Lifecycle analysis — most people in the field write it as life cycle assessment, or LCA — is the standardized method for measuring what a product costs the environment across its entire existence: extraction of raw materials, manufacturing, transport, years of use, and disposal. The framework is defined by ISO 14040 and ISO 14044, and its basic currency is the functional unit: a measure of service rendered — “one year of smartphone use” rather than “one phone.” That distinction sounds pedantic. It is not. It decides everything downstream.
Digital products are among the method’s hardest test cases. Their supply chains cross dozens of jurisdictions. Their manufacturing concentrates impacts in places far from the buyer. Their use phase depends on electricity grids that vary by a factor of ten or more across the regions where devices actually run. For readers of this site, the problem is concrete: Latin America sits at both ends of the digital product lifecycle — the lithium of the Salar de Atacama, the copper of Peru and Chile, the gold that moves through Amazon supply chains, the e-waste and refurbishing economies of São Paulo — yet it appears in most lifecycle inventories only as a proxy or a blank.
In this article I set out where the numbers are weak, what that weakness does to the decisions built on them, and what better practice would require. The short version: the stages with the thinnest data are, inconveniently, the stages located here.
What lifecycle analysis of digital products measures — and where it stops working
Every ISO-compliant study runs through four phases: goal and scope definition, inventory analysis (the LCI), impact assessment (the LCIA), and interpretation. Digital products strain each phase in a specific, predictable way.
- Scope. The functional unit choice. A laptop assessed over three years and the same laptop assessed over eight give opposite conclusions about whether buying new is defensible. Same device. The number flips on an assumption.
- Inventory. Semiconductor allocation. A fab runs 24/7 regardless of what it produces. Dividing its electricity and ultrapure water across hundreds of chip designs is an accounting choice, not a measurement. Published estimates of embodied energy per chip vary by an order of magnitude between studies.
- Impact assessment. Location. A liter of water consumed in the Atacama and a liter consumed in Ireland are not equivalent, but most inventories treat them as if they were. Carbon intensity of electricity runs from under 0.1 tCO2e/MWh on Brazil’s hydro-heavy grid to roughly 0.7–0.8 on the diesel microgrids of the Amazon.
- Interpretation. Uncertainty is real and rarely quantified. Single-point results get quoted. The ranges that produced them do not.
None of these complaints are exotic. They are the standard critiques inside the LCA community, argued at conferences and in methods papers. What makes them worth restating here is the asymmetry: the stages with the weakest data are the stages located in this region.
Three numbers that explain the credibility problem
Manufacturing dominates a phone’s footprint — and that footprint is the least-measured part
Apple’s product environmental reports attribute roughly three-quarters or more of a flagship phone’s lifetime emissions to production, before a buyer charges it once. At sector scale, estimates of ICT’s share of global greenhouse gas emissions span 1.4% — Malmodin and Lundén’s 2018 assessment — to 2.1–3.9% in Freitag and colleagues’ 2021 critique. Both are published. Both are defensible. The spread is the message: the inventory layer is thin precisely where the impact is concentrated.
The manufacturing itself happens in a handful of places — Taiwan, South Korea, the United States. None of them here. But the feedstocks come partly from here: lithium, copper, gold, tin, tantalum. They enter the inventories as coarse global averages, which is a polite way of saying the local conditions vanish.
The streaming correction: two orders of magnitude on one method choice
In 2019 a widely circulated French estimate put an hour of 4K streaming at several kilograms of CO2. Subsequent analysis by the IEA put the figure near tens of grams per hour — lower by roughly two orders of magnitude. The gap did not come from fraud. It came from method choices stacked on each other: outdated data-center energy intensities, generous device power assumptions, no marginal accounting.
I cite this case not to mock anyone. I cite it because it shows how a defensible-looking method produces a number two orders off, which then anchors public debate for years. And the lesson cuts both ways. Bad numbers can inflate alarm; they can also launder complacency. When the underlying inventories are thin, neither direction of the claim deserves much confidence.
E-waste: the least-measured stage, concentrated where data is weakest
The Global E-waste Monitor 2024, from UNITAR and the ITU, counts 62 million tonnes of e-waste generated in 2022, with 22.3% formally collected and recycled. Brazil generates on the order of 2.4 million tonnes a year — the largest flow in Latin America, second in the Americas after the United States.
Most product LCAs cut their boundary at the factory gate (“cradle to gate”) or at a generic disposal scenario. The flows that define end of life in this region — informal collectors, board exports for gold recovery, refurbishing for resale — sit outside the model. Whatever the model does not include, it scores as zero. That is not a metaphor. That is how the arithmetic works.
Where the Latin American data holes sit
Extraction: order-of-magnitude disagreement
Published water-use estimates for brine lithium in the Salar de Atacama range from tens of thousands of liters to more than two million liters per tonne of lithium carbonate equivalent. The spread is methodological. Whether brine counts as “water.” Whether co-produced potassium salts share the burden. Whether the source is operator reporting or the gauge network of Chile’s water directorate, the DGA. A materials engineer in Antofagasta who works on brine monitoring put it to me on background: “We reconcile operator submissions against monitoring wells. The disagreement is routinely tens of percent, and that is before you decide allocation.”
Copper is better served, and worth naming as the exception: Cochilco publishes water and energy use per tonne of refined copper — one of the few public, process-level series in the region. Gold in the Amazon basin is the worst case. Informal and illegal production appears in no inventory, though it enters electronics supply chains and shows up in regional mercury exposure, in Madre de Dios in Peru and Roraima in Brazil. LCA cannot include what no agency measures.
