Around 2015, automakers and consumers arrived at an unspoken bargain: buy the car and you can connect your phone. Apple CarPlay and Android Auto projected familiar maps, music and messaging onto the dashboard. Automakers outsourced an experience they struggled to keep current; drivers brought a digital life they already understood.
In October 2025, GM chief executive Mary Barra told Decoder that the company would eventually phase out CarPlay and Android Auto across its lineup. GM later clarified that existing vehicles would not change, gasoline vehicles would retain projection for the foreseeable future, and the transition would be gradual as a new centralized-compute platform arrives from 2028. [The Verge] [GM clarification]
That remains an unusually large bet. McKinsey found in 2023 that 45% of connected-car users used phone projection regularly and another 40% used it periodically; nearly half of surveyed buyers said they would not purchase a vehicle without CarPlay or Android Auto. [McKinsey]
The replacement makes the story more interesting. GM is removing Android Auto, a Google product, while relying more heavily on Android Automotive OS, Google built-in and Gemini. This is not Google deliberately killing Android Auto—Google is still developing it, and began rolling Gemini out to Android Auto in late 2025. [Google] What is happening is subtler: an automaker is giving up Google's projection layer while inviting Google into a deeper embedded layer.
That shift captures the evolution of Google's relationship with the auto industry: from a suitcase carried into the car to part of the building's infrastructure. “Landlord” is a metaphor for control over digital infrastructure, not legal ownership of the vehicle, and it does not mean Google automatically receives every stream of vehicle data.
Act I: the suitcase era and the bill behind the win-win
Projection remains one of the most successful features in automotive history. Apple said in 2022 that 98% of new cars sold in the United States supported CarPlay and that 79% of US buyers would only consider a car with it. Those were Apple's own figures, not an independent global measurement, but they captured how quickly projection became expected equipment.
The bargain meant different things to each party. Drivers received an interface that improved with their phones. Automakers avoided an endless cockpit-software refresh cycle. The technology companies gained something more strategic: time inside the car and influence over how drivers defined a good cockpit. Whenever a driver chose projection, the automaker's own navigation, media and app layer disappeared behind the phone.
The economics sharpened that tension. MarketsandMarkets estimates the connected-car market could reach $26.4 billion by 2030. [MarketsandMarkets] At a March 2026 investor presentation, GM described roughly $2.7 billion in 2025 digital-services revenue and $5.4 billion in related deferred revenue—the second number is a balance-sheet liability to be recognized over time, not annual revenue. [GM Investor Relations] Stellantis has targeted about €20 billion in incremental annual software-related revenue by 2030. [Stellantis]
Revenue and data control are therefore deep motives behind the move away from projection. Safety and integration are not merely excuses: an embedded system can use battery state, range, sensors and vehicle controls in ways a phone connection often cannot. But those engineering benefits coexist with a commercial ambition—to keep the customer relationship and recurring revenue inside the automaker's own product.
Act II: the fitted-interior era and its paradox
Automakers then tried to reclaim the cockpit. The problem was what to build it with. Maintaining a global operating system, map, voice layer, app ecosystem and security-update program is expensive, and many in-house projects have struggled. Android became the most visible path out.
A 2022 S&P Global Mobility forecast illustrated the slope: Android Automotive OS, or AAOS, was expected to grow from roughly 1% of infotainment systems in 2022 to about 18% in 2027. [S&P Global Mobility] That does not mean AAOS simply “ate” QNX. AAOS commonly runs the infotainment experience, while QNX may handle real-time, safety-critical or hypervisor layers; both can exist in the same vehicle. QNX says its software is embedded in more than 275 million vehicles. [QNX]
AAOS is also not the same thing as Google built-in. AAOS is the open-source operating-system base. Google Automotive Services, or GAS, adds Google Maps, Google Play and Google's assistant layer under a commercial agreement; that package is marketed to consumers as Google built-in. An automaker can use AAOS without buying GAS.
