The Map Wars · Part 6
On July 13, 2026, the Overture Maps Foundation put out a release: membership had passed 50 organizations, nearly double 2024's count.
The number isn't the point. The point is the headline — "Industry Converges on Open Data to Ground AI."
Google is not mentioned once.
Recall what everyone understood when this organization launched in December 2022: AWS, Meta, Microsoft and TomTom gathering under the Linux Foundation to build open map data. Translated: a group of companies unwilling to let Google set the rules, pooling together to turn the base map into a free public good.
Three and a half years later, "anti-Google" has been dropped. The subject is now AI.
When an organization changes its story, either the old story stopped working, or the new story buys something the old one couldn't. Overture is the second case. And the player who has played this board most completely is the only professional mapmaker among the four founders.
A Partnership of Separate Agendas
The founding structure decides what it is today.
The base map is a capital-intensive commodity. Google burns billions of dollars a year on maps and can afford to, because billions of daily users vote with their feet and the data flows back for free. Meta, Microsoft and Amazon had no interest in replicating that bill individually — and less interest in resting their location capabilities on an adversary they fight across ads, cloud and commerce.
So the division of labor was set. The internet giants contribute "data exhaust": Meta's AI-extracted building footprints, Microsoft's own building datasets, and later members like Uber and Tripadvisor pouring in pickup points, ratings, foot traffic — byproducts of their businesses.
TomTom contributes the craft. Open data arrives in bulk and in chaos — mixed sources, inconsistent schemas, unguaranteed error rates. Between "downloadable" and "usable in production" lie validation, normalization and entity deduplication: the dirty work professional mapmakers have done for thirty years. Open base for scale, professional curation for quality; drop either side and the combination fails.
The Linux Foundation contributes the neutral shell. Without it, four companies with misaligned interests never sit at one table.
By the July 2024 GA, the estate was: 2.3 billion building footprints, 54 million POIs, divisions and base datasets, transportation in beta, addresses in alpha (200 million across 14 countries). Free. Anyone can download. No membership required.
The agendas were never unified: Microsoft feeds Copilot, Meta feeds Llama and smart glasses, Amazon wants cheaper map procurement, TomTom sells Orbis. Beneath the public-good coat are four private ledgers. That isn't a defect — it's precisely why the structure runs.
Three Steps by One Company in 2022
Widen the timeline and three moves around 2022 come from a single hand, composing one strategic action.
Step one, outside the courtroom. In 2022 TomTom complained to the German Federal Cartel Office about Google Automotive Services' bundling — Reuters confirmed the complainant. The result landed on April 9, 2025: the unbundling decision took force. Google Maps, Play and Assistant must be sold separately; exclusivity-linked revenue shares deleted; default-placement obligations deleted; commitments valid for ten years, with an independent trustee watching enforcement. Cockpit software's rules were rewritten from "bundle versus bundle" to "layer versus layer."
Step two, on data. December of the same year: co-founding Overture, pushing Google's heaviest asset — the base map — toward commoditization. Your wall is a contract; I dismantle contracts. Your moat is data scale; I turn data into a public good.
Step three, on product. The same period: Orbis Maps — Overture's open base underneath, TomTom's proprietary live-traffic, lane-level and automotive-grade quality layers on top, sold as commercial services: rendering, search, routing, navigation, real-time traffic. Microsoft's Azure Maps runs on TomTom — the most mature commercial outlet in the whole system.
Three steps, one logic: turn the giant's strongest asset into a public good, and drag the competition back to the layer where craft still decides outcomes. In the history of European technology companies versus American platforms, offense this deliberate is rare.
Three Reasons to Change the Story
Back to the narrative upgrade.
The most immediate reason is commercial. "Anti-Google" is a defensive posture with a low ceiling — the companies willing to fund that cause can be counted worldwide, and 2024's roughly 26 members marked the story's natural limit. "AI infrastructure" is an offensive posture: in theory, everyone building AI is a prospective member. Read the new-member list: beyond Grab, Uber and Samsara appear Fresno County — a county government — and UC Santa Cruz. Governments and universities aren't joining to fight Google; they're seeking a neutral spatial-data baseline. Nearly all of the doubling came from the new story.
