The Map Wars · Part 8
On March 12, 2026, Google gave Maps its biggest overhaul in more than a decade: the search box became a Gemini conversation — "where can I get good coffee with a quiet corner to take a call" now returns a reasoned answer — and the flat map became a real-time 3D rendering built from Street View and aerial imagery. One fact went unmentioned at the launch: the company that has ruled the map industry for twenty years has never made money selling maps.
Everyone else sells maps as a product. Google raises the map as an organ of Search. That single difference in position explains everything about it — why free, why build-your-own, why it dares to raise prices, and why its every move rewrites the valuation function of the entire industry. This piece takes the engine apart, and then answers a more practical question: outside the giant's gravity well, how much space is left?
A Beginning That Was Bought
Google Maps was not invented at Google; it was assembled from three 2004 acquisitions: Where 2 Technologies — two brothers in Sydney — supplied the product prototype; Keyhole supplied satellite imagery (later Google Earth); ZipDash supplied phone-based live traffic. Launch: February 8, 2005.
Four months later the API opened — and the decision was forced. Hackers had "stolen" map data to build mashups like housingmaps; Google chose surrender-by-embrace, converting the illegal use into an official product. One of its deepest moats — the developer ecosystem — began with somebody else trespassing on it.
Graduation in Thirteen Months
When Street View launched in 2007 it was treated as a toy. It was a survey fleet. In September 2008, Google Maps switched exclusively to Tele Atlas data worldwide; at the same moment, the internal Ground Truth project ran at full speed — Street View imagery, public government data, crowdsourced fixes, thousands of human operators checking the output. In October 2009, the US market dropped Tele Atlas for Google's own data.
From signing a long-term exclusive to graduation: thirteen months. It never intended to pay indefinitely — it paid to buy time to build. This series has already stated the lesson: selling your data to a platform customer capable of building its own is tutoring your replacement.
The Economics of an Organ
October 2009 held another blade: navigation went free on Android. The PND category was sentenced on the spot, and the whole ecosystem that earned money from navigation began liquidation.
Free follows organ logic: the map feeds Search, Search feeds ads, ads sustain the map, and the organ itself needs no separate P&L. Until the day it did — 2018, when Google Maps Platform was repriced: API costs up as much as fourteen-fold, free tiers slashed, credit cards made mandatory. Developers left in waves, toward Mapbox, OSM, HERE and TomTom. Those refugees became the popular base of Overture four years later. To make the organ answer to a P&L, Google manufactured its own opposition coalition by hand.
And the organ's output is now vast: Google Maps generated roughly $11.1 billion in 2023, up from about $3 billion in 2019, with 82% from advertising — promoted pins and location-based placements. Eleven billion dollars is more than five times the combined revenue of TomTom and HERE. And it is one organ of Google.
The Flywheel
Today's estate: more than 300 million places; input from 500 million contributors feeding routes and recommendations in real time; ETA predictions around 97% accurate since DeepMind's integration; hundreds of billions of Street View images; and the implicit behavioral feedback of billions of users, daily.
No link in that flywheel carries an acquisition cost — users use the product for free, the product collects the data for free. Navteq's cost structure — hundreds of employees paid to drive every street — against "billions of people walking every street for you, free": that is not competition. That is a difference of species.
The AI Triple
Across 2025 and 2026, Google played three moves on the map, all pointing one direction. November 2025: Gemini replaced Google Assistant inside Maps. March 2026: Ask Maps and Immersive Navigation — the map turned from a query tool into a conversation partner. The most important move was the middle one: on October 17, 2025, Grounding with Google Maps landed in the Gemini API — developers can wire a model's reasoning directly to live data on 250 million places; the model auto-detects geographic intent, returns grounded answers with citations, ships a context token for an embeddable interactive map, and can be combined with Search grounding in the same request.
See it for what it is: the closed-source answer to the "AI spatial grounding layer" narrative. The open camp talks about a neutral registry of places; Google packaged the registry, the reasoning and the distribution into one API and put a price on it. Fast, deep, fresh — inside its own function, unanswerable.