The scale of the embedded layer became unusually visible in April 2026. GM said Gemini would roll out to about four million eligible model-year-2022-and-newer Cadillac, Chevrolet, Buick and GMC vehicles with Google built-in in the United States. [GM]
Now the paradox can be stated precisely. GM wants more control over data, experience and revenue, yet it is placing the operating-system base and some combination of maps, voice and apps on Google's stack. It is not transferring legal title to the cockpit, and the data Google can access depends on contracts, accounts, consent and configuration. But the dependency is deeper than it was in the projection era. To reduce its dependence on a phone platform, the automaker may become more dependent on embedded infrastructure.
Four doors into the same courtyard
Google's automotive portfolio is effective because it is not a single all-or-nothing door. It offers several levels of integration.
Door one: Google built-in. This is the fastest way to fill a cockpit's map, app-store and assistant gaps. BYD's ATTO 2 DM-i, unveiled in October 2025 and launched in Europe in 2026, became one of BYD's first models with Google built-in. It still supports CarPlay and Android Auto, useful evidence that embedded Google and phone projection can coexist. [BYD] BYD and Parkopedia are also preparing an AAOS ParkPay service for early 2027, initially on the Denza Z9 GT and BYD ATTO 3 EVO, with some older vehicles to receive it by OTA. The public announcement does not support the claim that every Denza model is included or that it covers 90 million spaces. [Parkopedia]
Door two: Google Maps Auto SDK. This suits automakers that want to preserve their own interface while using Google's map data and routing engine underneath. Rivian was an early deployment in July 2025. [Rivian] On July 16, 2026, XPENG became the first Asia-Pacific automaker to ship the SDK. The L03 opened orders simultaneously in 65 countries and regions—not 64—and XPENG and Google said the integration was completed in less than a year. [XPENG]
Door three: a branded AI agent. Google Cloud's Automotive AI Agent lets an automaker keep its own wake word and identity while using Gemini and Google Maps Platform. Mercedes-Benz introduced it in the new CLA's MBUX under “Hey Mercedes.” The concrete advantage highlighted in the announcement was access to more than 250 million places and over 100 million map updates a day. [Google Cloud]
Door four: projection. If an automaker signs none of the deeper agreements, an Android user can still bring Google into the car through Android Auto. In that case Google occupies the user-experience layer, but it does not automatically gain an OEM contract or access to the automaker's full vehicle-data stream.
These doors lead toward the same courtyard, but not to the same room. Each gives Google a position in part of the cockpit; the rights, data flows and switching costs are different in every case.
The conversion at the border: the China variable
There is a large market where Google's proprietary consumer stack cannot operate in the same way. In S&P Global Mobility's 2022 baseline, roughly 32% of infotainment systems were classified as “generic Android,” and the forecast still put the category above a quarter of the market in 2026. That was a dated baseline and forecast, not a current measured share. It describes a world in which many automakers—especially in China—adapt the Android Open Source Project without Google's proprietary services.
Google's consumer services are not normally accessible in mainland China, and Google Cloud lists no mainland-China cloud region. [Google Cloud locations] But it would be wrong to say every car in China is unrelated to Google: AOSP still sits in the lineage of many systems, and suppliers operate across borders. The accurate point is that Google's proprietary services and commercial cloud stack do not participate in mainland cockpits on the same terms as they do in Europe.
The conversion happens on export programs. Rivian arrived in July 2025, BYD's first Google built-in European programs followed, and XPENG shipped Maps Auto SDK in July 2026. Some Chinese automakers are keeping a self-developed Android stack at home while adding Google's services to selected overseas vehicles. That is not yet a universal rule, but each export program that adopts Google converts an AOSP-based cockpit into commercial Google automotive business.
The attraction is real. Overseas buyers already expect navigation at phone quality; domestic suppliers often lack equivalent global POI depth, local services and language coverage; and the Google name can reduce trust friction for a new Chinese brand. The trade is classic: exchange some control and bargaining leverage for speed, coverage and market acceptance.