The second reason lives in the technology, and the gap is real. Large models and agents have no stable reference for physical places. The same restaurant carries different names, coordinates and IDs across data sources; when alignment fails, the model hallucinates about the real world — recommending a restaurant that closed last year, routing a car to a charging station that doesn't exist.
Overture's answer is GERS — the Global Entity Reference System: a globally unique ID for every spatial entity, letting every data source align around it. The June 2026 Places update refreshed operating-status fields for US POIs, using members' anonymized aggregate signals to judge whether a business is still open.
The ambition, stated plainly: GERS wants to be the DNS of physical places. The internet needs the domain-name system to keep "addresses" uniquely resolvable; the agent economy needs an equally neutral registry of places. Whoever holds that registry is root infrastructure for AI's spatial layer. A map shifting from "a picture rendered for humans" to "an entity store machines can cite" is this category's largest redefinition in twenty years. The previous one was 2005, when Google turned the map from a navigation tool into a local-search gateway.
The third reason is the stealthiest: the new story starts a data flywheel the old story couldn't. The release's most notable mechanism is called heartbeat signals — Meta, TomTom, Tripadvisor and Uber continuously injecting pickup points, ratings, foot traffic and traffic patterns to keep the data fresh. Google Maps' true moat was never the base map; it is freshness. An open base map can't touch that moat — but "grounding AI," as a public-good narrative, gives commercial companies a respectable reason and a neutral rulebook for continuously donating business data. The heartbeat mechanism is the alliance accumulating a freshness flywheel of its own.
Two Angles for Readers in China
Fifty members, and no mainland Chinese company on the list — Grab is Southeast Asian, Hyundai is Korean. The reasons are unsurprising: surveying licenses, data sovereignty, geopolitics; China runs its own parallel system.
But Chinese carmakers going global are precisely this system's natural customers. Export programs need a non-Google worldwide data base — partly cost, partly hedging: staking the overseas cockpit's lifeline entirely on Google means wiring yourself into the US–China crossfire, and last year's opening-then-suspension of the SAMR antitrust probe into Google demonstrated the shape of that risk. Since the German unbundling decision took force, "Play store plus third-party navigation" is a lawful architecture with a cartel-office decision behind it; and an Overture base plus a neutral European supplier's commercial layers is the most readily available data foundation under that architecture — one contract for global coverage, GDPR-native compliance, zero competitive overlap with any carmaker's own data assets. Cockpit-sourcing question lists are due for a rewrite.
The second angle is inside the car. The cockpit agent is one of the highest-cost venues for spatial hallucination: booking restaurants, finding chargers, planning waypoints all require entity alignment across apps and data sources. Recommending the wrong restaurant is an experience bug; routing a low-battery car to a nonexistent charger is an incident. Unified entity IDs — infrastructure that sounds dull — will become a hard requirement in cockpit AI first. In the next round of cockpit data-foundation competition, this is the position few are watching and worth occupying early.
Ignore the Press Release; Watch Two Numbers
Whether this upgrade succeeds won't be settled by communiqués. Watch two things.
One: GERS's citation surface — how many non-member AI products, how many agent frameworks, adopt GERS IDs as the default identifier for places. DNS is DNS because you can't be online without it; GERS is far from that seat, but whether the direction holds will be visible within a year or two.
Two: the contribution structure of heartbeat signals — whether members donating data come to outnumber members merely consuming it. A public good's worst fate isn't disuse; it's being used by everyone and fed by no one.
Three and a half years ago, four companies gathered to dismantle Google's wall; now they say they're laying foundations for AI. The dismantling is real; so is the foundation-laying. The first buildings are already going up on that foundation — Orbis is the earliest one standing. The quality of those buildings will set the foundation's valuation, and decide whether "AI spatial infrastructure" ends as a category leap or as a beautifully written press release.
Sources: Overture Maps Foundation / Linux Foundation announcement of July 13, 2026; July 2024 GA release notes; Bundeskartellamt decision B7-25/22 (April 9, 2025); TomTom Orbis product materials.
