FREE MUSK MILLIONAIRES MASTERCLASS

The economics of plastic waste just changed. Enzymatic depolymerization of PET, the polymer in beverage bottles, food trays, and polyester garments, is moving from lab to industrial demonstration. But the specific cost figures in play are still the debate: the article’s €63 per ton enzymatic-cost claim, the €250 per ton conventional-recycling claim, and therefore the implied €187 per ton gap are not supported by the same kind of public, apples-to-apples disclosures. Treat the cost advantage as directional until a like-for-like techno-economic assessment is cited side-by-side.

What closed the performance gap was not incremental chemistry. It was protein design powered by machine learning.

The Design Problem AI Solved First

Natural enzymes were never built for synthetic polymers. PET persists for decades in the environment because nothing in the microbial world evolved specifically to dismantle it at industrial speed. Nature operates on evolutionary timescales, and facing a biosphere saturated with synthetic polymers, computational biology stopped waiting for evolution and began inventing solutions from scratch.

Scientists at Murdoch University’s Bioplastics Innovation Hub are combining machine learning and biochemistry to filter through large enzyme candidate sets, using pipelines trained on previously characterized enzymes to predict how protein structures interact with and degrade target pollutants. The scale of that search is the point: Murdoch itself describes “millions upon millions” of potentially relevant enzymes sitting in global databases, largely untested. Machine learning converts that raw inventory into a ranked shortlist for lab validation.

On the engineering side, published research in The FEBS Journal engineered the current gold-standard PET-hydrolyzing enzyme, LCC-ICCG, into a mutant called LCC-ICCG-C09. The paper reports a 3.5°C increase in melting temperature versus LCC-ICCG and, under optimal conditions, roughly two-fold higher efficiency hydrolyzing amorphous PET into terephthalic acid. Two-fold efficiency on a platform already operating at industrial demonstration scale is meaningful compression of the cost curve.

From Bench to Industrial Floor

Carbios, listed on Euronext Growth Paris, is the furthest along commercially. On July 21, 2026, the company said it had completed 100 production batches at its Clermont-Ferrand industrial demonstration plant and expanded its licensing offering to the textile market. In the same announcement, Carbios said that since 2025 it has been extending its enzymatic biorecycling process to textile waste rich in PET polyester fibers, citing global textile consumption of roughly 67 million tonnes per year and describing that as nearly twice the size of the PET packaging market. That expansion reframes the addressable volume entirely.

Research published in May 2026 highlights the promise of designing enzyme systems capable of processing more heterogeneous waste mixtures, potentially reducing the need for rigorous waste-stream sorting. Sorting is often where economics break down in practice. Enzymes that tolerate contamination and blended materials remove that barrier.

The Startup Layer

Berkeley-based Rhea’s Factory is attacking the same problem from a purer AI-first position. In its June 8, 2026 announcement, the company said its process runs at mild temperatures with low energy input and returns polyester to virgin-quality building blocks at greater than 99% purity, while also handling contaminated and blended materials that make much textile waste hard to recover. In June 2026, Rhea’s Factory was named one of ten winners of the H&M Foundation’s Global Change Award, validating its early-stage model before it has demonstrated commercial scale.

What Investors Should Watch

• Carbios licensing velocity: Its July 2026 textile expansion signals a shift toward a high-margin licensing model that does not require Carbios to build every plant itself. Watch how quickly signed licenses convert to operating royalties.
• Enzyme immobilization progress: Reusable, immobilized enzymes are the lever that compresses per-ton cost further. Lab results already show significant degradation rate improvements over free enzyme equivalents.
• Regulatory tailwinds: EU recycled-content mandates are tightening. Per the OECD’s Global Plastics Outlook, plastic waste is projected to almost triple by 2060, and recycling is projected to remain below one-fifth even then; OECD also estimates recycling was about 9% in 2019. Mandatory minimum recycled-content rules reward whoever can supply virgin-quality monomers at scale.
• Carbios Longlaville timeline: The company has described its Longlaville project as targeting production start in H1 2028, conditional on financing, and in August 2026 said the previously announced target for financial close by September 30, 2026 would not be met, which would push the timeline. Any acceleration or slippage can move the stock.

Bottom Line

The question for this market is no longer whether AI-designed enzymes can dissolve industrial plastic. The published science and the demonstration-scale milestones support that they can. The question is who captures the monomer supply chain that this technology unlocks. Carbios holds the longest operating track record at industrial demonstration scale. Rhea’s Factory represents the hypothesis that a pure machine-learning enzyme platform can close the gap faster than a decade of conventional protein engineering did. Both are worth watching. The cost advantage over conventional recycling is the number that will eventually draw larger capital into this space, but investors should demand like-for-like comparisons before underwriting a precise per-ton spread.