☰

Arbiom

Wood-to-protein scale-up with traceable data

Industry
Fermentation and food biotechnology
Headquarters
Paris, France
Public information as of
February 2026

A4BEE prepared this analysis from publicly available sources. It reflects our own reading of Arbiom's published strategy and is not endorsed by, or produced in cooperation with, Arbiom. Company website

Strategic priorities

Arbiom operates a continuous Torula yeast (*Cyberlindnera jadinii*) fermentation that converts lignocellulosic sugars from forestry and agricultural residues into single-cell protein. The company has accumulated more than 25,000 hours of fermentation data at its demonstration facility in France and has scaled to 200,000 litres of working volume. In February 2022 the French government awarded a EUR 12 million France Relance grant to build the company's first commercial plant, designed for 10,000 metric tons per year and expected to create more than 40 direct jobs in the Auvergne-Rhône-Alpes region.

Two protein products sit on top of the platform. SylPro is positioned for aquaculture and pet food, with a 55 to 60 percent protein content, 98 percent digestibility, and a 20 percent validated inclusion level in rainbow trout diets; in feline feeding trials, 9 out of 10 cats preferred a SylPro-based diet to a chicken-meal control. Yusto extends the same fermentation into human food applications, supplying umami enhancement and texturising for alternative meats, sauces, snacks and processed foods, with EFSA Novel Food authorisation and FDA GRAS designation on the regulatory path.

The company's strategic pillars sit in this order: commercial-scale industrialisation in France, dual-market expansion across animal and human nutrition, regulatory market access in the EU and the United States, and a sustainable protein-independence narrative built on 1.5 to 7 times lower CO2 emissions and 250 times lower water use than conventional protein sources. The co-location with an upstream pulp and paper partner (Norske Skog's Golbey mill sits inside the SYLFEED consortium that validated the technology) is the feedstock-economics backbone of the model.

The immediate operational question is whether the digital backbone — process data, biological outcomes, regulatory evidence and feedstock composition — can be read together as one body of evidence once the commercial plant is running. With 25,000 hours of fermentation history, a multi-jurisdiction regulatory submission in flight, and a feedstock stream that varies by source and season, the data work is what sits between a pilot success and a multi-thousand-ton commercial reality.

Challenges we see

  • Manufacturing Operations

    Running continuous fermentation at commercial scale

    The demonstration plant has run continuously at 200,000 litres for more than 25,000 hours. The commercial plant moves to a 10,000-metric-ton annual throughput, which implies a substantially larger and differently shaped bioreactor system operating under the same continuous regime.

    Where biological performance is proven at pilot scale and the next step is commercial production, the operating envelope that held at 200,000 litres has to be re-established at the larger one — and that means reading the running process, not relying on release tests after the fact.

  • Supply chain Operations

    Matching feedstock quality to fermentation needs

    The fractionation step takes forestry and agricultural residues with variable cellulose, hemicellulose and lignin composition, plus seasonal moisture and inhibitory-compound variation. The SYLFEED consortium with Norske Skog's Golbey pulp and paper mill provided a co-located feedstock stream during validation; the commercial plant intends to extend that co-location model.

    Where sugar yield depends on incoming biomass composition, the supply stream becomes part of the fermentation rather than a separate cost line, so the question of how a new batch of residues will behave upstream becomes a fermentation question rather than a purchasing question.

  • Regulatory Compliance

    Preparing multi-jurisdiction safety dossiers

    Yusto is in flight for both EFSA Novel Food authorisation, which under 2024 to 2025 guidance requires DIAAS (Digestible Indispensable Amino Acid Score) nutritional profiling and advanced allergenicity assessments using targeted serum pools, and FDA GRAS designation in the United States. Torula yeast has decades of food use, which is the load-bearing piece of the safety argument.

    With two frameworks to satisfy and stricter EFSA guidance in force, the bottleneck in market access sits in how quickly evidence can be assembled against the right dossier template — not in whether the safety work itself exists.

