Instrumenting infrastructure with networked sensors is no longer speculative. Great Britain now has more than 41 million smart and advanced meters installed, the US Department of Energy counted almost 65 million American customers with smart meters by 2015, and one Tennessee utility has published more than a decade of measured outage data. The record those deployments left behind is less tidy than the pitch: a British program the National Audit Office costed at 13.5 billion pounds after three missed deadlines, a Chattanooga build with gains measured by national laboratories, and a botnet assembled entirely from factory passwords.
Forty-one million meters, and the bill that came with them
The Department for Energy Security and Net Zero publishes quarterly statistics on the rollout. Its Q1 2026 report, published on 28 May 2026, states that over 41 million smart and advanced meters were in homes and small businesses across Great Britain, and that 72 percent of all meters are now smart or advanced, with over 38 million operating in smart mode or classified as advanced. A further 5.3 percent sit in traditional mode despite being smart-capable, which is to say the hardware is installed and the telemetry is not arriving.
The pace is falling. In the same quarter, large suppliers installed 590,000 smart and advanced meters, down 1.0 percent from the previous quarter and 18 percent from the same quarter in 2025.
The cost record comes from a different publisher, and the distinction matters. The National Audit Office, Britain’s independent public auditor, reported in November 2018 that the original 2016 estimate of 11.0 billion pounds had already increased by at least 0.5 billion pounds, an extra 17 pounds per household. It put the cost at 374 pounds per dual fuel household against household savings of 18 pounds a year. Its June 2023 follow-up revised the program cost to 13.5 billion pounds.
The 943,000 meters that stopped talking
The most instructive part of the British experience is not the total spend but the fleet management failure underneath it. First-generation SMETS1 meters were tied to the supplier that installed them, and installation continued long past the point the plan called for. Every figure below is the National Audit Office’s, not a supplier’s.
- The department planned for 5.4 million first-generation SMETS1 meters. Around 12.5 million were actually installed.
- Around 70 percent of SMETS1 meters go dumb when a household switches supplier, and 943,000 had already lost their smart functionality when the NAO reported in 2018.
- The first SMETS2 meters were installed in July 2017, over three years later than first planned.
- By March 2023 approximately 3 million meters, roughly 9 percent of the 32 million-plus then installed, were not operating as intended.
- The completion deadline slipped three times: from 2019 to 2020, then to 2024, then to 2025.
Chattanooga’s numbers were measured by national laboratories
EPB, the municipal utility in Chattanooga, is the best-documented single utility IoT deployment in the United States, largely because federal laboratories studied it. Lawrence Berkeley National Laboratory records 1,200 automated circuit switches and sensors on 171 circuits at a total cost of about 48.4 million dollars, built on a 2009 grant under the American Recovery and Reinvestment Act. LBNL measured the outcome: SAIDI fell 45 percent, from 112 to 61.8 minutes per year, and SAIFI fell 51 percent, from 1.42 to 0.69 interruptions per year. It puts normal-operations savings at about 26.8 million dollars annually for customers, and calculated an avoided customer interruption cost of 23 million dollars for one severe weather event in July 2012. The federal money also compressed the build from 10 to 12 years down to 2.5 years.
Oak Ridge National Laboratory went further into the plumbing. Report ORNL/TM-2017/246, published in June 2017 by Michael Starke, Ben Ollis, Jim Glass, Alex Melin, Guodong Liu and Isha Sharma, describes an AMI network of more than 175,000 meters, 72,328 intelligent power controllers, 214 motor-operated switches used to isolate faulted sections, roughly one 12 kV intelligent recloser per 150 customers, and more than 167,000 points of data with 20,000 points available for control. ORNL itemized returns as well as outlay: theft detection worth about 5 million dollars per year, meter reading savings of approximately 2 million dollars per year by 2014, conservation voltage reduction cutting demand charges by between 2,434,140 and 2,681,014 dollars, and power factor savings of 198,230 dollars in 2015. Against that it lists 1,500,000 dollars for the distribution management system, 1.5 million dollars for SCADA software, 5,854,056 dollars for smart grid management implementation and 8,062,324 dollars for demand management.
EPB also publishes its own account, and it should be read as such. The utility’s 2025 community benefit study, peer-reviewed and conducted by Bento Lobo of the University of Tennessee at Chattanooga, claims 5.3 billion dollars in community benefit over 2011 to 2024, more than 417.7 million minutes of outage prevented, an annual average outage reduction of about 59 percent, and 10,420 jobs. Those are larger claims than the laboratory measurements, and they come from the operator.
Mirai showed the failure mode was a password list
The largest documented consequence of mass device deployment was not a clever exploit. Antonakakis and colleagues from Georgia Institute of Technology, Akamai, the University of Illinois, the University of Michigan, Google and Cloudflare analyzed the Mirai botnet at the 26th USENIX Security Symposium in Vancouver in August 2017. Across 1 August 2016 to 28 February 2017 they observed a peak of 600,000 infections, a steady state of 200,000 to 300,000, and 55.4 million IP addresses scanning.
The original malware carried 62 username and password pairs; variants eventually used 371 unique passwords across 48 distinct dictionaries. The infected population was DVRs, IP cameras, routers and printers, with Dahua, Huawei, ZTE, Cisco, ZyXEL and MikroTik among the top manufacturers. Traffic reached 623 Gbps against Krebs on Security on 21 September 2016, and the same botnet hit OVH that September and Dyn on 21 October 2016.
Peter Fuhr of Oak Ridge National Laboratory, writing about sensor research across EPB’s 600 square mile territory, put the lesson plainly: ‘It is difficult to overstate the important role of cybersecurity in the communications link between such deployed sensor systems.’ Nothing in the Mirai record required sophistication. Default credentials on unmanaged embedded devices were sufficient.
Which figures deserve which weight
Reading this material well means sorting it by who produced it. Five kinds of evidence appear above, and they are not interchangeable.
- Independent audit. The National Audit Office reports on a program it does not run, which is why its cost and failure figures outweigh departmental commentary.
- Official statistics. DESNZ publishes the quarterly meter counts on a fixed schedule, naming its next publication date, so the series can be checked over time.
- National laboratory measurement. LBNL and ORNL measured EPB’s reliability and itemized its costs, and the measurer had no revenue riding on the answer.
- Federal program self-reporting. The Department of Energy’s final Smart Grid Investment Grant report, dated 20 December 2016, records 3.4 billion dollars invested under the 2009 Recovery Act inside a total of 7.9 billion dollars, more than 1,300 phasor measurement units deployed, 19 projects saving 316 million dollars in operations and maintenance over three years, and customers on time-based rates cutting peak demand by up to 23.5 percent. Useful, and reported by the funder.
- Operator self-publication. EPB’s 2025 study may well be accurate, but it is the utility describing itself, and blending it with the ORNL measurements misleads the reader.
Sources: Department for Energy Security and Net Zero · National Audit Office · Lawrence Berkeley National Laboratory · Oak Ridge National Laboratory · USENIX Security Symposium · US Department of Energy
