Starbucks ended fiscal 2025 with 40,990 stores, 16,864 of them in the United States and 8,011 in China, and it closed 627 stores under the restructuring it announced on 25 September 2025 (Starbucks Q4 fiscal 2025 results, 29 October 2025). Those numbers are the reason the Starbucks store locator is one of the most-read location datasets in retail analytics: site-selection teams, competitors, real-estate researchers and mapping projects all want to know where the stores are, and the answer changes by hundreds of stores a quarter. Add menus and prices that differ by country and by store, and seasonal items that exist in one market and not another, and you have the jobs people put proxies in front of Starbucks for.
We run country-pinned and metro-pinned residential traffic for exactly these jobs, and on 24 August 2026 we also looked at starbucks.com from a datacenter server so the front-door section below is observed.
Which proxy type fits Starbucks?
A residential exit in the country whose site you are reading, driving a real browser, with slow pacing on the locator. For store-level prices, metro-pinned US residential exits with one sticky session per store or delivery address, because those prices live in the app and on the delivery marketplaces rather than on the website. Logins get a static ISP proxy each, one account per address. A datacenter address will load a page (we checked), and it is also the first address metered when it pages through thousands of locator queries.
What Starbucks is as a target
Starbucks runs company-operated and licensed stores across dozens of markets, each market with its own website, its own menu and its own app. Three surfaces carry the data. The store locator on each country site returns the stores near a set of coordinates, with address, hours and services such as drive-through and mobile ordering, which makes it a coordinate-paged database of the whole estate. The menu pages show what each market sells, including regional and seasonal items. And the app shows a specific store's prices once that store has been chosen, which is also where Starbucks Rewards lives; on delivery marketplaces, the same store's delivery prices appear per address.
The website, in other words, is where the locations and the menus are. The prices are one layer deeper, behind a store choice.
Jobs a proxy is the right tool for
- The location dataset. Sweeping the locator across a country to build or refresh a list of every store with hours and services. Site-selection analysts use it to model trade areas, competitors map it against their own estates, researchers correlate it with everything from rents to foot traffic, and 627 closures in one restructuring is why the sweep repeats.
- Menu and price comparison across countries. What a latte costs in Tokyo against London against Seattle, in local currency, over time. The country sites carry the menus; store-level prices come through the app or the marketplaces per store.
- Seasonal and regional item tracking. Items launched in one market, limited-time returns, exclusives that never leave a country. Food media and competitor marketing teams track them per market, which means reading each market from inside it.
- Store-level price monitoring in the United States. Prices differ between stores in the same city. Reading them at scale runs through delivery-marketplace storefronts one address at a time, the method laid out in proxies for DoorDash and proxies for Uber Eats.
- Localization testing. The Japanese site in Japanese with yen where shown, the right seasonal campaign, the right legal text, read from a Japanese exit. The general method is in geo-testing with proxies.
Jobs a proxy is not the tool for
Rewards accounts. They attach to a phone number or email and to the app on a device. Creating accounts to reuse offers runs against Starbucks' terms, and none of the links that would need to change is the network address.
Gift cards and stored value. Checking or probing card balances in bulk is fraud, not research. We do not support it, and we say so here so nobody reads this page as a hint.
App offers by location. The app chooses the store from the phone's location. Behind a proxy the network sees a new address and the GPS sees nothing different.
The front door, observed
The Starbucks homepage came back to our datacenter server in full on 24 August 2026: HTTP 200 to one request with a normal browser user-agent, assets loading from Starbucks' own content domain. No challenge page, no block. The delivery marketplaces, measured the same evening, challenged the same client (Cloudflare on DoorDash and Uber Eats), and McDonald's stalled it outright; Starbucks simply answered.
An open homepage says nothing about the locator's limits. Locator endpoints are the natural place for per-address metering, because a human asks for stores near one point a few times and a scraper asks for stores near ten thousand points in an hour. That limit is not published and not something we have measured, so the honest instruction is to log the first rising 403 or 429 rate per exit and treat it as the number. What a 403 or a 429 means, and how to back off, is in 403 errors through a proxy and 429 rate limits.
