Submission to the Call for Evidence for the EU Biotech Act II

Executive Summary

Europe’s bioeconomy strategy remains grounded in today’s production realities: established industrial infrastructure and valorisation of by-products. This grounding is essential; but bio-digital capabilities, the technologies that will define the next decade, appear only narrowly, mostly as workforce training rather than core enablers. Other actors are moving decisively: China positions AI, bioengineering and synthetic biology as the foundation for replacing fossil-based products. Biotech Act II is the moment to add the future-oriented layer the strategy lacks. Our submission sets out four priorities.

1. The bio-stack and cross-cutting bio-digital capabilities

Industrial biotechnology, like semiconductors, is a stack of interdependent layers (feedstocks, organisms, fermentation, analytics) bound by a cross-cutting bio-digital layer. Public returns and strategic autonomy is captured only by holding the whole stack. The highest-leverage investments are platform capabilities reusable across products: foundation models for biology, computational protein design, predictive scale-up twins, AI process control. We recommend the Commission make the bio-stack the organising concept of Biotech Act II and extend Biotech Act I’s strategic-project model to industrial biotechnology.

2. Strategic scale-up infrastructure

Scale-up failure is a binding constraint on Europe’s bioeconomy. Titre, rate and yield often collapse from lab to factory, a living-systems problem as much as engineering, and precisely the long-horizon challenge public investment exists for. Learning from the US and EU Chips Acts, we recommend funding scale-up as three compounding layers: a federated pilot-plant network, an embedded data layer, and a NIBRT-equivalent training institution for non-health biotech.

3. Data standards

These tools only deliver value if bioprocess data can be read, compared and trusted across sites, instruments and scales, which today it cannot. Because a standard creates value only at sector-wide adoption, no single firm can establish one, and an incumbent-owned standard would entrench enclosure. Building on the EU’s BIOINDUSTRY 4.0 project, we recommend mandating a modular, openly governed Union-wide bioprocess data and metadata standard, with conformance required only at points of exchange and security and IP protection designed-in.

4. Regulatory innovation

Regulation should be adaptive and predictable, built on foundations for learning, coordination and trust. This is design thinking applied to regulation: a framework prototyped, tested against real cases and iterated, not specified once and left. For increasingly cross-cutting, platform-based biotechnology, that shift is a precondition for the framework working at all. It has three components. First, a risk-based principle: unless assessment scales to risk rather than process or category, no downstream flexibility delivers adaptability. Second, mechanisms for regulatory learning, of which sandboxes are one controlled instrument and not the only one, by which regulators learn what flexible frameworks should contain. Third, mechanisms to embed that learning, through pre-submission advice, sandbox-to-dossier data flow, and a standing body with statutory review clauses, so what is learned changes binding requirements.

Across all four priorities the proposed logic is to fund the shared, pre-competitive capabilities that no single firm will build alone, so that public investment compounds rather than dissipating.

Point 1 – The bio-stack as a conceptual framework and support for cross-cutting capabilities

1.1 The bio-stack – Europe’s missing industrial framework

As with value chains such as semiconductors, industrial biotechnology is a stack of distinct, interdependent layers, and value is captured only by holding the whole bio-stack, not isolated parts. Four layers of the bio-stack are physical, with material passing through them in sequence: 

(i) feedstocks European biological systems can viably use; 

(ii) the engineered organisms; 

(iii) the fermentation and processing that scale them; 

(iv) the analytical methods that verify output. 

A fifth, the bio-digital layer, runs through all four: it specifies the organism before it is built, informs which feedstocks are worth pursuing, predicts behaviour at scale, and interprets the analytics, feeding optimised parameters back into the next design cycle.

Every layer of the bio-stack needs its own science, infrastructure, and funding. The layers depend on one another, so public investment only compounds when it is planned across the whole stack rather than spent project by project. Backing one layer without the others bottlenecks the rest, and the public return never arrives. Strategic autonomy in critical sectors, and a competitive biomanufacturing ecosystem more broadly, depends on coordinated capability across the whole stack.