The trust problem: Google's two clocks
Automakers' deepest psychological objection is Google's record of shutting products down. The mobile driving experience is a useful case file. Assistant Driving Mode was announced at Google I/O in 2019, arrived in preview in late 2020, expanded in 2021, then began losing its Dashboard in late 2022. More features disappeared in 2024, and the remaining Maps launcher was removed in April 2025. [9to5Google] The exact definition of “launch” changes the lifetime calculation, but the direction is unambiguous: dismantling began roughly a year after the fuller experience arrived.
The embedded record is different. In 2026, existing Google built-in vehicles received Gemini by OTA: GM's eligible fleet begins with model year 2022, while Polestar said all of its models except Polestar 1 would receive the assistant without new hardware. [Polestar] That does not prove every automotive service will live forever. It does show that a multi-year vehicle upgrade promise can operate on a different clock from a consumer app.
The procurement question is therefore not simply “Will Google kill it?” Google has killed products and supported long-lived vehicle programs. The better question is: which clock is written into this contract? What is voluntary product goodwill, and what is a binding obligation covering support, security, data portability, fallback operation and migration? That distinction is worth an entire legal team's attention.
Cold water: installing Google does not win the cockpit
The first evidence is not romantic. Rivian's early Google Maps rollout produced owner reports of frozen navigation and repeated soft reboots. [Carscoops] A strong map foundation does not eliminate launch risk.
The second warning is in the product list itself. BYD kept CarPlay and Android Auto beside Google built-in. That is not a contradiction; it is an admission that users value choice and that an embedded stack has to earn daily use.
Most important, Google's EV-routing promise depends on vehicle data: state of charge, consumption, charger compatibility and destination conditions. [Google Maps] If an automaker shares too little data—or integrates it poorly—the result can still feel generic. Google supplies the entry ticket, not immunity. Differentiation moves to the depth at which the vehicle and the Google stack are fused, precisely the part that is hardest to outsource.
Three watchpoints and one old rule
GM's 2028 wager. The new centralized-compute generation beginning with the Escalade IQ will be the cleanest large-scale test of whether consumers accept an OEM cockpit without projection. The Escalade IQ already lacks CarPlay and Android Auto today; 2028 refers to the new computing architecture, not the first GM vehicle to remove projection.
The late-2026 comparison. BYD's Google built-in route, XPENG's Auto SDK route and Rivian's first full year with the SDK will begin producing comparable owner evidence. The data will not be perfectly symmetrical, but it will give the industry its first useful side-by-side view of Google's two main integration paths.
Regulation squeezing from both directions. The European Union has designated Alphabet a Digital Markets Act gatekeeper and Google Maps a core platform service. That does not make GAS bundling unlawful, but it makes future scrutiny of tying, interoperability and data terms a reasonable inference. [European Commission] In the United States, the Commerce Department's connected-vehicle rule is more concrete: from model year 2027, manufacturers owned by, controlled by or subject to the jurisdiction of China or Russia are generally prohibited from selling connected vehicles in the US, even when the vehicle-connectivity hardware and software are not themselves Chinese, unless authorized. A Google stack does not neutralize the manufacturer's national nexus. [US final rule]
The endgame follows an old automotive rhythm: embrace a technology company, fear loss of control, build in-house, struggle with the cost, then embrace again. The industry has already swung through that cycle more than once. AI raises the cost of going alone and makes Google's bundle more complete, so this swing may last longer. It has not disappeared.
For Chinese automakers and suppliers, the whole argument reduces to one question: when an export vehicle crosses the border and its cockpit switches to Google, what do you gain—and what control, leverage and optionality do you give away? The right time to calculate that trade is before signing, not at renewal.
This article is based on public information. Facts were rechecked and updated on July 18, 2026; the argument and conclusions are those of TopChinaCar.
