Gigaton at Mokra: kiln optimisation needs a governed control loop

External-source research synthesis, not an OfferSignal client result. Based on the retained October 2026 research and source-specific dates.

A cement kiln must balance fuel, air, temperature, speed and raw-material chemistry while keeping clinker quality stable. A lower fuel setting is not automatically better: fuel mix, moisture, sensor behaviour and laboratory timing can make a recommendation unsafe, uneconomic or unsuitable for the required product.

Gigaton, the 2026 rebrand of Carbon Re, addresses this pyroprocess rather than cement sales or enterprise resource planning. Its strongest named public implementation is at Heidelberg Materials’ Mokra plant in Czechia, where Carbon Re AI was integrated with ABB Ability Expert Optimizer. The case describes a plant workflow and an evaluation design, not merely an abstract promise. 1

What the Mokra case reports

Cambridge Enterprise reports that, during the first month of continuous operation, Mokra’s fuel-cost index fell 4.1%. Fuel-derived emissions were 4.5 kg CO2 lower per tonne of clinker, reported as a 2% reduction, and C3S variance fell 33%. A separate one-month-on and one-month-off closed-loop evaluation, controlling for fuel mix and raw-material quality, reported a 2.2% reduction in specific heat consumption. 1

These are the most specific public results in the record, but remain a university-linked/provider case rather than a public independent audit. The fuel-cost number is an index, not absolute currency; the emissions boundary is fuel-derived rather than total lifecycle emissions; and the C3S variance definition and statistical confidence are undisclosed. The on/off evaluation is short and has not been publicly replicated.

From plant data to an operating target

The described workflow combines plant historian, sensor, laboratory and chemical data. Public material names fuel flows, temperatures, pressures, atmospheric conditions, mechanical speeds, electricity and quality data; six months of history and existing automation are recommended. A kiln-specific, physics-aware and self-learning model forecasts candidate actions, which can become a dynamic target or closed-loop setpoint through existing advanced process control, with operator visibility. The kiln explanation describes oxygen-target optimisation, fuel-air balancing, automatic setpoint updates and retraining as fuel mix or hardware changes. 2

This places the system inside an existing control environment rather than treating a dashboard as the operating decision. But the public evidence does not document approval thresholds, interlock separation, rollback, audit logs, maintenance lock-outs or final authority. Resilience Media says plant staff still changed kiln weights and balances when measurements indicated a shift, so universal autonomous operation is not demonstrated. 4

Separate named evidence from portfolio claims

2026 announcements name Adani Cement, Heidelberg Materials and Holcim; Clean Growth Fund additionally names Mannok. Gigaton and investors claim annual savings of 1–3 million US dollars and approximately 30,000 tonnes of CO2 per plant. 3 CemNet and ESG Today repeat the funding, deployment and outcome claims, but provide no plant-by-plant audit. 5 6

A separate official Carbon Re case for an unnamed Brazilian producer reports 25% lower clinker-free-lime variance, eliminated XRD/XRF analysis lag and quality estimation between lab measurements; its plant, period, baseline, denominator and causal design are undisclosed. 7 An investor article claims about £1.3 million annual savings, 6,000 tonnes of carbon avoided and 33% lower output variability at one unnamed plant. These remain investor-reported claims, not general customer ROI. 8

An illustrative plant-readiness test

For an illustrative cement producer, start with one kiln and a bounded objective such as specific heat consumption while preserving clinker quality. Assemble aligned historian, sensor, fuel, laboratory and chemical records; document units, timestamps, fuel moisture and chemistry, hardware revisions and missing-data rules. The process engineer and control-room owner should define operating ranges, approval steps, override rights and stop conditions. The authorised plant team remains responsible for safety and release.

Replay representative historical periods, then run a controlled shadow or on/off evaluation. Measure the declared baseline, energy-price and emissions boundaries, throughput, clinker denominator, quality variance, operator interventions and rejected recommendations. Version the model and configuration, retain input lineage and record overrides. This tests a governed process, not whether a forecast sounds plausible.

What is not established

Gigaton is a B2B enterprise plant-deployment software/service described as cloud-based; public pricing, contract terms, implementation fees, service levels and ROI guarantees are undisclosed. The record does not show RFQ or CPQ, quoting, contracts, logistics, certification, invoicing or maintenance scheduling. The public GitHub organisation shows tooling and forks, not an industrial-control source, licence or self-host package. The apparent gigaton.ai domain currently serves unrelated music-creator software. 9 10

Sources and next reading

• Gigaton/Carbon Re resources — public informational index; no self-serve operational access or pricing.
• AI-readiness guidance — public guidance on plant data, automation and cloud requirements.
• Kiln-combustion explanation — public technical explanation; no authenticated control access.
• UK Companies House record — public legal-entity record, not product access.

Explore OfferSignal · Focused Business & AI Workshop · All research

References