Embodied Carbon Observatory
The grid vs. process problem: why most carbon improvement data is noise.
There are 40k+ environmental declarations in the industry’s largest open database for US concrete plants. Each one is a timestamped statement: here is the carbon intensity of this product, at this facility, as of this date.
Every one of those declarations is being misread.
Not by everyone, and not always. But systematically, by the analysts, credit teams, and capital allocators who rely on improving carbon numbers as a signal of operational progress. The misreading is structural. It’s baked into how the number is constructed. And it’s producing materially wrong conclusions about which companies are genuinely decarbonizing and which ones just happen to be located in the right place on the electricity grid.
This is what we built the Embodied Carbon Observatory to fix.
What the carbon number actually measures and what it doesn’t
The carbon intensity figure in a building materials environmental declaration is a cradle-to-gate number. It captures the emissions associated with producing one unit of a material (one cubic yard of ready-mix concrete, say) including raw material extraction, transport to the plant, and manufacturing.
Electricity consumed at the plant is a direct input to that number. The emissions factor for that electricity comes from the regional grid the plant sits on: the mix of coal, gas, solar, wind, and hydro generating power in that part of the country, updated annually by the federal government.
Here’s the consequence: when a region’s grid gets cleaner because a state mandates renewables, because a utility retires a coal plant, because wind capacity comes online: every plant in that region gets a lower carbon number automatically. Without doing anything. Without changing their process. Without any management decision whatsoever.
For plants on the Western grid (eg.: California, the Mountain West) this effect has been dramatic over the past decade. The grid has gotten substantially cleaner. Carbon numbers for concrete plants in those regions have improved as a result. That improvement is real in a physical sense. But it is not a signal of anything a company did.
For anyone making investment or procurement decisions based on it, it’s noise masquerading as signal.
The misallocation this produces
Consider a large materials sector fund running a decarbonization-tilt strategy. They’re overweighting cement and concrete companies whose carbon trajectories show consistent improvement. The thesis: these companies are ahead of the transition curve, they’ll face lower regulatory risk, they’ll have lower capital requirements as carbon costs tighten, and their long-run value should be adjusted upward accordingly.
That thesis is correct but only if the carbon improvement is driven by what the company actually did.
If the improvement is driven by the grid getting cleaner, the thesis inverts. The company has done nothing. When the grid plateaus which it will, because you can only retire so much coal, the improvement stalls. The company that looked like a decarbonization leader is actually a company with no operational advantage, facing the same capital requirements as everyone else, just later.
The credit team underwriting their bonds on a favorable risk assumption is wrong. The equity model assigning a premium is wrong. And the sustainability rating that put them in the top quartile of their peer group is wrong: not because of bad intent, but because the underlying number doesn’t separate what the grid did from what the company did.
No financial data terminal publishes this decomposition. No sustainability rating agency produces it. Industry analysts don’t have it. The signal is sitting inside the data. It just hasn’t been extracted.
The decomposition
The framework is straightforward once you see it.
For any plant with multiple environmental declarations across time, you have a carbon intensity timeline. You also have a grid emissions timeline for the region that plant sits in: the federal government publishes this annually. The attribution question is: of the total carbon change between two periods, how much is explained by the grid getting cleaner, and how much required something at the plant to actually change?
The grid contribution is mechanical. If the regional grid’s emissions dropped 20% and the plant’s electricity consumption per unit of output stayed constant, you’d expect carbon intensity to drop proportionally just from that. Anything above or below that expected drop is the process signal.
More precisely, the counterfactual asks: what would this plant’s carbon intensity be today if the only thing that changed was the grid? You take the plant’s baseline carbon number, scale it by how much the grid has changed since then, and that’s your counterfactual. The difference between the counterfactual and the actual number is the genuine contribution from plant operations: positive or negative.
A plant whose actual carbon intensity is below the counterfactual has done real work. They changed their mix designs. They sourced lower-carbon inputs. They improved process efficiency. That’s the signal.
A plant whose actual carbon intensity tracks the counterfactual closely has done nothing. The grid moved, they moved with it.
A plant whose actual carbon intensity is above the counterfactual has gotten operationally worse even while benefiting from a cleaner grid: the most dangerous case, because the headline number may still look acceptable while the underlying trajectory is deteriorating.
What we built
Embodied Carbon Observatory operationalizes this decomposition across 40k+ environmental declarations for US concrete plants.
For each plant, for each period between consecutive declarations, we compute the total carbon change, how much came from the grid, how much came from process, and a verdict: genuine improvement, grid tailwind, mixed, or getting worse.
And the counterfactual timeline: for every declaration in a plant’s history, what would the carbon number have been if only the grid changed from the starting point? That’s the line on the chart that lets you visually separate what the macro environment did from what the company did.
Aggregated across a company’s full plant fleet, this becomes a carbon quality score that reflects actual operational trajectory, something no existing data provider publishes with this level of specificity.
Why the database architecture matters
This is fundamentally a time-series join problem: plant carbon history on one side, regional grid emissions history on the other, matched by geography and year across a decade of data.
We built on TimescaleDB, a time-series database that automatically pre-computes annual carbon rollups as new declarations land, so queries that would take seconds run in milliseconds. Range scans across years of plant history stay fast even as the dataset grows. The counterfactual calculation runs as a single database query without a separate analytics layer on top.
The alternative would be separate systems for time-series data and relational data, with an pipeline stitching them together. Instead it’s one database, one query engine. The map loads in milliseconds.
The applications
Transition risk. The question for bond investors isn’t what is this company’s current carbon intensity. It’s whether the trajectory is durable. An operationally driven improvement will persist. A grid-driven one is contingent on continued grid decarbonization in that specific region, which the company has no control over and which will eventually slow. That distinction should affect how you price the debt and structure any sustainability-linked covenants.
Equity valuation. Cement and concrete manufacturers are long-duration assets. Long-run value assumptions are highly sensitive to where you set the carbon cost trajectory. A company with genuine operational improvement has a structurally lower cost path. A company riding grid tailwinds doesn’t. Getting this wrong in a valuation model isn’t a small error.
Procurement. The best long-run concrete supplier isn’t necessarily the one with the lowest carbon number today. It’s the one whose operational trajectory is genuinely improving. A plant with mediocre current numbers but real process improvement is a better procurement partner than one with good current numbers that has been coasting on a favorable grid.
Investment due diligence. For anyone evaluating materials sector exposure (whether at the fund level or the asset level) the process attribution scorecard is the first honest look at whether portfolio companies are actually decarbonizing or just appearing to.
What’s next
The immediate next step is company-level aggregation: rolling plant attribution up to manufacturer fleet level. That’s the view that matters for capital markets, where you’re underwriting a whole business, not one facility.
Steel is the next material category. The distinction between electric arc furnaces and traditional blast furnaces maps directly to grid intensity in the same way: making the same decomposition framework immediately applicable, and arguably more consequential given steel’s role in construction and infrastructure.
The demo is live.
The map shows full plant coverage across the US; detailed attribution data is available by request.
If you’re running a materials sector book, doing environmental commodities work, or building in the embodied carbon data space: ankur@dexdogs.earth