2015年前后，车企和消费者达成过一个心照不宣的握手协议：你买我的车，我让你连手机。CarPlay和Android Auto把地图、音乐和消息投到车机屏幕上。车企外包了自己难以持续更新的体验，用户带来了熟悉的数字生活，各取所需。
2025年10月，通用汽车CEO玛丽·博拉在Decoder播客里表示，GM最终会在全线车型中逐步取消CarPlay和Android Auto。随后GM澄清：已经支持投屏的车辆不会改变，燃油车在“可预见的未来”仍会保留投屏；真正的全面迁移将随新的集中式计算平台渐进发生，该平台计划从2028年起在Escalade IQ上率先部署。[The Verge] [GM后续澄清]
这仍然是一场很大的赌局。麦肯锡2023年的调查显示，45%的联网汽车用户经常使用手机投屏，另有40%不定期使用；接近一半的受访购车者表示，不会购买没有CarPlay或Android Auto的汽车。[麦肯锡]
更耐人寻味的是替代方案：GM砍掉的Android Auto是Google的产品，而顶上来的技术栈——Android Automotive OS、Google built-in和Gemini——仍然来自Google。这并不是Google主动献祭Android Auto；Android Auto仍在继续开发，2025年末还开始接入Gemini。[Google] 真正发生的是：车企放弃了Google的投屏层，却把Google请进了更深的嵌入式层。
理解这步棋，就理解了Google和车企关系的演变：从被带进车里的行李箱，变成房子的数字基础设施。本文所说的“房东”是控制力的比喻，不是法律产权，也不意味着Google自动拿走全部车辆数据。
第一幕：行李箱时代，双赢背后的账
投屏至今仍是汽车史上最成功的数字功能之一。苹果2022年曾披露，美国98%的新车支持CarPlay，79%的美国购车者只会考虑支持CarPlay的车型。这是苹果自报的美国市场数据，不是独立的全球统计，但足以说明投屏多快变成了消费者的默认期待。
这笔交易对三方的意义并不相同。用户得到了一套随手机一起更新的界面；车企绕开了座舱软件持续迭代的无底洞；科技公司拿到的则更具战略性：用户在车内的时间，以及定义“好座舱”的影响力。用户一旦进入投屏，车企自己的导航、媒体和应用层就退到了手机背后。
经济账让这种张力更尖锐。MarketsandMarkets预计，广义联网汽车市场到2030年可达264亿美元——这里是“联网汽车市场”，不是单独的“车联数据市场”。[MarketsandMarkets] GM在2026年3月的投资者活动中披露，2025年数字服务确认收入约27亿美元，期末相关递延收入约54亿美元；后者是将在未来随履约逐步确认的负债，不能直接写成年度收入。[GM投资者关系] Stellantis则曾提出，到2030年实现约200亿欧元软件相关增量年收入的目标。[Stellantis]
所以，收入与数据控制确实是“去投屏化”的深层动力。安全和集成也不只是话术：嵌入式系统可以调用电池电量、续航、传感器和车控，手机连接往往做不到同样深。但工程收益与商业野心同时存在——车企想把客户关系和持续收入留在自己的产品里。
第二幕：精装修时代，悖论上演
于是车企开始收回座舱。问题是拿什么收。全球操作系统、地图、语音、应用生态和安全更新都要长期维护，不少自研项目都付出了高昂学费。Android成了最显眼的一条出路。
S&P Global Mobility在2022年给出过一条很陡的预测曲线：AAOS在车载信息娱乐系统中的份额，将从2022年约1%上升到2027年约18%。[S&P Global Mobility] 但这不能写成“AAOS吃掉QNX”。AAOS通常承载信息娱乐体验，QNX则可能承担实时、安全关键系统或虚拟化层，两者完全可能出现在同一辆车里。QNX目前称其软件已嵌入超过2.75亿辆汽车。[QNX]
AAOS也不等于Google built-in。AAOS是开源操作系统底座；Google Automotive Services（GAS）是在商业协议下加入Google Maps、Google Play和Google助手层，面向消费者通常称作Google built-in。车企可以使用AAOS而不签GAS。