1.2 Cross-cutting bio-digital capabilities as strategic project priorities

What most amplifies the bio-stack is not more product-specific projects but cross-cutting, platform-level capabilities. These are being built, but at startup scale, dispersed, and with no mechanism to take them up across the Union. The highest leverage sits at the interface of biology and computation, in AI/ML, foundation models, and structured data generation applied to design, prediction and control, because the capability is reusable across feedstocks, organisms, and processes rather than tied to one product. Some examples, mapped to the bio-stack layers they serve:

• Foundation models for engineering biology guiding target selection, pathway design and strain optimisation end-to-end (enzyme → pathway → strain → process).
• Computational protein and enzyme design coupling protein language models with biophysical simulation for novel, manufacturing-optimised biochemistry.
• Predictive scale-up digital twins modelling how a process behaves across scale transitions, from lab to production and production to industrial vessels, before capital is committed to the next scale.
• AI process-control layers that turn standard bioreactors into self-optimising systems, adjusting feeding, aeration and pH against real-time signals.
• Integrated bioprocessing systems unifying upstream, downstream and analytics in a single data environment, collapsing the siloed instruments that fragment the design space.

The Call for Evidence frames investment predictability as transparent market-pull timelines and access to funding. It overlooks the technical side of predictability. Historically, a common cause of scale-up failure has been titre, rate and yield (TRY) not holding at industrial scale: organisms that perform in a two-litre flask behave differently in a 100,000-litre fermenter, and can lose engineered properties over a production run as selection works against them (Rugbjerg et al. 2018). For example, the company KiOR projected 67 gallons of fuel per tonne of biomass and delivered far less; similarly, Amyris reached 15% lab-scale yield and could not replicate it in production (First Bight Ventures, 2025; Stumpf, 2026). Reduction of the TRY at industrial scale is a key driver of the lab-to-commercial cost gap and the constraint that cuts across every other part of the system: poor fermentation performance forces larger reactors, drives up downstream costs, and amplifies feedstock waste, compounding cost across the value chain (Santos-Navarro et al. 2021; Banerjee et al. 2020).

Scale-up carries its own science, not just engineering: these failures surface only at industrial scale, spanning strain robustness, genetic stability and downstream process behaviour. Predictive tools are what predict reliable behaviour of engineered microorganisms at scale, work companies do not naturally fund given diffuse returns and long timelines, and precisely the case for public investment.

TRY is not the only such constraint. Feedstock economics can cap a process before fermentation matters, and first-of-a-kind plants carry capital costs that depreciated petrochemical facilities do not (Kim et al. 2026). These are preconditions that finance and market-pull do not address. 

Predictive capability across the bio-stack is therefore not only a lever on whether bio-based production reaches cost-competitiveness in the first place, but a precondition for investment predictability and investor attraction.

These capabilities need mechanisms for implementation at scale: First, license capabilities into shared infrastructures so their operational data accrues as shared EU intelligence rather than staying locked in individual firms. Second, maintain a Union-level database of these tools and platforms, priced fairly for SME use, building on the Act’s existing requirement of “open, non-discriminatory, transparent and criteria-based access… with particular attention to SMEs, start-ups and scale-ups.” [COM(2025) 1022, recital 16] Together these turn dispersed startup tools into a Union-scale capability base, and stop Europe rebuilding the same playbook firm by firm.

1.3 A systems-level predictive twins as de-risking tools for scale-up investment

Where Point 1.2 funds prediction layer by layer, cost-competitiveness is dependent on many variables at once: strain performance, fermentation behaviour, downstream recovery, feedstock cost and variability, market price, and the economics and life-cycle impacts that follow. A bioprocess extends well beyond the bioreactor, and it is the integration of upstream and downstream steps that ultimately determines viability; optimising any one in isolation does not produce a viable process (Kim et al. 2026; Espinel-Ríos et al. 2025).

