← Events← 返回日历
EVENT BRIEF · 事件摘要与解读
Qualcomm FY26 Q3 Earnings高通 2026 财年三季度财报
2026-07-29
Earnings · 财报
Brief v1
OFFICIAL MATERIALS官方材料直达
KEY NUMBERS · SOURCE: 8-K FILING (◆ = MARKET CONTEXT, NOT FILING DATA · ⊘ = ASKED FOR, NOT IN THE FILING)核心数字 · 取自 8-K 备案(◆ 为市场数据非备案 · ⊘ 为我们想看但备案没给)
MARKET CONSENSUS IN ONE PARAGRAPH市场共识一段话
◆ Market context, not filing data: the street went into this print at roughly $9.7B of revenue and $2.22–2.23 of adjusted EPS, having cut the EPS number about 8–9% over the prior 90 days, with options implying a ~9% move in either direction. What landed is an awkward shape for a beat-hunting market — revenue came in above consensus and, per the CEO's own words, at the high end of guidance, while non-GAAP EPS of $2.21 sat at or just under the number. A four-quarter streak of small beats did not extend on the line that matters to the multiple. The sell-side agenda going into the call was automotive durability, handset timing and data center; the filing's own framing pushes the conversation somewhere else entirely — toward input costs the company says it will recover in pricing "over time," and toward a fiscal-2027 non-handset acceleration to greater than 60% that no automotive design-win curve can produce on its own. For an automotive audience the useful question on the call is not whether the quarter beat, but who pays for the memory and packaging inflation in a business where prices were agreed years before the cost curve moved.◆ 以下为市场数据,非备案内容:进入财报前,卖方大致停在营收约 $97 亿、调整后 EPS $2.22–2.23,且过去 90 天已把 EPS 预期下调约 8–9%,期权市场隐含双向约 9% 的波动。落地的组合对一个习惯「找超预期」的市场来说很别扭——收入高于共识、且按 CEO 原话处于指引区间高端,而非 GAAP EPS $2.21 只是持平甚至略低于共识。连续四个季度的小幅超预期,恰恰在决定估值倍数的那一行断了。会前卖方的议程是汽车可持续性、手机复苏节奏和数据中心;但备案自己的叙事把话题带去了别处——一是公司称将「随时间」通过定价回收的投入成本,二是 FY27 非手机业务加速到 60% 以上,而这个斜率不可能靠汽车定点转量单独产生。对汽车读者,电话会上真正有用的问题不是「有没有超预期」,而是:在一门价格早于成本曲线数年就谈定的生意里,存储与封装的涨价由谁承担。
AUTOSIGNAL READ · ANALYSISAUTOSIGNAL 解读
Editorial analysis built on primary data (this 8-K exhibit, our own signal archive incl. the Tesla, GM and Mobileye Q2 2026 briefs published earlier this month). Market and litigation items are marked ◆ and are not filing data. Not investment advice.基于一手数据(本份 8-K 附件、自有信号档案——含本月早前发布的特斯拉、通用、Mobileye 二季度简报)的编辑部分析。市场与诉讼类信息标注 ◆,不属备案数据。不构成投资建议。
①Revenue beat, earnings didn't: a −41% operating quarter收入超预期、利润没超:一个经营利润 −41% 的季度
Hold two lines side by side: revenue −4%, operating income −41%. Cost of revenues actually ROSE 1.4% on a shrinking top line (computed gross margin 53.1% vs 55.6%), R&D added 17% and SG&A 27%, and non-GAAP EBT margin fell from 34% to 27% — seven points in four quarters. The filing is unusually candid about the cause, naming a broad-based increase across wafer fabrication, assembly, test, advanced packaging, memory and other materials. Then note the remedy language: pricing actions are expected to "benefit our gross margins over time as the pricing changes gradually come into effect." No quarter, no basis points, no attach to guidance — a sentence engineered so that it cannot be missed next quarter. The Q4 guide is where the honest version shows up: revenue midpoint rises sequentially while non-GAAP EPS midpoint of $2.15 falls below this quarter's $2.21. Management is telling you, in numbers rather than adjectives, that cost recovery does not arrive inside the guidance window. The automotive consequence is specific and under-discussed: auto SoC pricing is locked into multi-year OEM agreements signed long before the memory cycle turned. Of Qualcomm's revenue streams, automotive is structurally the slowest to reprice and the fastest-growing — which means the mix shift the market applauds is also, at least for now, margin-dilutive. The filing does not segment cost inflation by revenue stream; that split is the first thing to demand on the call.