嵌入式层的规模在2026年4月第一次变得非常直观。GM宣布，Gemini将通过OTA陆续进入美国约400万辆符合条件的2022款及以后凯迪拉克、雪佛兰、别克和GMC车型，前提是车辆已经配备Google built-in。[GM]
现在可以把悖论摆上桌：GM希望收回数据、体验和收入主导权，却把操作系统底座，以及部分地图、语音和应用能力放在Google栈上。这不等于把座舱的法律产权过户给Google；Google能接触什么数据，取决于合同、账号、用户同意和车企配置。但依赖关系确实比投屏时代更深了。为了减少对手机平台的依赖，车企可能增加对嵌入式基础设施的依赖。
四扇门，通向同一个院子
Google汽车产品组合的精妙之处，在于它不是一扇非黑即白的门，而是几种不同深度的接入方式。
第一扇：Google built-in。适合想快速补齐地图、应用商店和助手短板的车企。2025年10月亮相、2026年在欧洲上市的比亚迪ATTO 2 DM-i，成为比亚迪首批内置Google built-in的车型之一。它同时保留CarPlay和Android Auto，这也证明内置Google与手机投屏可以并存。[比亚迪] 比亚迪与Parkopedia还计划在2027年初上线基于AAOS的ParkPay，首批车型是腾势Z9 GT与比亚迪ATTO 3 EVO，部分旧车随后通过OTA获得。公开通稿没有支持“腾势全系”或“覆盖9000万车位”的说法。[Parkopedia]
第二扇：Google Maps Auto SDK。适合想保住自有HMI、但在底层使用Google地图数据与路径引擎的车企。Rivian在2025年7月率先落地。[Rivian] 2026年7月16日，小鹏成为亚太首家量产交付该SDK的车企；L03在65个国家和地区同步开订，不是64个。小鹏与Google称整个集成不到一年完成。[小鹏]
第三扇：品牌化AI agent。Google Cloud的Automotive AI Agent允许车企保留自己的唤醒词和品牌身份，底层使用Gemini与Google Maps Platform。梅赛德斯在新CLA的MBUX中以“Hey Mercedes”首发。通稿真正强调的硬能力，是可调用超过2.5亿个地点的数据，以及每天超过一亿次地图更新。[Google Cloud]
第四扇：投屏。即便车企没有签更深的协议，Android用户仍可以通过Android Auto把Google带进车里。这种情况下，Google占据的是用户体验层，并不自动获得车企合同或完整车辆数据。
四扇门通向同一个院子，却不是同一个房间。每一种方式都让Google占住座舱的某一层，但权利、数据流和切换成本并不相同。
国境线上的转化：中国变量
这张地图上有一个Google专有消费服务无法以同样方式进入的大市场。S&P Global Mobility的2022年基准数据中，约32%的车载信息娱乐系统属于“Generic Android”；其当时预测，到2026年这一类别仍会超过四分之一。这里是历史基准和预测，不是2026年的实时实测份额。它描述的是大量车企——尤其中国车企——拿Android开源项目自行改造、但不使用Google专有服务的现实。
Google消费服务在中国大陆通常无法正常访问，Google Cloud公布的区域列表里也没有中国大陆区域。[Google Cloud区域] 但说“境内每一台车都与Google无关”也太绝对：许多系统的技术谱系里仍有AOSP，供应商也跨境经营。更准确的说法是，Google的专有消费服务与商业云栈，不能像在欧洲那样参与中国大陆座舱。
转化发生在出海项目上。Rivian在2025年7月落地，随后是比亚迪首批欧洲Google built-in项目，再到2026年7月的小鹏Maps Auto SDK。部分中国车企正在形成一条双轨：国内车型继续自研Android栈，部分海外车型接入Google。这还不是所有车企的统一路线；但每一个采用Google的出海项目，都把一套AOSP座舱转化成了Google的商业汽车业务。