《地图战争》系列 · 第6篇
2026年7月13日，Overture Maps Foundation发通稿：成员突破50家，比2024年翻了将近一倍。
数字不是重点。重点是标题——"Industry Converges on Open Data to Ground AI"，行业正在汇聚到开放数据上，为AI做接地。
通篇没提谷歌。
这个组织2022年12月成立的时候，所有人都心知肚明它是干什么的：AWS、Meta、微软、TomTom四家凑在Linux基金会下面共建开放地图数据。说白了，就是几家不想让谷歌定游戏规则的公司，合伙把底图做成免费公共品。
三年半过去，反谷歌不提了，改谈AI。
一个组织换故事，要么旧故事讲不下去了，要么新故事能换来旧故事换不来的东西。Overture属于后者。而这盘棋里下得最完整的，是创始四家里唯一的职业地图商。
各怀算盘的合伙生意
底图是个资本密集的商品。谷歌一家每年在地图上烧几十亿美元，烧得起，是因为每天几十亿用户用脚投票，数据自己回流。Meta、微软、亚马逊谁都不想单独复制这笔开销，更不想把位置能力押在一个广告、云、电商全面开战的对手身上。
所以分工是这么摆的。
互联网大厂出"数据废气"。Meta贡献AI提取的建筑轮廓，微软拿出自家的building footprints，后来Uber、Tripadvisor把上下车点、评分、人流这些业务副产品也注了进来。
TomTom出制图的手艺。开放数据的原始形态是散装的，来源杂、口径乱、错误率没人担保，从"能下载"到"敢用在生产上"，中间隔着质量校验、格式统一、实体去重这堆又脏又累的活。职业地图商干了三十年的就是这个。开放底图管规模，专业策展管质量，缺哪边都不成立。
Linux基金会出中立的壳。没有这层壳，四家利益不一致的公司坐不到一张桌上。
2024年7月正式GA，家底是23亿栋建筑轮廓、5400万POI、行政区划和基础地物数据集，交通数据beta，地址数据alpha（14个国家2亿条）。免费，谁都能下载，不用入会。
各家的算盘从来没统一过：微软要喂Copilot，Meta要喂Llama和智能眼镜，亚马逊想压低地图采购成本，TomTom要卖Orbis。公共品的外衣下面是四本私账。这不是毛病，恰恰是这种结构才转得动。
2022年，同一家公司的三步棋
把时间线拉开，2022年前后有三件事出自同一家公司，凑成一个完整的战略动作。
第一步在法庭外。2022年TomTom向德国联邦卡特尔局投诉谷歌车载服务（GAS）捆绑销售，路透社的报道确认了投诉方身份。结果落在2025年4月9日：卡特尔局的拆绑决定生效，Google Maps、Play、Assistant必须能拆开单卖，排他分成条款删除，默认位义务删除，承诺有效期十年，还配了个独立受托人盯着执行。座舱软件的规则从"整包对整包"改成了"图层对图层"。
第二步在数据上。同年12月共创Overture，把谷歌最重的资产——底图——推向商品化。你的城墙是合同，我拆合同；你的护城河是数据规模，我把数据变成公共品。
第三步在产品上。同期发布Orbis Maps：Overture开放底图打底，叠上自有的实时交通、车道级数据和车规级质量图层，作为商业服务卖出去，可视化、搜索、路径规划、导航、实时路况都在里面。微软的Azure Maps就由TomTom驱动，这是整个体系里最成型的商业化出口。
三步是同一个逻辑：把巨头最强的资产变成公共品，把竞争拉回到手艺还能定胜负的图层上。欧洲科技公司对美国平台的博弈史里，这种主动进攻不多见。
换故事的三个理由
回到叙事升级。
最直接的理由是生意上的。反谷歌是防御姿态，天花板太低——愿意为这件事出钱出数据的公司，全世界数得过来，2024年的26家左右就是这个故事的自然上限。"AI基础设施"是进攻姿态，理论上所有做AI的人都是潜在成员。看新增名单就懂了：Grab、Uber、Samsara之外，出现了加州弗雷斯诺县政府和加州大学圣克鲁兹分校。政府和学界不是来反谷歌的，是来找一个中立的空间数据基准的。成员翻倍的增量，几乎全来自新故事。