把两行数字并排看:收入 −4%,经营利润 −41%。收入在缩,营业成本反而 +1.4%(据备案计算的毛利率 53.1% vs 55.6%),研发 +17%、销管费 +27%,非 GAAP EBT 利润率从 34% 掉到 27%——四个季度掉 7 个百分点。备案对成因交代得罕见地直接:晶圆制造、封装、测试、先进封装、存储及其他材料全面涨价。但接着看它的「解法」措辞:提价预计将「随时间(over time)推移、在价格调整逐步(gradually)生效后」改善毛利率。没有季度、没有基点、不挂钩指引——这是一句被设计成「下季不可能被判定为落空」的话。诚实的版本写在 Q4 指引里:收入中值环比上升,非 GAAP EPS 中值 $2.15 却低于本季 $2.21。管理层用数字而非形容词告诉你:成本回收不会发生在指引窗口内。对汽车的含义具体且被低估——车规 SoC 的价格锁在多年前签的车企长协里,那时存储周期还没转向。在高通所有收入线中,汽车结构上最难重新定价,同时增长最快;也就是说,市场为之鼓掌的结构转移,至少眼下同时是毛利率的稀释项。备案没有按收入线拆分成本通胀——这个拆分是电话会上第一个该追问的东西。②Three layers of flattery — and this time GAAP is the flattered one三层美化——而且这次被美化的是 GAAP
Our GM brief flagged a $2.16 gap where adjusted EPS flattered a GAAP loss on write-downs. Qualcomm is the mirror image, and it is the rarer and more instructive case. GAAP EBT fell only 17% while non-GAAP EBT fell 24% — because QSI, the strategic-investment segment, contributed $768M of GAAP EBT and $0.57 of GAAP EPS, with investment and other income at $1,014M against $358M a year ago. Strip QSI and Q3 GAAP EPS is roughly $1.30. The company itself validates the read: the Q4 guidance table lists diluted EPS attributable to QSI as $—, and the Q4 GAAP midpoint of $1.32 lands almost exactly on this quarter's ex-QSI figure. So the single best-looking line in the GAAP column is the one management does not expect to repeat. Layer two is the nine-month view, where net income of $12,377M against $8,658M looks like 43% growth and is substantially a $4,136M income tax benefit — deferred tax assets went from $743M to $5,679M, while the cash flow statement shows income tax payments exceeding the provision by $5,550M. Accounting benefit up, cash out the door; this exhibit does not explain the item's origin, and the 10-Q is where that gets read. Layer three is quieter: 2.7% of the diluted share count retired year over year, which is why net income −25% becomes EPS −23%. None of this is improper — QSI exclusion is exactly what the Non-GAAP note says it does. The discipline is simply to price the operating business off $2,776M of non-GAAP operating income and 27% EBT margin, and to treat both the investment gain and the tax benefit as what they are: real cash-and-book events with no bearing on whether Snapdragon Ride wins its next platform.
我们的通用简报标记过 $2.16 的差距:调整后 EPS 掩盖了减值造成的 GAAP 下滑。高通是它的镜像,而且是更少见、更有教学价值的一种。GAAP EBT 只跌 17%,非 GAAP EBT 跌 24%——因为战略投资分部 QSI 贡献了 $7.68 亿 GAAP EBT、$0.57 的 GAAP EPS,投资及其他收益从去年 $3.58 亿跳到 $10.14 亿。剔掉 QSI,本季 GAAP EPS 约 $1.30。公司自己给这个读法背书:Q4 指引表把「归属 QSI 的摊薄 EPS」列为 $—,而 Q4 GAAP 中值 $1.32 几乎正好落在本季剔除 QSI 后的水平。也就是说,GAAP 栏里最好看的那一行,管理层自己不预期它会重演。第二层是九个月口径:净利润 $123.77 亿对去年 $86.58 亿,看着 +43%,实质上很大一块是 $41.36 亿的所得税收益——递延税资产从 $7.43 亿升到 $56.79 亿,而同期现金流量表显示实缴税款超出计提 $55.5 亿。账面收益向上、现金向外走;这份附件没有解释该税务项的来由,要读 10-Q。第三层更安静:摊薄股数同比减少 2.7%,所以净利润 −25% 变成 EPS −23%。这些都不构成不当——排除 QSI 正是非 GAAP 说明里写明的做法。真正需要的纪律很简单:给这门生意定价要用 $27.76 亿非 GAAP 经营利润和 27% 的 EBT 利润率;至于投资收益和税务收益,承认它们是真实发生的账面与现金事件,但与「Snapdragon Ride 能否拿下下一个平台」毫无关系。③$1,588M and +61%: the number is real, the explanation is missing$15.88 亿、+61%:数字是真的,解释是缺的
This is the most important automotive datapoint of the quarter anywhere in the industry, so it deserves both the credit and the caveat. Credit first: $1,588M in a single quarter, up 61%, now 18.7% of QCT versus 10.9% a year ago, extending 23 consecutive quarters of double-digit growth. Scale it against our archive — Mobileye's entire company revenue for a quarter ending one day earlier (June 27) was $508M and flat. Qualcomm's automotive line alone is now roughly 3.1× the pure-play ADAS champion's total business, and growing 61% while that champion is flat with a falling ASP. The center of