Industries / Technology / Software and SaaS / Cloud and data centers
Edge computing
Edge computing is the work of running computer programs on machines placed near where data is produced, in factories, shops, vehicles and cell towers, instead of sending everything to a distant data center.
1999
In April 1999 Akamai began selling FreeFlow, a service that delivered customers' web files from Akamai servers placed inside other companies' networks; Mahadev Satyanarayanan, writing in IEEE Computer in January 2017, traces the roots of edge computing to it.
39
Milestones achieved across 6 domains.
15
To watch for: announced, not yet confirmed outside.
14
Open deliverables. Nobody has shipped them.
In one minute
Where it started
In April 1999 Akamai began delivering customers' web files from its own servers placed inside internet providers' networks, close to the people asking for them. In 2002 Akamai and IBM set out to run customers' programs on those servers as well, a service later sold as "Akamai EdgeComputing". In 2009 four researchers argued that phones should hand heavy work to "cloudlets", small computers one wireless hop away, and in 2014 ETSI, Europe's telecom standards body, set up a group to put computing inside mobile networks.
Where it stands
IDC, a market research firm, puts worldwide spending on edge computing at $265 billion in 2025 and expects it to nearly double by 2029. Edge computers run in restaurants, stores and factories: Chick-fil-A keeps a small cluster in each of nearly 3,000 restaurants, and Walmart counts 10,000 edge nodes in its buildings. Computing inside phone networks has grown more slowly than its backers expected: data centers at tower bases stalled around 2021, and AWS's list of zones inside 5G networks now names five, none at Verizon, its first partner. The newest work is running AI models on site, where shared tests are thin: the latest MLPerf inference round had one board-sized edge system and no power figures.
What is open
Much of what the industry knows about its equipment comes from the makers: rated temperatures, promised delays, support periods, update reliability. The open jobs are public tests and records anyone can repeat: fanless computers held at their top rated temperature, power cut during software updates, energy per answer for small AI models on cheap boards, measured delays to cloud zones, and registers of edge sites and of how long makers promise security updates.
Domains
AchievedWatch forOpen
Domain
Progress
Last achieved
Watch for
Open
Hardware
This is the work of designing, building and shipping the computers that run where data is made: sealed boxes with no fans for factory floors, small servers for back rooms and cell sites, and cloud companies' racks placed inside customers' buildings.
7 · 3 · 2
Last achievedRaspberry Pis go to industry2024
Watch forRugged servers for −40 °C
2
openSites
This is the work of choosing, building and running the places where edge computers sit: inside internet providers' networks, in cloud providers' city zones, at cell towers and telephone exchanges, in shop back rooms and on factory sites.
6 · 2 · 2
Last achievedPlant servers replace line PCs2023
Watch forAI chips across delivery sites
2
openSoftware
This is the software that installs, runs and updates programs on computers outside big data centers, such as a shop's back-room server, a factory gateway or a delivery network's city site, and keeps their data in step with the cloud when the network comes and goes.
6 · 2 · 3
Last achievedAn offline sync service ends2025
Watch forWebAssembly functions at scale
3
openRunning AI at the edge
This is the work of running trained AI models that see, hear or handle language on small computers on site, in a camera, a shop or a vehicle, instead of sending the data to a distant data center. It covers the accelerator chips and boards, the software that shrinks and runs the models, and the shared tests that measure them.
7 · 3 · 3
Last achievedAccelerator slower than its host2026
Watch forOpen NPU design and compiler
3
openNetworks
This is the work of putting computers inside or beside telecom networks, and of writing the rules and software that let apps find those computers and reach them with little delay.
7 · 3 · 2
Last achievedOne set of network APIs2023
Watch forA live AI-RAN call
2
openManaging many devices
This is the work of keeping thousands of small computers in shops, plants, vehicles and towers set up, updated, secure and counted, without sending a person to each one.
6 · 2 · 2
Last achievedA device cloud shuts down2023
Watch forVersion 2.0 of FDO
2
open
Advances to build with
1 research note
Open in edge computing
3 of 14 open deliverables
Open · Hardware
Heat-soak tally for fanless PCs
A test that holds ten popular fanless edge PCs, one at a time, at each unit's rated top temperature for 24 hours under full load, logging processor speed, internal temperatures, power drawn at the wall, errors and shutdowns. Edge boxes that run AI models all day stay at full load for hours, which makes this matter more than it did.
Built with what exists
A lab temperature chamber, or an insulated box with a heater and a PID temperature controller; a thermocouple data logger; a plug-in power meter; stress-ng for processor load; llama.cpp running a small open model as a steady AI load (see the Note "A first stable number for local AI").
Open · Sites
A register of edge sites
A public, dated table of every announced edge site, starting with AWS Local Zones and Wavelength Zones and then tower-base sites from American Tower and Vapor IO. Each row gives the site's status (announced, live or gone), the dates it was first and last seen, and a link to the page snapshot behind each change.
Built with what exists
The AWS command line tool (aws ec2 describe-availability-zones --all-availability-zones lists the zones each region offers); providers' location pages saved in git for dated snapshots; an AI model that reads press releases and pulls out the site, city, date and status (see the Note "The jobs agents repeat"); Browser Use 1.0, a web agent that drives Chrome, for pages that need clicking (see the Note "Deleting 80% to start over"). Cheap extraction of fields from documents by AI models is what makes the upkeep affordable for one person.
Open · Software
Power cuts on small clusters
A test bench that cuts power to cheap single-board computers running each lightweight Kubernetes (K3s, k0s, MicroK8s and a KubeEdge edge node) 100 times each at random moments, and records whether the machine and its containers come back without anyone touching them, and how long that takes.
Built with what exists
Raspberry Pi 4 or 5 boards, booting from SD cards and from USB solid-state drives; a relay board or network-switched power strip; K3s with its SQLite store and with embedded etcd; k0s; MicroK8s; KubeEdge; a script on a second computer that cuts power and polls each cluster with kubectl.
Where it sits
Broader
