BRENNEN

Industries / Technology / Software and SaaS / Cloud and data centers / Edge computing

Networks

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

Milestones achieved, from 2014 to 2023.

3

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 2014

    A standards group for edge

    In September 2014 six companies published a white paper on "mobile-edge computing": placing servers in the radio access network, the part of a mobile network nearest the user, so apps run close to phones. On 24 September 2014 ETSI, Europe's telecom standards body, formed an Industry Specification Group to write common rules for it.

    etsi.orgportal.etsi.org

    Credit

    ETSI; Huawei, IBM, Intel, Nokia Networks, NTT DOCOMO and Vodafone; Nurit Sprecher of Nokia, who led the group

  2. 2015

    Exchange rebuilt as data center

    In June 2015 AT&T and ON.Lab showed CORD ("Central Office Re-architected as Data Center") at the Open Networking Summit in Santa Clara. A central office is a local telephone exchange building; in CORD it was rebuilt from ordinary servers and plain "white-box" switches running open-source software.

    LessonIn November 2017 AT&T still called CORD a reference platform that "is not yet production-ready", while its backers said 70% of operators planned to deploy it; a working demonstration is not a working network.

    lightwaveonline.comrcrwireless.com

    Credit

    AT&T and ON.Lab

  3. 2017

    "Mobile" becomes "multi-access"

    On 28 March 2017 ETSI renamed the group Multi-access Edge Computing, keeping the initials MEC, and widened its work beyond mobile networks to LTE, 5G, fixed lines and Wi-Fi.

    LessonEdge computing did not need to be tied to one kind of network.

    etsi.org

    Credit

    ETSI ISG MEC; Alex Reznik of Hewlett Packard Enterprise, elected chair

  4. 2019

    Germany licenses private 5G sites

    From November 2019 Germany's regulator, the Bundesnetzagentur, let organizations apply for local licences in the 3.7 to 3.8 GHz band, so a firm can run its own 5G network, and the computers behind it, on its own site; by November 2025 it had granted 484. Mercedes-Benz's Factory 56, opened on 2 September 2020 with O2 Telefónica and Ericsson, sends all its data to a separate 5G core network on site, where it is processed.

    LessonThe largest group of licence holders is research and public bodies (31%), ahead of IT and telecom firms (28%) and metal and electronics firms (23%).

    bundesnetzagentur.debundesnetzagentur.detelefonica.de

    Credit

    Bundesnetzagentur

  5. 2020

    Cloud servers in phone networks

    On 6 August 2020 AWS and Verizon opened the first two AWS Wavelength Zones, in Boston and the San Francisco Bay Area: Amazon cloud servers placed inside Verizon's 5G network, so phone traffic reaches them without crossing the wider internet. Outside tests disagree on what that buys. AvidThink found no meaningful difference in November 2020 between the Bay Area zone and AWS's nearby regular region, while a 2023 study on Verizon's mmWave 5G, its short-range high-frequency band, measured 17.27 ms on average to a Wavelength server against 44.08 ms to the region.

    LessonDemand grew slowly; in February 2024 STL Partners, an analyst firm, called network edge computing "still nascent" and found that 51% of customers ran only two or three servers.

    press.aboutamazon.comavidthink.compar.nsf.govstlpartners.com

    Credit

    Amazon Web Services and Verizon; measured by Roy Chua (AvidThink) and by Fezeu and colleagues (PAM 2023)

  6. 2021

    Operators hand cores to clouds

    On 30 June 2021 AT&T agreed to move its 5G core, the central software that runs a mobile network, onto Microsoft's Azure cloud, and Microsoft bought the Network Cloud platform the core ran on. On 4 June 2024 Light Reading reported that Microsoft was cutting staff in Azure for Operators and halting previews of its own 5G core product.

    LessonAn operator that hands its core to a cloud firm depends on that firm's priorities, and those changed within three years.

    news.microsoft.comlightreading.com

    Credit

    AT&T and Microsoft

  7. 2023

    One set of network APIs

    On 27 February 2023 the GSMA, the mobile operators' trade body, launched Open Gateway with 21 operators: a shared set of network APIs (application programming interfaces, fixed requests one program can send another), written in the open-source CAMARA project run by the Linux Foundation. The first set included "quality on demand", which lets an app ask for a better connection for a set time, and "edge site selection", which finds the nearest edge server. In September 2024 Ericsson and 12 operators formed Aduna, a joint venture to sell such APIs.

    LessonA shared API on paper is not one every network offers; in March 2026 broadcasters were still asking operators to switch quality on demand on.

    rcrwireless.comericsson.comstreamingmediaglobal.com

    Credit

    GSMA, its member operators and the Linux Foundation; Ericsson and operators (Aduna)

Watch for

  1. A live AI-RAN call

    The RAN, or radio access network, is the radios and computers at and behind cell towers; AI-RAN means running AI work on the same computers. T-Mobile US, Nokia and NVIDIA say they made the world's first AI-RAN voice call, over T-Mobile's live network from a commercial cell site in Bellevue, Washington, with the radio processing on NVIDIA Aerial RAN Computer servers. Running AI inference alongside the radio work is a next step, and Nokia's commercial AI-RAN products are not expected until 2027.