gravity in ADAS and cockpit silicon has moved to the general-purpose compute vendor, and this quarter is where that stopped being a forecast. Now the caveat, and it is not small. The 8-K gives no organic-versus-acquired split, no Snapdragon Ride versus cockpit breakdown, no design-win pipeline figure and no automotive AI compute metric — every one of the four things the desk flagged in advance lives on the call or at Investor Day, not in this document. Meanwhile goodwill rose $2,916M since fiscal year-end on $1,573M of nine-month acquisition cash, and this exhibit allocates none of it to a revenue stream. A jump from a long "double-digit" cadence to +61% is a break in pattern that the filing does not explain, and until the 10-Q resolves it, +61% should be read as a number, not yet as a rate. One more piece of arithmetic worth banking: this quarter sets a $1,588M comparison base, so keeping the streak alive in Q3 FY27 requires more than $1,747M. Streak metrics are chosen because they sound unbreakable; the surest way to break one is an exceptional quarter.
这是本季全行业最重要的一个汽车数字,所以该给的肯定和该打的问号都要给。先说肯定:单季 $15.88 亿、同比 +61%,占 QCT 从去年 10.9% 升至 18.7%,把双位数增长的连续纪录延到 23 个季度。用我们自己的档案标定它的量级——Mobileye 截止日只早一天(6 月 27 日)的整个公司季度收入是 $5.08 亿,同比持平。高通仅汽车这一条线,已约为这家纯血 ADAS 冠军全部业务的 3.1 倍,而且在对方持平、ASP 下滑的同时增长 61%。ADAS 与座舱芯片的重心已经移向通用算力供应商,而本季是这件事从预测变成事实的节点。再说问号,而且问号不小:8-K 没有给内生与并购的拆分、没有 Snapdragon Ride 与座舱的拆分、没有定点管道金额、也没有车规 AI 算力指标——编辑部会前标记的四个观察点,全都只存在于电话会或投资者日,不在这份文件里。同时,商誉较财年末增加 $29.16 亿、前三季收购现金流出 $15.73 亿,而本附件没有把其中任何部分分摊到具体收入线。从长期「双位数」的节奏一步跳到 +61%,是一次备案未加解释的节奏断裂;在 10-Q 给出答案之前,+61% 应当被读作一个数字,还不是一个速率。还有一笔值得存档的算术:本季把比较基数定在 $15.88 亿,因此 FY27 三季度要维持双位数,需要超过 $17.47 亿。「连续 N 季」这类指标之所以被选中,是因为听起来不可打破;而打破它最可靠的方式,恰恰是一个特别好的季度。④Automotive won the quarter and lost the narrative汽车赢了季度,输了叙事
Read the press-release headline against the segment table. Automotive grew 61%; IoT grew 9%; the headline chose "Combined QCT Automotive and IoT Revenues Grew 28%." A company does not average down its single best number by accident. The reason is in the CEO quote: the organizing frame is now non-handset revenue of $40B by fiscal 2029 — nearly double the November 2024 target — with growth accelerating from 24% in fiscal 2026 to greater than 60% in fiscal 2027, described in the filing as "including Data Center." Do the sanity check an automotive analyst should do: automotive revenue converts on OEM program timelines, three to five years from design win to start of production, which makes a one-year step-change to >60% arithmetically impossible to source from automotive. That acceleration is underwritten by data center. Automotive's role has shifted from hero of the diversification story to ballast inside a bundle — 23 quarters of dependable double-digit growth is exactly what you put underneath a more speculative number. For SBD's clients this is the sentence that matters: nine-month capex doubled to $1,578M and R&D is up 17%, and this filing tells you nothing about where either went. An OEM signing a 2030 platform is now buying from a supplier whose fastest-growing internal claimant on advanced-node capacity, packaging allocation and top AI silicon talent is a business with higher volumes, faster refresh and none of automotive's qualification burden. The right ask on the call is not "how is automotive doing" — it plainly is doing well. It is whether automotive's own fiscal-2029 target — ◆ $10B, given at the June 2026 Investor Day and absent from this filing — gets restated and funded now that data center owns the acceleration, because a target that stops being repeated is a line item losing its claim on capital.
