One of the biggest surprises at WWDC last year was Apple’s announcement that its new map would cover “the entire U.S. by the end of 2019”:

Four years in the making, Apple’s new map covered just 3% of the U.S. at its launch in September 2018:

And in the eight months that followed, up until WWDC 2019, Apple was only able to expand its new map to an additional 8.2% of the U.S.:

8.2% over eight months is faster than 3% over four years—but even at that rate, Apple wouldn’t cover the U.S. until 2026:

So why was Apple confident it’d cover the remaining 88.8% of the U.S. with less than seven months left in 2019?

But Apple’s WWDC announcement wasn’t the only surprise regarding the new map. Equally surprising was that it had launched with a very noticeable coverage break across California:

Surely, Apple didn’t mean to ship it that way? It probably just ran out of time?

But that’s why it was so surprising: after four years of mapping, Apple couldn’t finish a full state?

Whatever the reason, it seemed as if Apple would expand to the rest of California next. But it didn’t. Instead, Apple’s first expansion was to a group of islands half an ocean away (and as far from Cupertino as parts of the East Coast):

It’d be another two months until Apple completed California, erasing the coverage seam:

And then two months later, Apple expanded from California into the Southwest:

But still, why didn’t Apple expand to Southern California first? Why Hawaii?

The simplest answer is that Hawaii was actually the first state completed by Apple’s data collection vehicles:

Apple wouldn’t finish driving Southern California until two months later, which makes sense given that Southern California is roughly fives times larger than Hawaii:

But what’s interesting about Southern California being fives times larger than Hawaii is that the third area Apple expanded to (the Southwest U.S.) is roughly five times larger than Southern California:

And when looked at this way, a pattern emerges. Not only were Apple’s second and third expansions bigger than the expansions before—they were roughly five times bigger:

So Apple was accelerating—which helps explain Apple’s confidence going into WWDC:

And then in the months after WWDC, Apple’s new map continued expanding—doubling in size every couple of months:

And by late January, it had grown to cover the entire Continental U.S.:

So Apple missed its goal by just 30 days—a far cry from finishing the U.S. in 2026.

But if you look at everything else Apple was doing over the past year, you start to see why it might’ve just missed its deadline. That’s because Apple wasn’t just expanding—it was also adding new detail to the areas it had already mapped, such as here in San Francisco:

When the new map launched in 2018, Apple replaced its old San Francisco building dataset with a new, home-grown one:

But since launch, Apple has updated many of these buildings again—increasing their accuracy and detail:

And some of these upgraded structures are spectacularly detailed, such as Grace Cathedral:

The Cathedral of Saint Mary of the Assumption:

And the Dutch Windmill in Golden Gate Park:

Apple also adjusted all of the buildings with the height and detail oddities I wrote about in “Apple’s New Map”:

But even more surprising than Apple adjusting these buildings was the speed at which it had done it.

When we compared Google and Apple Maps in San Francisco in 2017, it took Apple three weeks to add a missing park shape and eleven weeks to add the shapes of missing park paths:

This time, however, Apple fixed the shapes within a week:

Even more impressive was the speed at which Apple corrected the POI issues in Markleeville and Bridgeport. Back in 2017, it took Apple almost two weeks to address the POI issues mentioned in “A Year of Google & Apple Maps”:

But this time, Apple corrected them in 48 hours:¹

Here we see why it was so important for Apple to start owning its own data: errors are fixed in hours and days, instead of months and weeks. And Apple is getting so good at this that its San Francisco map now generally seems more up-to-date than Google’s.

For example, here’s Chase Center in San Francisco (the new home of the Golden State Warriors) on Google and Apple Maps back in May:

Apple not only had a shape and label for Chase Center before Google, it also had the realigned road on Chase Center’s east side:

We see the same at the San Francisco Police Department’s headquarters (completed in 2015). Apple, again, had the building before Google:²

Meanwhile, here’s California Pacific Medical Center’s new campus, located along one of San Francisco’s busiest streets (U.S. 101):

Opened last March, this twelve-story hospital occupies an entire city block in central San Francisco. Apple has had its shape since 2018, but Google still has the shape of the Cathedral Hill Hotel, demolished five years ago.