The size of the Starbucks estate, fiscal 2025
- 40,990
- Stores worldwide
- 16,864
- United States
- 8,011
- China
- 627
- Closed in the 2025 restructuring
Source: Starbucks Q4 and full fiscal year 2025 results, 29 October 2025
Which proxy type fits, by job
| Starbucks job | Proxy type | Why |
|---|---|---|
| Locator sweep for a location dataset | Residential in the country, slow pacing | Thousands of coordinate queries from one address is the metered pattern |
| Menu and price comparison across countries | Residential in each country | Each market serves its own site to the country it sees |
| Store-level US price monitoring | Rotating US residential, metro-pinned, sticky per address | Prices are per store, read through marketplace storefronts |
| Seasonal and regional item tracking | Residential in the target country | The market decides what the menu shows |
| Any login or partner portal | ISP (static residential) | One fixed, trusted location per account |
| A one-off manual look at one page | Datacenter, or a fresh free proxy | The homepage answered a datacenter client; the address will not survive a sweep |
Datacenter vs residential proxies has the background on the two types. For the locator job specifically, proxies for Google Maps covers the coordinate-grid method on the other big location source and most of it transfers.
Free versus paid for Starbucks
Because the homepage answers datacenter clients, a fresh proxy from our free proxy list can load a menu page or a handful of locator results once, which makes Starbucks friendlier to free than the delivery marketplaces. It is still not a plan: free proxies are shared datacenter addresses that die within minutes, and a locator sweep is the exact pattern that gets a shared address metered. Use free to test your parser against one market, verify each proxy in the proxy checker first, and read when free proxies are fine for the line. For the sweep itself, our residential proxies start at $0.44/GB, pay as you go, no KYC, and pin to any market Starbucks operates in.
How to set it up
- One country, one exit. Pin the residential exit to the country whose site you are reading. The country sites are separate properties, and the seasonal campaigns, language defaults and risk checks all key off where you appear to arrive from.
- Sweep the locator on a grid, slowly, and cache. Locator queries are coordinate-based. Cover the country with a grid sized to the locator's result radius, ask each cell once at a human pace, and diff against the last sweep rather than re-reading everything. A closed store shows up as a missing row, which is the whole point of the schedule.
- Use a real browser where the page renders client-side, and keep a firm timeout. Playwright or Puppeteer with the proxy set at launch; the Playwright proxy guide walks through the options.
- For store-level prices, go through a store choice. Either the app, which is one store at a time on a device, or the delivery marketplaces, one address per sticky session with a metro-pinned exit. Record the store number and the timestamp with every price.
- Pace and watch the block rate. Keep gaps between reads, randomize their length, skip images, and slow down at the first rise in 403 or 429 responses. The wider set of rules is in avoiding IP bans while scraping.
How many IPs
Size by grid cells and stores, not by menu items.
US locator sweep, monthly:
grid cells to query ~6,000 at a 15 km radius
queries per exit per hour ~30, paced
exits in rotation ~20-30 US residential
Country menu index, 20 markets, monthly:
pages per country ~15
exits needed 1-2 residential per country
Residential is billed by traffic. Locator responses are small, so a
full US sweep is a modest bill; the pacing, not the bandwidth, sets
how long it takes.
Because the balance on our pricing never expires, a monthly sweep costs a month of traffic and nothing in between.
Where a proxy stops
Strip it down and the proxy is an address in the right country or metro, nothing more. The app still locates by GPS, accounts and stored value stay off the table (and we will not help anyone put them on it), and Starbucks' terms, which prohibit scraping, apply through any exit. What residential exits make possible is narrower and real: a legitimate data job reading the public locator, menus and marketplace storefronts the way local customers do, one market or one store at a time, at a human pace, with a block rate you watch.
For the chain where the same three layers come with a much harder front door, see proxies for McDonald's, and for a general treatment of scheduled reads across any retailer, proxies for price monitoring.
Sources
- Starbucks Corporation, Q4 and full fiscal year 2025 results, 29 October 2025 (40,990 stores; 16,864 in the United States and 8,011 in China; 627 stores closed under the restructuring announced 25 September 2025): investor.starbucks.com
- HProxy observation, 24 August 2026: one direct request to starbucks.com from a datacenter server returned the homepage (HTTP 200).