Cracks in the Throne
Five cracks; four are about maps, the fifth is about the neck.
China is zero — the world's largest single market, absent. Korea's export restrictions on high-precision data block depth. Europe presses on two lines: the DMA constrains self-preferencing of Maps inside Search, and the German cartel office's two decisions dismantled the automotive-services bundle and the map-mixing restrictions — "using Google" no longer requires "using all of Google." The fourth crack is a precedent of its own making: 2018 proved that when the map is told to answer to finance, Google will spend ecosystem trust — which means post-unbundling à-la-carte pricing may well repeat "compliant but expensive," and ship a second cohort of customers to its rivals.
The fifth cuts deepest: AI self-cannibalization. Ask Maps answers directly, so users stop browsing lists and stop tapping promoted pins — and 82% of the revenue is built on users browsing lists. Google's own AI is dissolving Google's own revenue interaction. Organs obey the whole: when Search itself is rebuilt around AI, the map's KPIs get rewritten too. An organ's fate is to have no will of its own.
Beyond the Gravity Well
Now the opening question.
Inside Google's function — map value equals traffic times advertising — nobody beats Google, and that war is not worth fighting. But the market permanently contains demand for a second function: map value equals liability times compliance times composability. The territory that function covers is exactly where the advertising function cannot reach, and every block of it is structural, not leftovers.
Block one: sovereignty. Carmakers will not hand their customer relationships and behavioral data to an advertising company — the in-house cockpit programs of Mercedes, BMW and Volkswagen, and the non-Google global stacks Chinese exporters need, are that demand's visible shape. The advertising function conflicts with data sovereignty by nature; no price Google sets can dissolve the conflict.
Block two: liability. The EU's mandatory intelligent speed assistance, Level-3 activation authority, automotive-grade SLAs — these businesses are, at bottom, "if it's wrong, I pay." They contract for liability, not traffic. The advertising function doesn't sign liability: free things carry no indemnity clause. Google can make the map 97% right, but its business model doesn't answer for the other 3%.
Block three: institutionalized composability. The German case granted the right to swap; Overture supplied the raw material; NDS supplied the socket. With all three, "beyond Google" stops being a posture and becomes a constructible architecture — with ten-year commitments and an independent trustee underneath. This is space drawn by regulators' own hands. Google cannot enter it, because the space is defined as not-Google.
Block four: the customers Google ships out itself. The 2018 repricing refugees built Overture; if post-unbundling pricing replays "compliant but expensive," a second wave ships. Every time the giant answers to finance is customer-acquisition season for the neutrals.
Block five: the Agent era. However good Grounding with Google Maps is, Apple will not use it, Amazon will not use it — no platform that treats agents as strategy will hand its places layer to its largest rival. The agent economy needs a neutral spatial grounding layer, and by definition that seat cannot carry Google's name.
Added up, these five blocks are not a "beat Google" market. They are a "beyond Google" market.
Why the Name Is TomTom, Not HERE
One more question is owed: two non-Google, automotive-grade mapmakers stand in this space — why does it belong to one of them? On the surface HERE's credentials run older: a larger installed base, tens of millions of vehicles running its driver-assistance data across Europe and North America. The answer is that every factor of the second function eliminates it, one by one.
Composability first. This space was institutionalized only recently — drawn by the German case, Overture and NDS — and across those three rewritings of the rules, TomTom was complainant, co-founder and participant; HERE was absent three times, appearing only as a named beneficiary in other people's commitments. More fundamental is architecture: a composable market wants "open base plus pluggable layers," and Orbis was built to that shape; HERE's fully proprietary vertical stack is precisely the non-composable kind. The space rewards those who build ships to its shape, not spectators.
Liability second. The essence of the liability function is "if it's wrong, I pay — through 2035." Signing that requires an entity that will live to the payout date. A company whose shareholders book impairments annually and revisit "why do we still hold this" every quarter cannot credibly commit to ten-year SLAs; carmaker procurement risk reviews dock it at the "supplier continuity" line. The liability business first compares who can sign, and only then whose data is better.