把新闻稿标题和分部表并排读。汽车 +61%、IoT +9%,而标题选择的是「QCT 汽车与 IoT 合计收入增长 28%」。一家公司不会无意间把自己最好的数字平均下去。原因在 CEO 那段话里:如今的组织性叙事是「FY2029 非手机收入达 $400 亿」——接近 2024 年 11 月目标的两倍,增速从 FY26 的 24% 加速到 FY27 的 60% 以上,备案原文写明「含数据中心」。做一遍汽车分析师该做的合理性检验:汽车收入按车企项目节奏转化,从定点到量产三到五年,因此一年内跃升到 60% 以上,在算术上不可能由汽车提供。那段加速是数据中心承保的。汽车的角色已经从「多元化故事的主角」变成「一个组合里的压舱物」——连续 23 季可靠的双位数增长,正是用来垫在一个更具投机性的数字下面的东西。对 SBD 的客户,真正要紧的是这句:前三季资本开支翻倍至 $15.78 亿、研发 +17%,而这份备案完全没告诉你钱去了哪里。今天签 2030 年平台的车企,买的是这样一家供应商:它内部对先进制程产能、封装配额与顶级 AI 芯片人才增长最快的争夺者,是一门量更大、迭代更快、且完全没有车规验证负担的生意。电话会上该问的不是「汽车做得怎么样」——它显然做得很好;该问的是,汽车那个属于自己的 FY2029 目标——◆ 2026 年 6 月投资者日给出的 $100 亿,本份备案里并未出现——在数据中心接手加速叙事之后,还会不会被重述、被配资源。因为一个不再被复述的目标,就是一条正在失去资本索取权的业务线。⑤Modular, and the silence where Arm used to beModular,以及 Arm 消失的那块空白
Only one strategic action shares the headline with the numbers: the completed acquisition of Modular Inc, framed as "creating an open software foundation for generative and agentic AI." The filing gives no price and no segment attribution — but the direction is legible, and it is the automotive AI thesis in one move. The fight is migrating from TOPS to toolchain. Whoever owns the compiler and runtime owns portability, and portability is precisely what an OEM will and won't pay for. Set the three archive positions side by side: Mobileye's moat was a closed EyeQ-plus-software bundle, and our July brief watched that moat turn into channel conflict the moment it became an operator; Nvidia's moat is a proprietary software stack; Qualcomm has just bought a deliberately vendor-neutral one. That is a bet that OEMs will pay a premium for the right NOT to be locked in — a coherent position for the merchant supplier, and one the desk should test against actual OEM software-defined-vehicle procurement rather than press-release adjectives. There is also a revenue-recognition asymmetry to keep in view. Qualcomm books automotive at shipment; our GM brief showed $6.3B of deferred OnStar-line revenue and 30–40% Super Cruise attach after the prepaid trial. Qualcomm's silicon rides inside somebody else's subscription, and captures none of it. Finally, the silence. Arm appears nowhere in this document — no update, no reserve, no contingency, only boilerplate risk factors about third-party IP claims and adverse rulings. ◆ Public record, not filing data: Qualcomm was granted a complete victory in the Arm litigation in late 2025 over the Nuvia-derived custom cores, with Arm signalling appeal and Qualcomm pressing its own claims. The absence is itself the disclosure: management no longer considers architecture-license risk material enough to name. For an OEM underwriting a seven-year program on custom-core silicon, "no longer worth mentioning" is a meaningful answer — but it is management's judgement of materiality, not a court's finality, and the 10-Q legal-proceedings note is where to verify it.