And Google, oddly, didn’t start labeling the hospital on its map until October:

A similar situation exists at the site of 181 Fremont, San Francisco’s third tallest building. Apple has the skyscraper—but Google still has the site’s former structures, demolished six years ago:

Around the block from 181 Fremont, Apple had a full 3D model of Salesforce Tower, San Francisco’s tallest building, more than a year before Google—even though Google had detailed 3D models of the surrounding towers:

And Apple even has buildings that aren’t finished yet—such as 1500 Mission, pictured below (with crane):

Apple has a flat 2D shape for the structure’s base, while Google still has the site’s former buildings, demolished in 2015:

Whenever we compared Google and Apple in San Francisco over the past few years, Google was always more up-to-date-than Apple. But now it’s the opposite. (And keep in mind, every building mentioned above is within two miles of Google’s San Francisco office.)

But Apple hasn’t just been expanding and adding detail—it’s also been removing detail...

One of the strangest things regarding Apple’s new map at its launch was that Apple had included shapes for private, backyard tennis and basketball courts:

Even though these features were visible in Apple’s “Satellite” mode, their addition to the map seemed at odds with Apple’s privacy stance (especially since user privacy was one of the stated motivations behind Apple’s mapping initiative).

But days after I mentioned them, they quietly disappeared:

And as the new map expanded to Beverly Hills, Brentwood, and Bel Air, they never resurfaced:

Given the circumstances, it seems as if Apple had never intended to add them. And across the new map, we see a number of other peculiarities that seem unintended.

For example, similar to what we saw in San Francisco in “Apple’s New Map”, a number of Los Angeles highrises also have height oddities, such as Bank of America Plaza:

And City National Plaza:

And sometimes Apple adds new data on top of old data, such as here in Napa Valley where Apple has been adding buildings on top of a pond:

The pond was part of a winery (August Briggs) that was demolished in 2015—and a luxury resort is now under construction in its place:

Even though they were removed, Apple still has the winery’s old pond and parking lot and has been adding the new buildings on top of them:

While it’s unclear how much of Apple’s new map is human-made vs. machine-made, these don’t seem like errors a human editor would make. Instead, they suggest that Apple’s mapmaking process is increasingly automated—much like Google’s.

Given the pace of Apple’s U.S. expansion, it seems that Apple’s initial map, which it had spent four years making (and showed hints of human labor), served as a training set for the extraction algorithms that are now accelerating Apple’s mapmaking effort.

And if that’s the case, it explains why Apple’s San Francisco map is now seems more up-to-date than Google’s—and why so many of Apple’s buildings have been adjusted since launch:

In other words, Apple is catching up with Google by using the same computer vision and machine learning techniques we saw in “Google Maps’s Moat”:

But these aren’t the only parts of Google’s mapmaking playbook that Apple is following. Apple also adopted Google’s collection vehicles and appears to be driving the U.S. faster than Google did:

And like Google, Apple now collects data on foot:

Apple is also, like Google, using probe data to add new features to its map. When the new map expanded to Los Angeles, a number of walking paths disappeared:

But in the days that followed, the paths reappeared one-by-one (likely via iPhone probe data):

And we see the same back at the demolished Napa Valley winery. Even though it’s a fenced-off construction site—without paved roads—Apple appears to getting probe data from the iPhones of workers and adding little paths:

But Apple isn’t just adopting Google’s mapmaking strategies, it’s also adopting Google’s release strategies. The state-by-state rollout of Apple’s new map is reminiscent of how Google Maps was released country-by-country:

And Apple has been pre-releasing sections of its new map to random subsets of its users, weeks before releasing them to its larger user base—a strategy Google adopted in the wake of Apple Maps’s infamous 2012 launch.

Much of what we’ve learned about Google’s mapmaking process has come via exclusive stories with Wired, The Atlantic, and Fast Company:

And even here Apple is mirroring Google, revealing its new map back in 2018 via a TechCrunch exclusive (rather than during a keynote):

But as Apple’s map-making process becomes increasingly like Google’s, Google’s look and feel is, in recent months, becoming more like Apple’s.

Perhaps the most noticeable difference between Apple’s old and new maps is all of the vegetation detail Apple has added:

And now Google is adding it too:

Meanwhile, when Apple Maps launched in 2012, it featured hand-drawn icons for landmarks in New York and San Francisco:

And Apple has since added hundreds more:

But now as of last Fall, Google is adding them too:

And two months after Apple started adding sidewalk detail in New York and other U.S. cities...

...Google now appears to be doing the same:

So can Apple become more like Google faster than Google can become more like Apple?

__

1  Apple made the corrections so quickly that some who had read “Apple’s New Map” in the days after it was published assumed I had gotten the Markleeville POI issues wrong.

↩︎

2  All of the screenshots of Google Maps and Apple Maps in this section were taken during August 2019.

By the time you read this, the maps may have changed. ↩︎