Neutrality third — the fatal one. HERE is owned by Audi, BMW and Mercedes; its "neutrality" holds only toward its own shareholders. For their competitors — above all the Chinese carmakers now fighting BBA hand-to-hand in global markets — feeding fleet data back into a supplier controlled by rival shareholders is a competitive taboo at the procurement table. TomTom carries no OEM shareholder and stands equidistant from every carmaker. In the second function, neutrality must be unconditional; conditional neutrality isn't neutrality.
Cost last. In an era of base maps trending to zero, a fully proprietary stack keeps carrying the entire base map's maintenance bill alone, and that bill slowly crushes the second function's margins. TomTom has already paid the conversion tuition — Genesis rebuilt the production line, the base swapped to the commons, and the savings sit in the dynamic and responsibility layers.
So the accurate statement is not that TomTom beat HERE. It is that the second function's four factors — composability, liability, neutrality, cost — eliminated HERE item by item. The space chose the company that built its ship to the space's shape: the Amsterdam company that swapped its base map for open raw material and rebuilt itself as an automotive-grade factory, holding a €2.4 billion order book, the rule-writer's standing of the German case's complainant, and a first batch of contracts countersigned by the Volkswagen system itself. The space is structural; keeping it is a matter of execution — but the space itself, Google cannot take, and HERE cannot hold.
Forty years of map wars come down, in the end, to a war between two valuation functions. The king's function cannot be taken from him. But the king's function also has places its light doesn't reach. That territory is not large — but it is large enough to raise an oligopolist.
Sources: Google developer blog (Grounding with Google Maps launch, October 17, 2025); Google's March 2026 Ask Maps / Immersive Navigation launch materials; ElectroIQ Google Maps statistics (2023 revenue and mix); Bundeskartellamt decision B7-25/22; Search Engine Land coverage of the 2009 data switch and 2018 GMP repricing; earlier installments of this series.
