A chip’s carbon footprint is closer to an allocation than a measurement, and the method a supplier chooses can move the number as much as anything happening on the factory floor. A semiconductor maker can know its fab’s total electricity use and greenhouse gas emissions without holding a directly measured figure for any single device it ships.
Getting from the factory total to a product figure takes another layer of accounting, in which shared power, process gases and equipment time are assigned to manufacturing routes, wafers and finished parts. For technology teams buying chips and the servers built around them, a supplier-specific product footprint is usually more useful than a spend-based estimate. It also carries allocation choices that can make two suppliers’ numbers hard to compare.
Pressure on those numbers is building. Market research firm TechInsights published a forecast projecting that global chip fabrication emissions will rise from 148 million metric tons of carbon dioxide equivalent in 2023 to 255 million by 2031. It expects emissions to grow faster than silicon shipments between 2026 and 2031. The figures are a commercial projection, not reported data. Even so, they give buyers good reason to look harder at what sits behind a per-chip figure.
Industry Groups Have Set a 2028 Target
Work toward consistency is under way. SEMI’s Semiconductor Climate Consortium (SCC) aims to develop, by 2028, a methodology and reporting framework across its value chain so companies can exchange relevant and reliable product carbon footprints. It pairs that with a separate goal of public reporting of relevant Scope 3 emissions by the end of 2028. Both are consortium ambitions and create no reporting deadline for any individual company.
The first concrete step is a SEMI product carbon footprint strategy paper, listed among the consortium’s November 2025 reports and available to members only. Its public description says a SEMI working group and the consortium’s Hyperscaler Peer Group started the process, with the aim of removing barriers to emissions reporting and outlining common methodological rules for chipmaking. The paper describes itself as the outcome of a first phase, with critical issues still to be resolved.
A Factory Total Cannot Describe Every Product
Inside a fab, emissions come from process tools, supporting facilities and the gases used in manufacturing, and different products run through different process sequences. Research institute imec reported in a 2023 analysis that lithography and etch account for 45% of Scope 1 and Scope 2 emissions for 3-nanometer logic wafers. Its work also showed that levers such as higher scanner throughput, lower exposure doses and fewer fluorinated etch gases change the result. Recovering process gases belongs to the same effort, and hydrogen recycling in fabs is already spreading.
Imec has since opened part of its imec.netzero virtual fab to the public. The model draws on data from imec’s own fab, benchmarked with equipment, materials and fab partners, and lets users compare emissions, electricity and water use for several logic technologies built in different locations. Spreading a factory’s annual emissions evenly across every device would erase the differences that kind of model exposes. Allocating by revenue brings its own distortion, because a price change could shift a product’s assigned share while the process stays the same.
Yield Changes the Per-Chip Result
Manufacturing yield adds another variable. Yield is the share of devices that come out usable, and it becomes the denominator whenever emissions are expressed per working chip. If a production run carries the same attributed footprint but delivers fewer good parts, each good part absorbs more of it. Real calculations also have to account for where failed devices leave the line and which later steps they skip.
Two per-chip figures can therefore differ because of manufacturing performance, allocation assumptions or both. Buyers need enough method detail to tell those explanations apart before rewarding one supplier over another.
The Boundary Has to Match Before the Numbers Can
Every comparison also depends on defining the product the same way, since a bare die and a packaged component sit at different stages of production. STMicroelectronics (ST), in a footprint study of a micro-electro-mechanical systems (MEMS) sensor published on its corporate site, describes front-end manufacturing as producing wafers and forming integrated circuits on them. Back-end manufacturing packages the chip so it can connect to a circuit board and stay protected. The study covers raw materials, production, transport, use and end of life, counting only the chip’s own electricity during use.
Purchasers should fix the boundary before requesting data, because a manufacturing-only figure cannot be lined up against a full life cycle estimate that includes electricity consumed in use. ST’s page adds that its results should be considered in isolation unless the study conditions can be fully replicated, and it notes the work has not undergone external critical review. Even a precise-looking result needs its context.
One Data Package Could Replace Duplicate Questionnaires
Incompatible customer questionnaires create extra calculation work without necessarily producing more comparable results. Buyers can ease the load by agreeing with suppliers on a single reusable data package. The essentials are the product and production period a figure describes, its life cycle boundary, and the allocation and yield treatment behind it. Marking which inputs were measured and which were modeled helps the package stay credible as it is updated.
What to Watch Before 2027 Sourcing Decisions
Watch the SCC’s methodology work first. A published rule set for allocation and yield would give buyers a shared reference point, and progress toward the 2028 target will show how soon supplier figures can be compared directly. Modeling platforms deserve a look too. They show how impacts shift across technologies and assumptions, but a modeled technology profile is not a measurement of a particular supplier’s delivered product.
Price will stay in the same supplier conversations, with the rising cost of mature-node chips already weighing on industrial buyers. Chipmakers that already treat environmental limits as planning inputs may be best placed to answer with method detail attached. For 2027 sourcing, the footprint and the method deserve to be assessed together, and a comparison holds up only when suppliers’ figures describe equivalent products and boundaries.