  • Automation Digital infrastructure

    Connecting bioreactor data to plant operations

    The new commercial facility requires real-time monitoring and predictive analytics to run continuous fermentation at scale. Industry initiatives such as the Margo edge-interoperability standard from ABB, Rockwell Automation, Schneider Electric and Siemens point to where connected process equipment is heading.

    Where SCADA captures the equipment and the laboratory systems sit beside it, the work of correlating the two is what scales yield — and that correlation is done by hand unless the historian and the lab share a model.

  • Commercial Market

    Differentiating ingredient value in B2B sales

    Arbiom's commercial position runs through application-specific evidence: rainbow trout trials at 20 percent inclusion, feline feeding trials at 9 of 10 cats preferring SylPro, and Yusto's sensory profile across alternative meats, sauces, snacks and processed foods. The application data sits across formulation records, feeding trial reports, and customer feedback channels.

    Where ingredient buyers compare specifications and price rather than carbon and water, the route to a premium position runs through application-specific evidence that the alternative-protein story alone does not generate.

Opportunities, by urgency and business impact

Each bubble is one opportunity, numbered to match the list below. Further right means it bites sooner; higher means a bigger effect on the business. A bigger bubble means a bigger implementation effort.

Source: A4BEE analysis of public sources

  1. Reading fermentation signals while the run is in progress

    Continuous fermentation generates temperature, pH, dissolved oxygen, nutrient feed and cell density data across 25,000-plus hours of operation, but the data is not connected to the biological outcomes (protein content, amino acid profile, yield) in a single model.

    An industrial data platform that unifies the fermentation historian with biological KPIs (key performance indicators) lets the running batch be compared against the best historical trajectories and lets drift be flagged while the culture is still on the line.

    • Arbiom technology and scale-up data, arbiom.com
    • Arbiom press release on EUR 12 million France Relance grant, February 2022

  2. Building the evidence set for EFSA and FDA in parallel

    EFSA Novel Food and FDA GRAS submissions each require their own data package, and the 2025 EFSA guidance adds DIAAS profiling and advanced allergenicity assessments on top of the standard toxicology work. Manually aggregating the underlying records is the bottleneck.

    A regulatory data lake that maps quality, nutritional and process records to jurisdiction-specific templates, with narrow AI agents drafting the dossier sections from source records, lets EFSA and FDA packages move forward from the same evidence base.

    • EFSA 2025 novel food regulation changes, food science press, January 2026
    • Arbiom product and regulatory positioning, arbiom.com

  3. Simulating the commercial bioreactor before commissioning

    Scaling continuous fermentation from 200,000 litres to 10,000 metric tons a year involves different mixing, oxygen transfer and heat removal behaviour. Physical trials at commercial scale are expensive and time-consuming, and the EUR 12 million facility budget cannot absorb many of them.

    CFD-based (computational fluid dynamics) digital twins of the commercial bioreactors, combined with prediction models trained on the existing 25,000 hours of pilot data, let mixing, oxygen transfer and nutrient distribution be checked before the steel is committed.

    • Arbiom demonstration and commercial scale data, arbiom.com
    • Arbiom protein independence industrial project, Bio-Based Industries Joint Undertaking, May 2022

  4. Predicting fractionation behaviour from incoming biomass

    Lignocellulosic feedstock varies in cellulose, hemicellulose and lignin ratios, plus moisture and inhibitory-compound content, and the fractionation pre-treatment step has to compensate for that variation to keep fermentable sugar yields stable.

    Inline NIR (near-infrared) spectroscopy and moisture sensing at feedstock intake, with a model that maps incoming composition to the right pre-treatment recipe, lets the plant adjust ahead of the run rather than diagnose after it.

    • SYLFEED consortium feedstock and fractionation work, 2021
    • Arbiom co-location strategy, arbiom.com

  5. Capturing application data as a sales asset

    SylPro and Yusto are validated across aquaculture, pet food and human food applications, but formulation experiments, feeding trial results and sensory panel data are managed across disconnected records rather than as one body of application evidence.