Leading researchers now advocate a systems-level framework integrating all of them: AI coupled with systems metabolic engineering, linking strain design, process operation and techno-economic (TEA) and life-cycle (LCA) evaluation in one loop, so that minimum product selling price, net present value and environmental metrics become design criteria rather than retrospective checks. Digital twins are the operational form, used to monitor, optimise and control a process from design through to production, and uncertainty-aware models flag how confident each prediction is, so the output can support risk-informed investment (Kim et al. 2026; Espinel-Ríos et al. 2025; Shariatifar et al. 2025; Matei 2025).

This is a de-risking instrument that would help prioritise the commitment of public and private funding at scale. The bioeconomy is hard to model and predict, with too many interacting variables to optimise by hand; only systems-level thinking of this kind can identify which processes are actually viable (Kim et al., 2026).

Point 2  – Strategic scale-up support: building the infrastructure

2.1 Biotech Act II should learn from the CHIPS Acts

Both the US CHIPS and Science Act and the EU Chips Act made the same architectural choice: the bulk of public funding went not to individual firms but to shared capability no single firm could build alone: pilot lines bridging laboratory and fabrication, the data accumulating across them, and the workforce running them. The EU’s four pilot lines were each mandated to serve start-ups and SMEs. The same bottleneck constrains biotech, and the same logic should structure where Biotech Act II concentrates public investment: across three layers: physical, digital, and knowledge infrastructure.

2.2 How to structure it: physical, digital, and knowledge layers

Physical infrastructure. Pilot and demonstration capacity, roughly 100 to 5,000 litres for process validation and a harder jump to 5,000 to 25,000 litres for industrial scale, is too expensive and too intermittently used for any single firm to own, so firms under-invest and the capacity is never built. The US has committed over $200M to shared pilot infrastructure at exactly this scale; Europe’s CBE Joint Undertaking targets scaleup but one project at a time. Shared not-for-profit plants are the most direct route, and Bio Base Europe Pilot Plant in Ghent has proven the model since 2008. But political commitment can be fragile: The UK’s Vaccine Manufacturing and Innovation Centre, conceived as a non-profit national asset and built with around £200m of government funding, was sold to a private US manufacturer before it became operational when political priorities shifted. What Biotech Act II needs is a coordinated network of pilot plants designed against each failure mode: durable political and financial commitment, governance that resists incumbent drift, and capacity that keeps pace with the frontier. 

Intelligent digital infrastructure. Every scale-up campaign produces operational data no single firm can afford to generate alone. Pooled across a network rather than siloed, it becomes a federated bioprocess intelligence that sharpens with every campaign, giving SMEs access to intelligence that otherwise only larger firms can build. The case is sharpest at the frontier: AI-designed biologics, engineered organisms in long production runs, and non-traditional hosts lack the operational playbook of established processes, so for them the data trail is the playbook. Without infrastructure to capture and share it, every firm rebuilds it from scratch. The US National Semiconductor Technology Center embeds exactly this layer alongside its pilot lines; Biotech Act II should build the equivalent.

Knowledge infrastructure. Shared scale-up plants and the data they generate are one half of successful scale-up. The other is the trained operators, engineers, and scientists who resolve the problems that arise when laboratory processes meet industrial scale. Ireland’s National Institute for Bioprocessing Research and Training (NIBRT), established in 2011 with €57m from Ireland’s inward-investment agency, is the clearest working example: a biopharma institute running as a four-university consortium (UCD, Trinity, DCU, IT Sligo) that is not only a training centre but an applied R&D test bed, so cohorts learn to develop new processes, not just operate established ones. It trains over 4,000 operators and engineers per year. 

The returns are disproportionate: Ireland is the world’s third-largest pharmaceutical exporter, with €116bn in annual export revenue and 85,000 sector jobs, and NIBRT is consistently named as the reason multinationals chose Ireland over comparable jurisdictions. Its curriculum, facility design, and certification regime have been licensed to the Jefferson Institute for Bioprocessing in Philadelphia ($10m partnership) and to K-NIBRT in Songdo (2,000 Korean trainees per year), each deal generating revenue and embedding NIBRT-pioneered processes as the operating standard abroad. 