这笔交易对中国车企确实有诱惑。海外消费者对“手机水准的导航”已有既定期待；国内供应商往往缺少同等深度的全球POI、当地服务与多语言覆盖；Google的品牌还能为新进入欧洲的中国车企降低一部分信任摩擦。交易的本质很典型：用一部分控制权和议价能力，换速度、覆盖与市场接受度。
信任问题：Google的两套时钟
车企最深的心理障碍，是Google砍产品的前科。手机侧驾驶产品是一份有用的档案：Assistant Driving Mode在2019年I/O大会宣布，2020年末以预览形式落地，2021年扩大推出，2022年末开始拆掉Dashboard；更多功能在2024年消失，最后的Maps启动器于2025年4月移除。[9to5Google] 对“正式上线”的定义不同，会算出不同寿命；但方向没有争议——较完整体验推出约一年后，核心功能就开始拆解。
车内嵌入侧的记录却不一样。2026年，现有Google built-in车型开始通过OTA获得Gemini：GM的适用车从2022车型年算起；Polestar则宣布，除Polestar 1外的全部车型都能在不增加硬件的情况下获得Gemini。[Polestar] 这不能证明每一项汽车服务都会永久存在，却说明多年期车辆升级承诺确实可能运行在不同于消费应用的时钟上。
所以车企该问的不是一句“Google会不会砍”。它砍过产品，也兑现过长期车辆项目。更好的问题是：我签的合同挂在哪个时钟上？哪些只是Google自愿提供的产品善意，哪些是覆盖支持、安全、数据可迁移性、离线降级与替代方案的白纸黑字义务？这一问，值一个法务团队的年薪。
冷水：装了Google，不等于赢了座舱
最初的实证并不浪漫。Rivian接入Google Maps能力后的早期阶段，出现了车主反馈导航冻结、需要反复软重启的问题。[Carscoops] 更强的地图底座，不会自动消灭量产落地风险。
第二盆冷水就写在产品配置表里：比亚迪在Google built-in旁边继续保留CarPlay和Android Auto。这不是逻辑冲突，而是承认用户重视选择，也承认嵌入式系统必须靠每天的体验赢得使用。
最重要的是，Google的电动车路径规划承诺依赖车辆数据：电池电量、能耗、充电兼容性和目的地条件。[Google Maps] 如果车企开放的数据太少，或者集成做得不深，结果仍可能很通用。Google给的是入场券，不是免死牌。差异化会转移到“车辆与Google栈融合得多深”——这恰恰是最难外包的部分。
三个观察点，和一个老规律
2028年的GM赌局。从Escalade IQ开始的新一代集中式计算平台，将是“没有投屏，消费者买不买账”的一次大规模检验。需要说清的是：Escalade IQ现在就已经没有CarPlay和Android Auto；2028指向的是新计算架构，不是GM第一辆取消投屏的车。
2026年下半年的口碑对撞。比亚迪的Google built-in路线、小鹏的Auto SDK路线，以及Rivian采用SDK满一年的反馈，会开始产生可比较的真实车主证据。三组数据不会完全对称，但足以让行业第一次较像样地并排观察Google两条主要集成路线。
监管的双向挤压。欧盟已把Alphabet认定为《数字市场法》意义上的gatekeeper，并把Google Maps列为核心平台服务。这并不等于GAS捆绑已经违法，但未来围绕捆绑、互操作性与数据条款的审视，是合理推论。[欧盟委员会] 美国的联网汽车规则则更具体：从2027车型年起，受中国或俄罗斯所有、控制或司法管辖的联网汽车制造商原则上不得在美国销售联网汽车，即使车联网软硬件本身并非来自中国，除非获得授权。装上Google，不能消除制造商的国别关联。[美国最终规则]
终局仍然服从一个老规律：车企拥抱科技巨头，害怕失控，转向自研，承受自研成本，然后再次拥抱。行业已经摆过不止一轮。AI抬高了单干门槛，也让Google的包裹更完整，所以这次钟摆可能停得更久，但没有消失。
对中国车企和供应商，整篇文章可以浓缩成一个问题：当出海车型越过国境、座舱换上Google时，你得到什么，又让出了多少控制权、议价能力和未来选择？这笔账最好在签字之前算，而不是续约的时候。
本文基于公开信息写作；事实于2026年7月18日重新核查，观点与结论由TopChinaCar独立提出。