    A single application-data platform that ties formulation versions to feeding trial outcomes and sensory scores lets the B2B team answer a customer question with the existing evidence instead of running a new trial.

    • SylPro feeding trial results, Skretting, Laxá and Matís consortium reports
    • Arbiom product portfolio and applications, arbiom.com

What we'd propose

  • Enterprise AI

    Industrial data platform for continuous fermentation

    A unified data backbone that connects bioreactor SCADA (supervisory control and data acquisition), the process historian and the laboratory system, so the running fermentation batch can be compared against the best historical trajectories and biological KPIs.

    • Process historian integration

      Fermentation data on one timeline

      Connect bioreactor controllers, pH, dissolved oxygen, temperature and nutrient feed sensors via OPC UA (Open Platform Communications Unified Architecture) or MQTT so process values leave the equipment in a vendor-neutral form rather than staying inside a closed controller.

    • Golden-batch comparison

      Best historical run as the reference

      Overlay the current fermentation against the best-performing batches from the 25,000-plus hours of historical data, so drift shows up against the trajectory that worked, not against a static setpoint.

    • Process-to-yield correlation

      From parameters to outcomes

      A data pipeline that links physical process parameters with biological outcomes (cell density, protein content, amino acid profile) so the parameters that drive Torula yeast productivity are visible rather than inferred.

    • Deviations show up against the running batch instead of against the next release test.
    • One process data set serves fermentation engineering, quality and regulatory evidence.
    • The 25,000 hours of history become a reference for the commercial plant rather than a record of pilot work.

    We've done this before

    See how we do Enterprise AI
  • Agents

    AI agents for novel-food dossier drafting

    Narrow, reviewable agents that draft the recurring parts of EFSA and FDA submission documents — DIAAS profiling, allergenicity summaries, process descriptions — from the underlying records, with a named reviewer approving each output.

    • Drafting from source records

      First drafts from system data

      Generate the first draft of a dossier section directly from the underlying quality, nutritional and process records so the author edits and judges rather than assembles.

    • Jurisdiction template checking

      Gaps found before submission

      Check each draft against the EFSA Novel Food and FDA GRAS templates and the 2025 EFSA guidance on DIAAS and allergenicity, returning missing or inconsistent sections before the document enters formal review.

    • Cross-jurisdiction change search

      Which sections a new study touches

      When a new feeding trial, toxicology study or allergenicity result lands, find every dossier section that references it so the update scope is known on day one.

    • EFSA and FDA submissions draw on the same evidence base instead of two parallel compilations.
    • Submission queues move faster because documents arrive complete against the template.
    • Every output is traceable to the source records and signed off by a named reviewer.

    We've done this before

    See how we do Agents
  • Digital CDMO

    Bioprocess digital twin for factory scale-up

    A CFD-based simulation of the commercial bioreactor combined with prediction models trained on the existing pilot data, so mixing, oxygen transfer and nutrient distribution are checked before the EUR 12 million facility is committed.

    • CFD bioreactor model

      Mixing and transfer in-silico

      Build a computational fluid dynamics model of the commercial bioreactor geometry to predict mixing patterns, oxygen mass transfer, shear stress and nutrient distribution at design throughput.

    • Scale-up prediction from pilot data

      Pilot history as the training set

      Train scale-up prediction models on the 25,000-plus hours of pilot fermentation data so the parameters that most strongly drive yield at 200,000 litres are mapped to their predicted behaviour at the commercial volume.

    • Virtual commissioning environment

      Run the plant before it exists

      Simulate the control sequences, automation logic and operator procedures of the commercial plant against the digital twin, so commissioning starts from a tested baseline rather than an untested design.

    • The EUR 12 million facility spend is de-risked by checking the process before it is built.
    • Commissioning time is shorter because the control sequences have been run virtually.
    • Process knowledge from 200,000-litre operation is converted into design assumptions for the larger volume.

