MIT Survey: AI Is Spreading Fast in Supply Chains, but Emissions Aren't Falling With It

MIT's Center for Transportation and Logistics has released its 2026 State of Supply Chain Sustainability report, and the headline finding for anyone building AI into operations is uncomfortable: adoption is widespread and self-reported efficiency gains are high, but net emissions are not reliably moving. According to the report, 78 percent of North American organizations surveyed now use AI in sustainability initiatives, and 87 percent of adopters say it has helped their goals — yet among what the report calls mature agentic AI users, only 53 percent report a reduction in total greenhouse gas emissions, and 29 percent report an increase.

The survey, produced by MIT's Sustainable Supply Chain Lab, covered 1,810 supply chain professionals in 91 countries. It is the seventh edition, and it arrives with a clear through-line: the constraint on corporate sustainability work is no longer executive conviction. It is governance, measurement infrastructure, and the ability to absorb external shocks.

The efficiency-versus-emissions gap

The AI findings are the most instructive part of the report for technical teams, because they show a familiar pattern in a new setting. Respondents report concrete operational wins: 86 percent say AI has reduced material waste, 81 percent say it has reduced energy use, and 63 percent use AI to estimate supplier emissions data they don't actually have. Those are the kinds of results that justify a budget line.

But the same report finds that these wins do not aggregate into lower total emissions as often as you'd expect. The sharpest illustration is warehousing. Among advanced and fully automated warehouses using AI, 78 percent report higher energy use since implementation, and 76 percent report higher waste. Only 24 percent use AI-driven real-time monitoring to track the environmental consequences of what they deployed. MIT's framing is that many organizations may not notice their energy footprint climbing until long after the system is live.

That is a measurement problem as much as an emissions problem. If a company instruments throughput, cost per unit, and order accuracy but not facility-level power draw, an automation program can look like an unqualified success on the dashboard it has while quietly worsening the metric it claims to care about. The report does not break out how much of the energy increase is attributable to compute versus the robotics, conveyance, and climate control that automated facilities tend to require — a detail worth noting, because it changes the remedy.

The gap between "AI reduced our energy use" and "our total emissions went down" is also a reminder that rebound effects are real. Making a process cheaper and faster often means running more of it. The report does not attempt to quantify that mechanism, so the causal story here remains inferential.

Commitment is high; accountability structures are not

The report says 83 percent of respondents rate sustainability as important or very important to long-term business success — the highest reading in the report's seven-year history. More striking, 59 percent say they became more committed following recent U.S. policy shifts that deprioritize environmental concerns.

The structural picture is thinner. Among companies with publicly stated sustainability goals, only 51 percent have a formal cross-functional sustainability team, and only 33 percent place oversight at the board level. The report ties that directly to outcomes: companies with cross-functional teams say sustainability is "frequently or always" built into decisions 57 percent of the time, versus 13 percent where responsibility sits in a single function — a 44-point spread.

That correlation is presented as a return on governance investment, and it's plausible. It is also worth reading with care. Organizations that invest in cross-functional governance are likely to differ in other ways — size, maturity, regulatory exposure — that the report does not control for in the material released. Self-reported frequency of "building sustainability into decisions" is a soft metric.

Where sustainability does embed, it embeds in daily operational choices: transportation and logistics (39 percent), procurement (35 percent), warehousing (34 percent), and product design (33 percent). It thins out in capital investment, network design, and supply chain risk planning — the decisions that lock in emissions for a decade or more. MIT's reading is that companies are getting better at operating sustainably without getting better at designing for it.

Tariffs and geopolitics are eating the plans

The external pressure is substantial. Sixty-eight percent of respondents say tariff uncertainty has pushed them to delay, scale back, or abandon sustainability strategies, and nearly a third say tariff-related disruptions have actually increased their operational emissions — rerouting and resourcing under pressure rarely optimizes for carbon. More than eight in 10 respondents across North America, Europe, and Asia say ongoing geopolitical conflict is reducing the effectiveness of their climate action.

Freight shows how policy clarity translates into capital decisions. Forty-three percent of fleets still run mostly or entirely on conventional diesel, and only 8 percent have completed a full transition to low-emission energy. In Europe, 84 percent of freight companies say recent policy changes have positively affected their decarbonization strategy; in North America, 44 percent do. Battery-electric vehicles are at scale in 60 percent of European companies surveyed versus 36 percent in North America. Capital-intensive transitions stall when the signal is ambiguous.

The one area that moved

Scope 3 collaboration is the report's clearest year-over-year improvement. Participation in joint supplier engagement and decarbonization programs rose from 31 to 58 percent in a single year, partnerships to manage shared suppliers' emissions more than doubled from 23 to 55 percent, and the share of companies working entirely alone fell from 18 to 6 percent. MIT reads this as recognition that indirect emissions cannot be solved by any one firm.

A caveat on sourcing: the report was supported by the C.H. Robinson Foundation, Breakthrough, and CO2 AI and Dassault Systèmes — several of which sell into this market. MIT does not describe what role, if any, sponsors played in survey design, and the published material does not include subgroup sample sizes, the definition of a "mature agentic AI user," or the respondent mix by company size and region. Those omissions matter when interpreting percentages drawn from slices of a 1,810-person sample.

Sreedevi Rajagopalan, director of the lab and lead author, frames the next phase as converting commitment into accountability and incremental decarbonization into genuine transition. On the evidence in her own report, the first practical step is less ambitious and more fixable: instrument the systems you automate before you scale them.

What to watch

• Whether next year's edition separates AI compute energy from facility automation energy in warehousing, which would clarify whether the emissions increase is a model-serving problem or a robotics problem.
• Whether the 24 percent real-time environmental monitoring figure rises as vendors add energy telemetry to warehouse automation stacks — and whether buyers ask for it in RFPs.
• Whether the 44-point governance gap holds up under controls for company size and regulatory exposure, or shrinks once confounders are accounted for.
• Whether the North America–Europe divergence in battery-electric fleet adoption widens, which would make policy certainty the dominant variable in freight decarbonization.
• Whether the sharp one-year jump in Scope 3 collaboration persists or proves to be a reporting artifact from a changed survey population.