与数字同列标题的战略动作只有一个:完成对 Modular Inc 的收购,定位为「为生成式与 agentic AI 建立一个开放软件底座」。备案没给价格、也没做分部归属——但方向是可读的,而且这一步就是汽车 AI 的核心命题:战场正从 TOPS 迁向工具链。谁掌握编译器与运行时,谁就掌握可移植性;而可移植性恰恰是车企愿意、或不愿意付钱的东西。把档案里的三种站位并排看:Mobileye 的护城河是封闭的 EyeQ+软件捆绑,我们 7 月的简报正看着这条护城河在它下场做运营的那一刻变成渠道冲突;英伟达的护城河是自有软件栈;高通刚买下的是一个刻意保持厂商中立的软件栈。这是在赌车企愿意为「不被锁定的权利」付溢价——对一家商用芯片供应商是自洽的选择,但编辑部应当拿真实的车企 SDV 采购行为去检验它,而不是拿新闻稿形容词。另有一处收入确认的不对称需要留在视野里:高通的汽车收入按出货确认;而我们的通用简报显示 OnStar 口径递延收入 $63 亿、Super Cruise 预付期满后 30–40% 继续付费。高通的芯片坐在别人的订阅里,却分不到那份订阅。最后说那块空白。Arm 在这份文件里完全没有出现——没有进展、没有计提、没有或有事项,只有关于第三方知识产权索赔和不利裁决的通用风险表述。◆ 以下为公开记录而非备案数据:围绕源自 Nuvia 的自研核,高通在 2025 年底的 Arm 诉讼中获得完全胜诉,Arm 表示将上诉,高通亦在推进自己的诉求。这块空白本身就是披露:管理层已不认为架构授权风险重要到需要点名。对一家要用自研核芯片承保七年项目周期的车企,「已经不值得一提」是一个有意义的答案——但那是管理层对重要性的判断,不是法院的终局,要核实请看 10-Q 的法律诉讼附注。THE BOTTOM LINE一句话结论
Automotive delivered the best quarter it has ever had at Qualcomm — $1,588M, +61%, three times Mobileye's entire business — inside a company whose operating income fell 41%, whose GAAP EPS was lifted $0.57 by an investment gain it guides to zero next quarter, and whose new growth story explicitly belongs to the data center; the question for the industry is no longer whether Qualcomm can win automotive compute, but whether automotive can still command capital inside a supplier that has found a faster-growing customer than the car.汽车交出了它在高通历史上最好的一个季度——$15.88 亿、+61%、等于 Mobileye 整个公司的三倍——但它所在的这家公司,本季经营利润下滑 41%、GAAP EPS 被一笔下季被自己指引为零的投资收益抬高了 $0.57、而新的增长故事已明确写给数据中心;行业要问的问题不再是「高通能否赢下汽车算力」,而是:在一家已经找到比汽车增长更快的客户的供应商内部,汽车还能不能争到资本。
WATCHLIST (POST-CALL)后续观察清单
□ Organic vs acquired split of the +61% automotive line, and which deals landed in which revenue stream (10-Q) · □ Whether the ◆ $10B FY29 automotive target from the June Investor Day is restated, or stays folded inside "non-handset $40B" · □ Design-win pipeline figure — restated on the call, or quietly retired · □ Does automotive stay double-digit against the new $1,588M base (needs >$1,747M in Q3 FY27) · □ Input-cost pass-through: first quarter with a stated basis-point recovery, not "over time" · □ Whether Q4 QSI actually comes in at $— as guided · □ Origin of the $4,136M nine-month tax benefit versus $5,550M of excess cash tax payments (10-Q) · □ Inventory at $8,379M: cost-inflation pre-buy or demand slip · □ Where the doubled capex goes — automotive qualification capacity or data center · □ Modular integration: any automotive toolchain commitment, or purely data-center scoped · □ Arm: whether the appeal ever re-enters the legal-proceedings note□ 汽车 +61% 的内生/并购拆分,以及各笔收购分别落在哪条收入线(看 10-Q) · □ 投资者日给出的 ◆ 汽车 FY29 $100 亿目标是否被重述,还是继续打包在「非手机 $400 亿」里 · □ 定点管道金额——电话会上重述,还是被安静下架 · □ 在 $15.88 亿的新基数上汽车能否守住双位数(FY27 三季度需 >$17.47 亿) · □ 成本转嫁:哪一季开始给出明确的基点回收幅度,而不是「随时间」 · □ Q4 的 QSI 是否真如指引落在 $— · □ 前三季 $41.36 亿税务收益的来由,对照 $55.5 亿超额实缴税款(看 10-Q) · □ $83.79 亿存货:提前备货应对涨价,还是需求走弱 · □ 翻倍的资本开支流向——车规验证产能还是数据中心 · □ Modular 整合:有无面向汽车工具链的承诺,还是纯数据中心范围 · □ Arm:上诉是否重新出现在法律诉讼附注中