No equivalent exists at European level for industrial biomanufacturing, where training happens ad hoc, inside individual firms, on equipment that does not transfer. Europe hosts the world’s densest concentration of research-intensive universities, so the consortium model that delivered NIBRT is replicable with regional specialisation across Member States. What is missing is the institutional design that converts academic capacity into industrial capability.

Point 3 – Common data standards for bioprocess and biological data

AI, digital twins, and shared datasets can improve biomanufacturing only if data from one lab or reactor can be read, compared, and trusted elsewhere. Today it cannot: there is no common way to describe and exchange it, so results are not comparable across sites, instruments, or scales (Asin-Garcia et al., 2025; Müller et al., 2026). The problem is sharper in biology than in other industries, because living systems are variable and identical strains can yield different results, so without common reference standards the digital twins and AI controls now attracting major investment cannot be reliably calibrated (Müller et al., 2026).

A standard generates value only when adopted sector-wide, so no single firm can establish one, and an incumbent-owned standard would fragment the market and entrench enclosure, with a few firms accumulating proprietary data while SMEs are excluded (Asin-Garcia et al., 2025; Müller et al., 2026). This is a coordinating function only an actor with a Union-wide mandate can perform.

The EU-funded BIOINDUSTRY 4.0 project (2023–2026) has developed interoperable data and metadata standards, alongside real-time monitoring devices and digital twins, and a common framework linking microbial-strain databases across European research infrastructures (CORDIS, GA 101094287); feasibility is demonstrated. The main barriers are structural, not technical: the work fragments into isolated efforts into ‘digital islands’ that end with each grant; firms will not contribute data without a clear share of the benefits; and shared data infrastructure raises security and IP-protection concerns that must be designed in, not bolted on (Müller et al., 2026).

Point 4 – Regulatory innovation: mechanisms for adaptive regulation

Biotechnology needs regulatory processes, tools, and institutions that are agile, able to realign with scientific and technological progress as it happens rather than lock in the science of the moment and wait for the next political opening to update. The OECD makes the case directly for the sector: the goal is regulatory systems that are more adaptive and predictable, through agile governance whose success depends not on individual tools but on the broader foundations that enable learning, coordination, and trust-building across the regulatory system (OECD, 2025).

This is, at root, design thinking applied to regulation: treating a framework as flexible and something to be prototyped, tested against real cases, and iterated rather than specified once and left. The EU Policy Lab already advances this mindset in EU policymaking, looking beyond political cycles and turning future-focused thinking into proactive policy experiments rather than one-size-fits-all solutions. What has not happened is carrying that same approach into the design of regulatory frameworks, the legal instruments, data requirements, and authorisation pathways themselves. For biotechnology, where products are increasingly cross-cutting and platform-based, making that leap is not a refinement but a precondition for the framework working at all, and the Biotech Act is the instrument to do it. 

Adaptive regulation is not a single instrument but three components working together, and this section is structured around them:

• a risk-based assessment principle (4.1), the conceptual base: unless assessment scales to risk rather than to process or category, no amount of procedural flexibility downstream will deliver adaptability;
• mechanisms for regulatory learning (4.2–4.3), the means by which regulators discover what flexible frameworks should actually contain, of which regulatory sandboxes are one controlled instrument, not the only one;
• mechanisms to embed that learning (4.4–4.7), so that what is learned changes binding requirements rather than dissipating.

The ultimate goal tying them together is modular regulation: frameworks composed of reusable, independently updatable components, a platform assessment, a product assessment, a data-requirements module, that different authorities can recombine and fit to different end goals, rather than a monolithic dossier rebuilt for every product. Achieving it requires regulators to learn together and, explicitly, to re-design as they go.

References