales from advertising and cloud services," naming advertising first. The strengthening trigger requires more than $5B of incremental annual advertising revenue and Apple discloses none, so a record on an undisclosed base gives direction without magnitude.

C2 — Services margin durability above 70%

🟢 Affirmed → 76.3% over nine months, clearing the strengthening trigger.

Services gross margin was 75.6% in the quarter, flat year over year, and 76.3% across nine months against 75.5% a year earlier. That clears the 70% floor by 630 bps and satisfies the strengthening trigger, which requires above 74% for two or more consecutive quarters. The 110 bps sequential decline Parekh attributed to mix falls inside the noise the primer specified, which discounts single-quarter movements driven by content timing.

The procedural position improved against the primer's assumption that the linked-out injunction was settled at a 27% commission, with the Supreme Court agreeing to hear Apple's appeal. That is optionality rather than an outcome. Alongside App Store model changes in certain countries, the App Store still set a June quarter revenue record, which is the pattern Three Rents predicts. Regulation opens the gate, Apple reprices the equipment, and the blended margin absorbs jurisdiction-level intervention without visible compression.

The pressure on services arrived somewhere the condition does not look. Growth decelerated from 16% in March to 12% in June, guided to roughly 9.5% for September, on foreign exchange first, then the absent F1 theatrical comparison, then mobile gaming softness. Margin is holding while growth slows, the opposite composition to the one this condition was written to catch.

C3 — AI-driven upgrade cycle materialisation

🟡 Developing ↑ Outcome far ahead of the threshold, attribution not yet testable.

iPhone revenue grew 21.7% in the quarter and 22.4% across nine months against a pass threshold of 8% in FY2027 and FY2028. The 10-Q attributes the growth "primarily to higher net sales of Pro models," which inverts the primer's weakening trigger. That trigger anticipated flat or declining ASPs as buyers chose non-Pro models lacking full AI capability. Apple also set a June quarter record for upgraders, reported the iPhone active install base at an all-time high, and gained global share per IDC.

The outcome runs far ahead of the threshold and the mechanism stays untested, which is why this reads developing rather than affirmed. Siri AI was unveiled at WWDC during the quarter and ships in the autumn, so the iPhone 17 cycle producing this growth predates the AI capability the condition names as its cause, and the formal test period is FY2027 and FY2028. Asked directly about pull-forward, Cook stated that a 22% rate sustained year to date makes it "not obvious in the data" that demand has been borrowed.

Two developments complicate the forward read. The iPad and Mac price increases carry elasticity risk that stays unreadable until channel inventory clears. And the ceiling is now set by supply rather than order intake, with September iPhone growth guided to mid-teens against 21.7% delivered.

C4 — Capital return execution at $90B+ annually

🟡 Developing ↓ $82.2B trailing, below the pass and well above the fail.

Management stated $25.8B for the June quarter. The cash flow statement basis used here gives $25.1B, on settlement timing.

The shortfall is concentrated rather than structural. Trailing twelve-month repurchases of $82.2B run $7.8B below the required pace, and the March quarter alone came in $10.2B under it, which accounts for the entire gap and more. June returned above pace at $25.1B.

The composition matters more than the total. Nine-month repurchases of $62.1B are 12% below the prior year's $70.6B, in a period when nine-month free cash flow rose 52% to $110.2B. Apple returned 72% of trailing free cash flow against 107% in FY2025.

The balance sheet shows where the difference went. Net cash rose to $62.2B from $33.8B at the FY2025 close, as cash and securities grew $14.1B while total debt fell $14.3B with no new term debt issued. Both legs moved away from the net cash neutral framework the primer identifies as an executive-level commitment, and that framework was not restated on the call. The funding leg of the condition is intact, with TTM free cash flow of $136.7B clearing the required $110B with room. Deployment is what changed.

C5 — China installed base retention

🟢 Affirmed ↑ Nine months already past all of FY2025, and Apple Intelligence approved.

Greater China revenue grew 22.4% in the quarter and 30.0% across nine months, and at $64.8B for nine months has already passed the $64.4B the segment generated across the whole of FY2025. Apple set a June quarter revenue record in China Mainland, an all-time Mac revenue record in Greater China, and iPhone was the top-selling model in urban China per Worldpanel.

Two strengthening triggers fired. The primer wrote one in revenue terms, growth above 10% year over year for two or more consecutive quarters, now met three times over at 37.9%, 28.1% and 22.4%. The second is the China approval: Cook disclosed that Apple received clearance in the week before the call to ship the original Apple Intelligence features, with "more work required down the road for Siri AI." Partial, and the first movement on a gap that had run twenty months, which pushes back the weakening trigger written around approval slipping beyond FY2027.

Two qualifications sit against the green. The 10-Q attributes part of the growth to renminbi strength, so the underlying volume gain is smaller than reported. And Apple disclosed no China-specific installed base number, only the global 2.5B active devices, leaving the condition's actual subject unobserved for another quarter.

C6 — Edge-first AI competitiveness

🟡 Developing → Siri AI in beta with no engagement metrics, and the DMA now blocks the feature itself.

