Whoever owns the map that agents and robots read to understand the physical world owns critical infrastructure. We think the durable version is levered to one technology: the camera.
A reference for the thesis — the seven mapmakers we've backed, the public and private comps that frame them, and the axes we classify all of them on.
When Warren Buffett was our age, he bought shares in the Sanborn Map Company. The business was simple: as fire insurers expanded nationally, they needed to price risk block by block — without putting boots on the ground in every town. Sanborn's surveyors reported updates to a New York headquarters, whose cartographers printed thousands of highly-detailed maps into 50-pound books.
On a unit basis Sanborn spent about $175/yr to produce and maintain the map for a typical town; the insurers — of which there were at least two in any town — paid about $125/yr each for access. Costs scaled with coverage; revenue scaled with demand. The unlock was reselling each map to several customers at once.
The mechanics have barely moved. Instead of surveyors in cars, today's mapmakers collect data from cameras on drones, telescopes, billboards, smartphones, autonomous vehicles, and AR headsets — and resell to oligopolists in food and parcel delivery, telecom, airlines, ridesharing, e-mobility, and commodities trading.
But the thesis is called machine maps for a reason. We believe AI — agents, robots, autonomous systems — will grow to become the biggest buyers of maps of all. Increasingly sophisticated AI needs a richer, faster understanding of the physical world around its embodiment, whether that lives on your phone, your glasses, or your car.
AI challenges the mapmaker's model the way it challenged software. Token-based consumption means agents demand ever-more-granular maps — and selling many smaller maps makes the average map harder to resell several times, which is the whole engine of operating leverage.
The failure mode is a granularity mismatch: a gap between the level at which a mapmaker can control the cost of producing an incremental map, and the level at which it resells that map. Planet Labs' 200+ LEO satellites give it near-global daily coverage, so the incremental cost to point a satellite at a new angle for a new customer is low — days and thousands of dollars. Its cheapest API tier is $2.7K for a one-hectare hex: small enough to expand cheaply, large enough to resell many times. Matched.
Mapmakers are perpetually at risk of disruption, and the disruptors come from non-obvious places — the paper road-map makers were beaten by a defense contractor, then a search engine, then a non-profit, all inside thirty years. But some technologies are more extensible than others.
Passive optical sensors — cameras — are the single most powerful tool in the mapmaking toolkit. The sun is the strongest source of energy in the solar system, so harnessing it has structural cost advantages over almost any other method. Cameras, like solar panels, are an economic innovation as much as a technical one: commoditized hardware on the semiconductor cost curve, no software lock-in, billions of new units a year.
Our reverence for the camera lens has kept us from investing in machine maps levered to any other technology — radiowaves, lasers, radar — even when talented founders disagree with us.
At investment committee we keep asking the same question. Such a business would be granularity-matched — the hex small enough that expanding coverage stays cheap, large enough that each hex resells multiple times. It would own a map that gives agents and robots superpower-like capabilities relative to what users expect today. And it would be run by an entrepreneur with the focus and meticulousness to build a world-class map, plus the relationships and hustle to drive its utilization and monetization.
Mapmakers are converging on one model: an up-front SaaS fee (or minimum commitment) to get in the door, consumption-based fees on map access, and consumption-based fees on compute — with ever-more-complex rate cards to drive price discrimination. TomTom raised API-credit prices +50% this quarter; Mapbox raised its map API +10x in 2019 and its search API +2–4x in 2024.
Globally, we believe mapmakers broadly defined grow from roughly $10–20B of revenue at 20% operating margins today to $100B of revenue at 35% margins over a decade. We've made a concentrated, non-consensus bet on the camera-levered version.
Each sources data from cameras already in circulation — telescopes, phones, cars, drones — and each is granularity-matched by design. Open the taxonomy to see how they classify against the comps.
The listed mapmaker universe that frames the thesis — 14 public & formerly-public names across four categories, from orbital imagery to road-map databases. Valuation, margins, and where each sits on the operating-leverage curve. Pick a category to isolate its companies.
Our seven camera-levered mapmakers, plotted where each sits on the thesis: the space its map covers, against how granular that map is. Status is live from our CRM at build time — public view shows the fund and one-liner only.
Rows = spatial domain the map covers · columns = granularity (coarse → fine). Only companies we've backed appear here — public comps live in the Taxonomy explorer.
Public fundraise announcements across camera-levered mapmaking — registry companies and newly-surfaced startups. Amounts and rounds parsed from the announcement; duplicates across outlets collapsed.
Funding rounds, pricing moves, M&A, and product launches across the mapmaking space — aggregated per registry company and thesis keyword. Filter by event, company, or taxonomy axis.