    Achieved whenAI jobs run on the same servers as live radio traffic in routine service, and someone outside the three firms measures or reports it.

    sdxcentral.com

    Claimed by

    T-Mobile US, Nokia and NVIDIA, October 2026

  2. SoftBank to share radio servers

    SoftBank and Nokia plan a 2027 trial on SoftBank's commercial network in which Nokia's radio software runs on SoftBank's AITRAS platform and an orchestrator, software that hands out computing, splits the same machines between AI jobs and radio jobs. In July 2026 SoftBank still described bringing AITRAS into its commercial network "from fiscal year 2026 onward" as a goal.

    Achieved whenAITRAS carries paying customers' traffic while running AI work on the same servers, confirmed by an outside party.

    softbank.jpsoftbank.jp

    Claimed by

    SoftBank Corp. and Nokia, October 2026

  3. Network API revenue claims

    The GSMA says 86 operator groups, covering more than 300 networks, back Open Gateway, and that more than 300 instances of 20 CAMARA APIs have launched commercially in 65 markets, with "well over 130% growth in the revenues". It did not publish the amounts.

    Achieved whenoperators or an independent analyst publish the revenue amounts by API, so that low-latency APIs can be told apart from fraud-check APIs.

    telecomtv.com

    Claimed by

    GSMA (Henry Calvert), April 2026

Open

  1. 01

    Edge-versus-region latency map

    A public, repeatable test in which ordinary phones on ordinary plans, and ordinary home broadband lines, measure round-trip time to two small servers: one in an edge zone (a Wavelength Zone inside a phone network, or a Local Zone in a city) and one in the nearest full cloud region. Run it weekly in every city that has a zone and publish all raw results as open data, alongside the register of edge sites (see Sites).

    The brief
    Credit · unclaimed
  2. 02

    Testing quality on demand

    A test harness for an operator that sells the CAMARA quality-on-demand API to outside developers. It opens a session for a phone, then measures latency and jitter with and without the session at quiet and naturally busy times, and records whether the network sends its notice that it cannot deliver the requested quality. Publish the report and raw data. Do not create congestion; use real busy hours.

    The brief
    Credit · unclaimed

Open 01 · the brief

Edge-versus-region latency map

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public, repeatable test in which ordinary phones on ordinary plans, and ordinary home broadband lines, measure round-trip time to two small servers: one in an edge zone (a Wavelength Zone inside a phone network, or a Local Zone in a city) and one in the nearest full cloud region. Run it weekly in every city that has a zone and publish all raw results as open data, alongside the register of edge sites (see Sites).

Why it is still open

The best independent study found tested one operator, two locations and only mmWave, with a specialist diagnostic tool; AvidThink's 2020 test used one city and one phone. cloudping.info, a public browser tool, measures full regions only, not Local or Wavelength Zones, and AWS has published no measurements behind its "single-digit millisecond" promise for Los Angeles.

par.nsf.govavidthink.comcloudping.info

Built with what exists

The cheapest AWS EC2 server in each zone and in its parent region; aws ec2 describe-availability-zones --all-availability-zones to list zones; Android phones on the carrier AWS names for each Wavelength Zone, with ping and traceroute, which need no rooting; Termux on Android; iperf3 for throughput and jitter (how much the delay varies); cloudping's open-source code.

  1. First step

    Opt in to the Los Angeles Local Zone, start one small server there and one in its parent region, Oregon, and publish a week of pings from your own home line and phone.

  2. Done looks like

    For at least 10 Local Zones and 3 of the 5 listed Wavelength Zones, the median and 95th-percentile round-trip time to the zone and to its parent region, from at least 20 phones or home lines per city, over at least four weeks, published as CSV with a script others can rerun.

Open 02 · the brief

Testing quality on demand

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A test harness for an operator that sells the CAMARA quality-on-demand API to outside developers. It opens a session for a phone, then measures latency and jitter with and without the session at quiet and naturally busy times, and records whether the network sends its notice that it cannot deliver the requested quality. Publish the report and raw data. Do not create congestion; use real busy hours.

Why it is still open

STL Partners counts seven operators with quality-on-demand APIs, most "early-stage", and says the API offers "weaker guarantees than full network slicing" (slicing reserves a separate share of the network) because it depends on public network conditions. The results it cites come from operators and their partners; Orange's published results come from a joint test in Orange's own lab. CAMARA's definition lets the network send a notice when it "cannot fulfill the requested quality profile", but no outside count of how often that happens was found.

stlpartners.comdeveloper.orange.comgithub.com

Built with what exists

The CAMARA QualityOnDemand API definitions, version 1.2.0 on GitHub; an operator developer portal that offers the API, such as Orange Developer; Android phones; iperf3 in UDP mode; one small cloud server.

  1. First step

    Sign up for an operator's test access, read the quality-on-demand 1.2.0 definition, and run ten paired sessions in one place to prove the harness works.

  2. Done looks like

    For at least one operator, at least 100 paired sessions across at least three times of day, reporting median and 99th-percentile latency and jitter, the share of sessions in which the requested quality held, and the share that drew a notice, all with code.