第二个理由藏在技术里，而且缺口是真的。大模型和Agent对物理世界的"地点"没有稳定指称。同一家餐厅在不同数据源里是不同的名字、坐标、ID，对齐失败，模型就在真实地点上产生幻觉：推荐一家去年就关门的店，把车导到一个不存在的充电桩。
Overture的答案叫GERS，Global Entity Reference System，给每个空间实体一个全局唯一ID，所有数据源围绕这个ID对齐。2026年6月的Places更新刷新了美国POI的"经营状态"字段，用成员贡献的匿名聚合信号判断一家店还开不开门。
GERS的野心，说穿了是想做空间实体的DNS。互联网靠域名系统保证"地址"唯一可解析，Agent经济也需要一个中立的地点注册表。谁握着这个注册表，谁就是AI空间层的根设施。地图从"渲染给人看的图"变成"机器可引用的实体库"，这是这个品类二十年来最大的一次定义变更。上一次是2005年，谷歌把地图从导航工具变成本地搜索入口。
第三个理由最隐蔽：新故事能启动旧故事启动不了的数据飞轮。通稿里有个机制叫心跳信号（heartbeat signals），Meta、TomTom、Tripadvisor、Uber持续注入上下车点、评分、人流、交通模式，维持数据鲜度。谷歌地图真正的护城河从来不是底图，是鲜度。开源底图动不了这条护城河，但"AI接地"这个公共品叙事，给了商业公司持续捐数据一个体面的理由和一套中立的规则。心跳机制攒的，就是联盟自己的鲜度飞轮。
中国的两个切口
50家成员名单里没有中国大陆企业。Grab是东南亚的，Hyundai是韩国的。原因不难猜：测绘资质、数据主权、地缘环境，中国有自己的平行体系。
但中国车企出海，恰恰是这套体系的天然用户。出海车型需要一套非谷歌的全球数据底座，既是算账，也是避险——把海外座舱的命脉全押在谷歌身上，等于把自己嵌进中美博弈的火线，去年SAMR对谷歌反垄断调查的开与关，把这种风险的形状演示了一遍。德国拆绑决定生效后，"Play商店+第三方导航"是有卡特尔局决定书背书的合法架构；Overture底图加欧洲中立供应商的商业化图层，是这个架构下最现成的数据底座：一纸合同全球覆盖，GDPR原生合规，跟任何车企的数据资产都不构成竞争。座舱选型的问题清单，该翻篇了。
另一个切口在车里。座舱Agent是空间幻觉代价最高的场景之一：订餐厅、找充电桩、规划途经点，都要跨App、跨数据源做地点实体对齐。推荐错一家店是体验问题，把电量告急的车导到一个不存在的充电桩是事故。统一实体ID这种听着枯燥的基础设施，会最先在座舱AI里变成硬需求。下一轮座舱数据底座的竞争里，这是个很少有人看、但值得提前占位的点。
别看通稿，看这两个指标
判断这次升级成不成，通稿没用，盯两件事。
一看GERS的引用面：有多少非成员的AI产品、多少Agent框架，把GERS ID当作地点实体的默认标识。DNS之所以是DNS，是因为不用它就上不了网。GERS离这个位置还远，但方向对不对，一两年就见分晓。
二看心跳信号的贡献结构：捐数据的成员，数量能不能超过只用数据的成员。公共品最怕的不是没人用，是只有人用、没人喂。
三年半前，四家公司凑在一起拆谷歌的城墙；现在它们说自己在给AI打地基。拆墙是真的，打地基也是真的。地基上第一批房子已经动工，Orbis是最先立起来的那一栋。房子的成色决定这块地基的估值，也决定"AI空间基础设施"这个故事，最后是一次品类跃迁，还是一份写得漂亮的通稿。
数据来源：Overture Maps Foundation / Linux Foundation 2026年7月13日公告、2024年7月GA发布说明、Bundeskartellamt B7-25/22决定（2025年4月9日）、TomTom Orbis产品资料。