Electricity: one national average hides two Brazils
Brazil’s Ministry of Science, Technology and Innovation publishes an annual grid emission factor, and the number moves. Between wet and dry years it shifts by a factor of two or more — 2021’s drought pushed thermal dispatch and the factor with it. The national average also hides two Brazils. The interconnected system runs hydro-heavy at low intensity. The isolated systems of the Amazon run on diesel at roughly 0.7–0.8 tCO2e/MWh.
So a data center in São Paulo that applies a global average factor of around 0.45–0.5 tCO2e/MWh overstates its use-phase emissions severalfold. The same spreadsheet applied to an Amazon edge site or telecom installation understates them. With São Paulo and Fortaleza — the latter a subsea cable landing hub — absorbing new capacity, the choice of factor is now a material line in corporate reports, not a footnote.
Use and repair: the region extends device lifespans, and models don’t see it
European LCAs typically assume a first life of two to three years. In much of Latin America, the first local owner is the device’s second or third owner globally. An operator in the refurbishing cluster near Santa Efigênia in São Paulo described the standard flow to me: devices arrive with four to six years of prior use, get reconditioned, and run another three to five years locally.
Longer first lives divide embodied impacts over more service-years, which cuts the per-year footprint. Studies that assume short replacement cycles therefore overstate per-year impacts for this region — and erase the informal logistics that make the extension possible in the first place.
End of life: formal registries measure a fraction
The reporting instruments exist. Brazil’s Law 12.305/2010 and its reverse-logistics agreements. Chile’s Law 20.920 of 2016, which created extended producer responsibility. Colombia’s Decree 284 of 2018 on electrical and electronic waste. But registered volumes capture registered collection. The informal sector moves most of the mass, and flows like circuit-board exports for metals recovery are barely tracked. LCA software defaults to European end-of-life scenarios, where formal recycling rates are high; imported wholesale, they overstate formal recovery here by a wide margin. A “recyclable” conclusion built on a European scenario tells you nothing about São Paulo’s actual system.
What better lifecycle analysis would look like
The fixes are known and partly boring, which is a good sign.
- Regionalized inventories, not “Rest of World” proxies. The honest starting layer already exists: Cochilco’s water and energy series, DGA monitoring, MCTI grid factors, Brazil’s SINIR waste registries, Peru’s MINAM mercury inventories, and spatially explicit mining layers from MapBiomas across the Amazon basin. None of these were built for LCA. That is exactly why they are credible — they were built for enforcement and operations, not for storytelling.
- Primary data from the actors, in usable form. Hyperscalers publish PUE and site-level grid data. Manufacturers publish product reports. Miners publish sustainability disclosures. What LCA needs is unallocated, per-process data — the opposite of what corporate communications prefer to publish. Disclosure regulation is changing the format: the EU’s Digital Product Passport under the Ecodesign for Sustainable Products Regulation, in force since July 2024, will require structured product data for goods placed on the EU market, with the battery passport under the 2023 Battery Regulation arriving first, in 2027. Chilean lithium carbonate and Brazilian assembled electronics will feel it through customers’ compliance chains, not local law.
- Uncertainty as a first-class output. Publish ranges and sensitivity tables, not single points. If the streaming correction taught the field anything, it is that a number without its assumptions is a claim, not a result.
- Functional units that match behavior. Regional lifespan data exists in commercial channels — refurbishers, insurers, operators. It needs to enter the public datasets that studies draw on.
- Declared cut-offs. State plainly what is excluded: informal mining, informal recycling, second-hand export flows. Silence should never read as zero.
Methodology is not the bottleneck. Brazil’s PBACV and Chile’s Huella Chile have spent a decade on it. The missing piece is inventory data.
Where this column goes next
This article opens a running thread. Next up: what the Digital Product Passport will demand, concretely, of Chilean and Brazilian exporters. In parallel, I am building a public reference of the datasets behind these articles — which numbers we use, and why.
And an invitation. If you work in refurbishing, collection, or mine monitoring, and you see a published number that does not match your operations, write in. Mismatches are exactly what this column collects.
FAQ
What is lifecycle analysis of a digital product?
A standardized method, defined by ISO 14040 and ISO 14044, for quantifying a product’s environmental impacts across extraction, manufacturing, transport, use, and end of life, expressed per functional unit such as one year of smartphone service. A study runs through four phases: goal and scope, inventory analysis, impact assessment, and interpretation.
Why do published estimates for digital products differ so much?
Because results hinge on method choices: where the boundary is drawn (cradle-to-gate versus full life), how fab energy is allocated among chips, the assumed device lifespan, and the electricity grid used for the use phase. Sector-level estimates of ICT’s share of global greenhouse gas emissions range from about 1.4% to roughly 4%, and that spread reflects assumptions more than measurement.
How much of a smartphone’s footprint comes from manufacturing?
Manufacturer disclosures typically attribute about 70–85% of a flagship phone’s lifetime emissions to production, before a buyer charges it once. The precise share depends on the assumed lifespan and the use-phase grid, but manufacturing dominance holds across studies — which is why supply-chain data quality matters more than charging habits.
Does better lifecycle analysis mean digital products are bad for the environment?
No. It means the current numbers cannot support either dismissal or alarm. Better inventories would identify which products, which locations, and which life stages carry real impact — and which claims are noise.
Will the EU’s Digital Product Passport affect Latin America?
Yes, indirectly. The Ecodesign for Sustainable Products Regulation, in force since July 2024, will require structured product information for goods sold in the EU, with the battery passport arriving in 2027. Latin American exporters of materials, components, and assembled electronics will face these requirements through their customers’ compliance chains.