    We've done this before

    See how we do Digital CDMO
  • Digital CDMO

    IoT-enabled feedstock quality monitoring

    Inline NIR spectroscopy and moisture sensing at feedstock intake, paired with a model that maps incoming composition to the right fractionation recipe, so the plant adjusts to the new batch rather than compensating after it.

    • Inline NIR and moisture sensing

      Composition at intake

      Deploy NIR and moisture sensors at the feedstock intake point so incoming lignocellulosic biomass composition is captured as data rather than estimated from a periodic offline test.

    • Fractionation recipe recommendation

      Recipe from incoming composition

      A model that maps incoming cellulose, hemicellulose, lignin and moisture readings to the fractionation pre-treatment recipe — temperature, chemical loading, residence time — that keeps fermentable sugar yield stable.

    • Biomass-to-product traceability

      From forest to finished protein

      A digital chain of custody from the forestry or agricultural source through fractionation and fermentation to the finished SylPro or Yusto batch, supporting life cycle assessment (LCA) documentation and downstream product claims.

    • Fractionation performance is adjusted to the incoming batch rather than diagnosed after it.
    • A wider range of biomass sources can be processed without an offsetting yield penalty.
    • Life cycle and traceability evidence is produced by the process rather than compiled for it.

    We've done this before

    See how we do Digital CDMO

Where QB Systems fits

Alongside our services we build QB Systems, hardware and software for bioprocess control. QB Systems is a product brand of A4BEE Sp. z o.o.

Applications
Bioreactors

Software-defined control for a bioreactor — a new QB vessel, an upgrade to one you have, or a retrofit of the existing PLC.

Buffer & media preparation

Automated preparation of growth media and process buffers, so a recipe runs the same way every time without fixed infrastructure.

Automated sampling

Automated sampling from 4–18 sources, aseptic-capable and up to 72 hours unattended. Works with any vendor's bioreactor.

Scale
Pilot (50–300 L)

Stainless steel, where QB supplies the control software and integration and a certified partner builds the installation.

Digital maturity: today and target

Scored out of 100 across six dimensions. The target is what Arbiom's own published ambition implies — not a perfect score.

Source: A4BEE analysis of public sources
Data Integration 35 → 85
More than 25,000 hours of fermentation history exist at demonstration scale, but the historian, the laboratory system and the upstream fractionation data are not connected as one searchable body of evidence.
Process Automation 45 → 85
Continuous fermentation is operational at 200,000 litres with SCADA in place; the next step is real-time monitoring of biological outcomes and predictive control rather than setpoint following alone.
Analytics & AI 20 → 80
The pilot data set is large enough to train predictive models, but the published information describes no deployed ML (machine learning) system for yield optimisation or scale-up prediction.
Regulatory Compliance 30 → 90
EFSA Novel Food and FDA GRAS submissions are in flight, with 2025 EFSA guidance adding DIAAS and allergenicity requirements; the dossier preparation work is largely manual across two jurisdictions.
Supply Chain Digitalization 25 → 75
Feedstock quality assessment is described as relying on periodic offline testing, with no inline monitoring at intake and no predictive link between incoming composition and fractionation performance.
Customer Experience 30 → 75
Application evidence — rainbow trout trials, feline feeding trials, sensory panels — exists across multiple product lines but is held in disconnected records rather than as a single application-data asset for the B2B team.

Check this yourself

Our Service Portal has free self-assessments and market comparisons. These are the ones that line up with what we've read above — no sales call required.

Think we've read this right?

Talk to us

Related reading

This is an independent analysis prepared by A4BEE from publicly available information as of February 2026. It reflects A4BEE's own interpretation and opinion, is not affiliated with, endorsed by, or verified with Arbiom, and may be incomplete or inaccurate. All company names and trademarks are the property of their respective owners. To request a correction or removal, contact [email protected].