The startup that took GPS - the thing that tells your phone you're "near a coffee shop" - and turned it into something that lets a robot park itself, to the inch, in a vineyard in Castroville.
It's a Tuesday afternoon in May 2026. Somewhere over a corn field in central Illinois, a small autonomous drone is checking a power line, hovering inside a box twelve inches wide. Somewhere else - a parking lot in Hayward - a robotaxi pulls into a charging stall without scraping a sensor. In a basement in San Francisco, a server inside Point One Navigation watches both of them, and a few thousand others, all at once.
Point One is the company that decided, in 2016, that GPS - that miraculous Cold War leftover that tells you your phone is "somewhere on Mission Street" - had run out of resolution. Cars were starting to drive themselves. Tractors were starting to harvest themselves. Drones were getting jobs. None of it works at the precision of "somewhere on this block."
So Point One Navigation built a different navigation stack: a private RTK correction network called Polaris, an inertial sensor box called Atlas, and a piece of software called FusionEngine that mashes satellites, accelerometers and computer vision into a single number with a centimeter after the decimal point. It then sold all of this to OEMs as a subscription, the way Stripe sells payments. Today it serves customers in automotive, agriculture, drones, robotics, construction and survey - across six countries and counting.
That's the elevator pitch. The harder, more interesting story is what they had to ignore to get here.
Civilian GPS, in raw form, is accurate to about three meters on a good day. Three meters is fine for finding a sushi place. It is not fine for landing a drone on a delivery pad the size of a pizza box.
The fix - real-time kinematic positioning, or RTK - has existed in surveying since the 1990s. The catch is that RTK requires a base station within a few dozen kilometers, broadcasting corrections in a proprietary protocol, often over a radio modem. Surveyors lived with this. So did farmers willing to mount a $20,000 dome on a $400,000 combine. Everyone else did not.
By the mid-2010s a generation of engineers who had built the first autonomous cars - the ones that finished, and didn't finish, DARPA's Urban Challenge in 2007 - looked at the road ahead and saw a bottleneck. You could throw GPUs at perception. You could buy LiDAR. You could not, no matter how hard you tried, change physics: a moving vehicle needs to know where it is, in real time, with very small error, ideally without three engineers and a tripod.
It is, in the most literal sense, a question of where to begin. Hence the company name.
Aaron Nathan and Bryan Galusha had done this before, mostly. Nathan had helped build Cornell's DARPA Urban Challenge car - the school's entry in the original self-driving competition - and then spent years as chief architect at Coherent Navigation, a precision GNSS company quietly acquired by Apple in 2015. The pieces, in other words, were familiar.
Their bet was simple to describe and expensive to execute: build the entire stack themselves. Own the base stations. Own the firmware. Own the cloud. Sell the result as an API. Make it boring to use.
This sounds obvious in retrospect, the way most good bets do. At the time, it was contrarian. The conventional wisdom said precision positioning was a regional, surveying-shop business; that nationwide coverage was a job for incumbents like Trimble or government networks; that nobody outside agriculture would pay a subscription for location.
Point One disagreed on all three counts. They started planting base stations - in parking lots, on rooftops, on cell towers - until the map of North America filled in. Then Europe. Then Korea. The RTK network became, in effect, Polaris: a private, continent-scale grid of reference receivers feeding a cloud that any developer could subscribe to with a credit card and a SIM.
Point One sells one thing in three pieces. The pieces have names that sound like they were chosen by an astronomer with a deadline.
A network of thousands of base stations broadcasting GNSS corrections. True RTK and Virtual RTK live under one umbrella. 99.9% uptime. The boring infrastructure that makes the magic boring.
An automotive- and industrial-grade inertial navigation box. Combines GNSS, IMU and wheel data. Keeps your robotaxi located when it dives into a tunnel and the sky disappears.
The software brain. Sensor fusion across satellite, inertial and vision data, exposed as a clean SDK and a GraphQL device management API. The thing developers actually integrate.
Fleet-level observability for precision devices. Updates, health, status, and a single pane of glass for everything paying a subscription to know where it is.
There are two ways to evaluate a precision-positioning company. One is to read the data sheet. The other is to look at who is paying for the data sheet to be true.
Point One's customer roster, as of its 2025 raise, runs from Zipline (autonomous medical delivery drones) and DroneDeploy (mapping software) to Scythe (autonomous mowers), AeroVect (airport ground equipment), AGRA (precision farming) and Cyvl (infrastructure scanning). It is the sort of list that is more interesting for what it implies than what it shows: every one of those companies has a CTO who decided, after evaluating alternatives, that building a precision-location stack in-house was not worth it.
That is the quiet number behind the loud one. The loud one is the funding: $35M in a Series C in November 2025, led by Khosla Ventures, with existing investors IA Ventures, UP Partners and Alumni Ventures coming back for more. Total raised: north of $52M.
The quiet one is the integration count - roughly 10x growth in manufacturers shipping Point One-equipped hardware in the year before the round.
Point One's stated mission is to "empower the next generation of developers with reliable, easy-to-use precision location data." Which, translated out of startup, means: make centimeter accuracy cheap and trivial enough that nobody has to think about it.
That's the part that matters. The promise of physical AI - cars that drive, drones that deliver, robots that work outside warehouses - is built on a layer of assumptions about where things are. Pull that layer out and the whole tower wobbles. Make it durable, ubiquitous and inexpensive, and an entire generation of products gets to exist that otherwise would not.
"Physical AI" is the phrase Khosla Ventures used in its announcement. It is, predictably, the phrase of the moment. It also happens to be useful here: large language models are catching up on what to do; the harder, less-discussed problem is where to do it.
Back to that Tuesday afternoon.
The drone over the corn field finishes its sweep and lands back on its pad in eleven seconds. The robotaxi in Hayward swaps batteries and drives off to a fare. A surveyor in Yorkshire packs up a $300 antenna and a phone running a Point One app and goes to lunch, having done a day's work before noon.
None of those scenes existed at this scale five years ago, because none of them could afford the precision. Now they can. Multiply that across a hundred industries that don't yet know they need centimeters - construction site logistics, port automation, utility locating, autonomous agriculture, drone inspection, robot vacuums that mow lawns - and you get a sense of why Khosla wrote the check.
Most companies sell a thing. Point One Navigation sells a coordinate. The coordinate is going to be everywhere.
The drone, by the way, doesn't know any of this. It just knows where it is. Which, for a drone, has always been most of the job.