Siri AI was unveiled at WWDC during the quarter and released to developer and public betas, with general availability in the autumn. Cook characterised reception as overwhelmingly positive, and Apple again disclosed no engagement metrics for Apple Intelligence, the gap the primer flagged against Gemini's 750 million monthly actives.

The regulatory exposure changed shape. Siri AI will not launch in the European Union at rollout, and Cook stated that it "is offered or will be offered for the Mac there, because the Mac is not covered by the same regulations as the iPhone and the iPad." The DMA is now gating a product capability rather than a fee. Section V of the primer frames regulation as an intervention at the distribution gate that Apple reprices at the runtime layer. A rule that stops the feature shipping is a different mechanism, and it lands on the two device categories that carry the installed base.

The cost side is unresolved on management's own account. Cook stated "I don't want to say that we have a complete plan for that" on Siri AI compute costs, with an unsized iCloud+ upsell as the recovery route, and on capital intensity described a hybrid of third-party cloud and Apple's own data centres, noting "there are other locations on the P&L other than OpEx, like COGS etc., that also have AI expenditures." On the reported lines the model looks more capital-light than ever, with capex down 29.1% and trailing intensity at 2.15% of revenue against 3.06% in FY2025. The qualification is management's own.

C7 — Leadership transition continuity

⏸ Dormant → Cook's final call, and no capital allocation statement from Ternus.

The transition sits before its first test. On Cook's final earnings call Ternus spoke once, in roughly forty words, on competitive threats from OpenAI and SpaceX devices, with no statement on capital allocation, buyback pace or the net cash neutral framework. The November call remains the initial test. The one observation touching this condition sits in C4, where net cash nearly doubled while the buyback ran below pace, both under the outgoing CEO. Whether that is timing or the front edge of a reallocation is what November answers.

Thesis Standing

The capital-light model did what the primer said it does, at a higher rate than the primer assumed, on an installed base of 2.5B active devices against the 2.2B recorded at publication. Two conditions are affirmed, three developing, two dormant before their first testable dates, and none weakening. On its own terms the thesis is in better shape than it was in June.

The framework has a blind spot, and this quarter walked into it. All seven conditions are written around revenue. C1 and C2 track revenue regulation could remove. C3 and C5 track demand for hardware. C6 tracks the capability sustaining that demand. C4 and C7 track the disposition of the cash that results. Not one of them tracks what it costs Apple to build the products. That was defensible while component prices were stable and the whole point of the capital-light model was that Apple had moved its cost base onto somebody else's balance sheet.

Outsourced manufacturing converts a fixed cost into a variable one and removes the capital that would otherwise sit on Apple's own balance sheet. What it does not provide is a supply-side hedge. Apple can blunt an input shock through inventory, contracts, mix, pricing and the rest of the bill of materials, all of which it used this quarter, and it cannot capture the producer economics of the price it is paying. Samsung, which fabricates its own memory, sits on both sides of the cycle and carries the utilisation and capital-intensity risk that comes with it. The primer read that trade as unambiguously favourable, with ROIC above 50% against Samsung's 18% as the evidence. The June quarter is the first period in this coverage where the other side of the trade carries a number.

None of that breaks the thesis. A cyclical input cost belongs to a different risk class than the permanent revenue removal the load-bearing conditions describe. What the quarter establishes is narrower and more awkward: the framework cannot currently distinguish a thesis working from a thesis being squeezed.

Framework Modification

Proposed addition: component supply and cost pass-through. Two legs. On cost, whether Apple holds consolidated gross margin above a defined floor through the memory cycle, with pricing and mix as the offsets. On availability, whether advanced-node supply caps revenue growth below the rate demand supports. Thresholds should be set at the FY2026 restatement against a closed fiscal year rather than fixed here.

Proposed refinement: separate C3's outcome test from its attribution test. A condition that can pass without its stated mechanism operating cannot fail when that mechanism does not work. The revenue threshold should stand, with an attribution test alongside it, first readable in the December quarter once Siri AI has shipped.

One finding is logged without a proposal attached. The DMA now gates a product capability rather than a fee, which sits outside the taxonomy Section V of the primer rests on. Reworking that taxonomy is restatement work rather than a condition change.

What to Watch

The next ninety days carry more resolution than the last twelve months did. FY2026 closes in late October, the first Ternus call follows in November, and Siri AI ships between them.

Ternus's first earnings call, November 2026 C7's named test. Specifically whether the net cash neutral framework is restated, given net cash nearly doubled to $62.2B under the outgoing CEO while the buyback ran below its required pace.

December quarter, reported late January The first period in which Siri AI has shipped, the iPad and Mac price increases have cleared channel inventory, and the memory cost path can be read against September's actuals rather than management's forecast of them. :::

Three items run on longer clocks. Siri AI needs a separate China approval Cook described as at the front end of its process. The European Commission and Apple are negotiating terms that would let it ship on EU iPhone and iPad, with no timeline offered. And the Supreme Court schedule on the linked-out appeal now belongs to the court.

The memory cycle may influence the next four quarters more directly than any condition currently in the primer. Apple outsourced the balance sheet, not the bill.