《地图战争》系列 · 第8篇
2026年3月12日，谷歌给地图做了十多年来最大的一次改造：搜索框换成了Gemini对话——"哪有能安静打电话的咖啡馆"这种问题直接给出推理过的答案；平面地图换成了基于街景和航拍实时渲染的3D。发布会上没人提的一个事实是：这家统治地图行业二十年的公司，从来没有靠卖地图挣过钱。
其他所有人把地图当产品卖，谷歌把地图当搜索的器官养。这个定位差异解释了它的一切——为什么免费、为什么自建、为什么敢涨价、为什么它的每个动作都在重写整个行业的估值函数。这篇把这台引擎拆开，顺便回答一个更实际的问题：巨头的引力场之外，还剩多少空间。
买来的起点
谷歌地图不是谷歌发明的，是2004年三次收购拼出来的：悉尼两兄弟的Where 2 Technologies给了产品原型，Keyhole给了卫星影像（后来的Google Earth），ZipDash给了手机实时交通。2005年2月8日上线。
四个月后开放API——而这个决定是被倒逼的：黑客们"盗用"地图数据做出了housingmaps这类拼贴应用，谷歌选择投降式拥抱，把非法用法变成官方产品。它日后最深的护城河之一——开发者生态——起点是别人先侵犯了它。
十三个月的毕业
Street View在2007年上线时被当成玩具，其实那是一支测绘车队。2008年9月，谷歌地图全球独家切换到Tele Atlas的数据；同一时间，内部的Ground Truth项目全速推进——街景影像、政府公开数据、众包修正，加上数千名操作员的人工校验。2009年10月，美国市场弃用Tele Atlas，改用自有数据。
从签下长期独家到毕业，十三个月。它从来没打算长期付费——付费是为了争取自建的时间。这一课我们在这个系列里讲过：把数据卖给有能力自建的平台客户，等于给自己的替代品当老师。
器官的经济学
2009年10月还有另一刀：Android上的导航免费。PND品类当场被判死刑，靠卖导航挣钱的整个生态开始清算。
免费的逻辑是器官逻辑：地图喂搜索，搜索喂广告，广告养地图，器官本身不需要独立核算。直到它需要的那天——2018年，Google Maps Platform重新定价，API价格最高涨了十几倍，免费额度大幅收缩，强制绑定信用卡。开发者大规模出走，流向Mapbox、OSM、HERE和TomTom。这批"难民"，就是四年后Overture成立时的群众基础。谷歌为了让器官对损益表负责，亲手制造了自己的反对派联盟。
而这个器官如今的产出确实惊人：2023年谷歌地图收入约111亿美元，较2019年的30亿大幅增长，其中82%来自广告——promoted pins和基于位置的推广。111亿美元，是TomTom和HERE营收总和的五倍以上。这只是谷歌的一个器官。
飞轮
今天的家底：超过3亿个地点，5亿贡献者的输入在实时喂路线和推荐，接入DeepMind后ETA预测准确率约97%，加上千亿级街景影像和几十亿用户每天的隐性行为反馈。
这个飞轮的每一环都不花采购费——用户免费用产品，产品免费收数据。Navteq当年"雇几百人开遍每条街"的成本结构，对上"几十亿人免费替你走遍每条街"，这不是竞争，是物种差异。
AI三连
2025到2026年，谷歌在地图上打出三连招，方向一致。2025年11月，Gemini替换掉地图里的Google Assistant。2026年3月，Ask Maps和Immersive Navigation上线——地图从查询工具变成对话对象。最重要的是中间那一招：2025年10月17日，Grounding with Google Maps登陆Gemini API——开发者可以把大模型的推理直接接到2.5亿个地点的实时数据上，模型自动识别地理意图，返回带引用的接地回答，附赠可嵌入的交互地图组件，还能和搜索接地在同一请求里并用。
看清这是什么：这就是对"AI空间接地层"叙事的闭源回答。开放阵营讲中立的地点注册表，谷歌直接把注册表、推理和分发打包成一个API卖。快、深、鲜——在它自己的函数里，无懈可击。
王座的裂缝
裂缝有五条，前四条是地图，第五条是命门。
中国是零，全球最大单一市场不在场。韩国受高精数据出境限制，进不去纵深。欧盟两线施压：DMA限制它在搜索里给自家地图开后门，德国卡特尔局的两项决定拆掉了车载服务捆绑和地图混用限制——"用谷歌"不再必须"全用谷歌"。第四条是它自己留下的先例：2018年证明了，当地图被要求向财务负责时，它会牺牲生态信任——这意味着拆售后的单品定价同样可能"合规但昂贵"，同样可能再送一批客户给对手。
第五条最深：AI自蚕食。Ask Maps直接给答案，用户就不再浏览列表、不再点推广图钉——而82%的收入恰恰建立在"用户自己看列表"的交互上。谷歌自己的AI正在消灭自己的收入交互。器官服从整体战略：当搜索被AI重构，地图的KPI也会被重写。器官的宿命，是没有独立意志。
引力场之外
现在回答开头的问题。
在谷歌的函数里——地图价值等于流量乘以广告——没有人赢得了它，这场仗不用打。但市场上始终存在另一个函数的需求：地图价值等于责任乘以合规乘以可组合。这个函数覆盖的空间，恰好是广告函数照不到的地方，而且每一块都是结构性的，不是残羹。
第一块是主权需求。车厂不愿把客户关系和行为数据交给一家广告公司——奔驰宝马大众的自研座舱路线、中国车企出海需要的非谷歌全球数据栈，都是这个需求的形状。广告函数天然与数据主权冲突，这个冲突谷歌无法定价消除。
第二块是责任需求。欧盟强制智能限速、L3自动驾驶的功能授权、车规级SLA——这些生意的本质是"出错我赔"，签的是责任而不是流量。广告函数不签责任：免费的东西没有赔偿条款。谷歌可以把地图做到97%准，但它的商业模式不为那3%负责。
第三块是制度化的可组合权。德国拆绑案给了换的权利，Overture给了换的原料，NDS给了换的插座——三件套齐了，"谷歌之外"从一种姿态变成一种可施工的架构，而且有十年期承诺和独立受托人兜底。这是监管亲手划出的空间，谷歌进不去，因为空间的定义就是"非谷歌"。
第四块是谷歌自己送出来的。2018年的涨价难民建起了Overture，GAS拆售后的定价如果重演"合规但昂贵"，会送出第二批。巨头向财务负责的每一次，都是中立商的获客季。
第五块在Agent时代。Grounding with Google Maps再好，苹果不会用，亚马逊不会用，任何一家把Agent当战略的平台都不会把地点层交给最大的竞争对手。Agent经济需要一个中立的空间接地层，这个位置在定义上就不能姓谷歌。
这五块空间加起来不是一个"打败谷歌"的市场，是一个"谷歌之外"的市场。
为什么名字是TomTom，不是HERE
有必要多问一句：这个空间里明明站着两家"非谷歌"的车规图商，凭什么归其中一家？表面看HERE的资格甚至更老——装车存量更大，欧美有几千万辆车在用它的驾驶辅助数据。答案在于，第二函数的每一个因子，都在逐项淘汰它。
先看可组合。这个空间是刚被制度化的新空间——德国案、Overture、NDS三件套划出来的——而三次规则改写，TomTom是投诉方、创始方、在场方，HERE三次缺席，只在别人的承诺条款里被点名。更根本的是架构：可组合市场需要的是"开放底图加可插拔图层"，TomTom的Orbis就是按这个形状造的；HERE守着的垂直一体化全专有栈，恰恰是不可组合的那种模式。空间奖励按它的形状造船的人，不奖励旁观者。
再看责任。责任函数的本质是"出错我赔，赔到2035年"——签这种合同需要一个能活到赔付日的主体。一家股东年年减值、每季度都在回答"为什么还持有"的公司，无法可信地承诺十年SLA；车厂采购的风控评估会在"供应商存续确定性"这一栏直接扣分。责任生意先比谁能签字，再比谁的数据好。
第三看中立，这一条最致命。HERE由奥迪、宝马、奔驰持有——它的"中立"只对自己的股东成立。对股东的竞争对手，尤其是正在全球市场与BBA贴身肉搏的中国车企，把车队数据回流给对手股东控制的供应商，是采购桌上的竞争禁忌。TomTom没有任何主机厂股东，对所有车厂等距——第二函数里的"中立"必须是无条件的，有条件的中立不叫中立。
最后看成本。底图价格趋零的时代，全专有栈意味着继续独自背负整张底图的维护费，第二函数的利润率会被这笔成本慢慢压垮；TomTom已经付完了转换学费——Genesis重建产线、底图换成公共品，省下的钱全部押在动态层和责任层上。
所以准确的说法不是TomTom打败了HERE，而是第二函数的四个因子——可组合、责任、中立、成本——逐项淘汰了HERE。空间选择了按它的形状造船的那家：把底图换成开放原料、把自己重建成车规工厂的阿姆斯特丹公司，手里握着24亿欧元订单、德国案投诉方的规则书写者身份，和大众系亲手背书的第一批合同。空间是结构性的，守不守得住看执行——但空间本身，谷歌拿不走，HERE也接不住。
四十年地图战争，打到最后是两个估值函数的战争。王的函数无人能夺；但王的函数也有它照不到的地方。那里不大，但足够养出一个寡头。
数据来源：谷歌开发者博客（2025年10月17日Grounding with Google Maps发布）、谷歌2026年3月Ask Maps/Immersive Navigation发布材料、ElectroIQ谷歌地图统计（2023年收入及构成）、Bundeskartellamt B7-25/22决定、Search Engine Land对2009年数据切换与2018年GMP定价调整的报道、本系列前篇。