LanzaTech ferments captured industrial off-gases and waste carbon into ethanol and jet fuel, sidestepping the feedstock cost that sinks sugar-based fermentation, while spinning its LanzaX platform into a focused JV. Cheap or negative-cost feedstock is a genuine economic lever, but gas fermentation is finicky, capital-heavy, and still depends on carbon pricing and offtake that remain politically fragile.

Antheia brews high-value pharmaceutical ingredients that traditionally require agricultural extraction, and Geno targets specialty bio-based intermediates rather than bulk chemicals. Starting where the incumbent price is high buys margin to survive on — the explicit lesson of the graveyard — but the addressable volume is smaller, and each product is a fresh scale-up gamble.

Solugen's enzyme-and-metal Bioforge aims to hit petrochemical cost parity on specialty chemicals from plant sugars, treating the manufacturing process itself as the innovation. Cost parity at scale is the whole game, but 'cost-competitive at scale' has been the industry's promise — and epitaph — for fifteen years, and tolling capacity to prove it remains scarce.

Ginkgo is betting the company on Nebula, which it calls the world's largest autonomous lab, running customer science around the clock and targeting a doubling in 2026, while Emerald Cloud Lab and Culture Biosciences offer cloud-accessible automated experimentation. Removing humans from the loop is the throughput unlock, but the capital and reliability engineering are brutal — and Ginkgo's own foundry revenue never scaled on the earlier version of this promise.

DNA Script's benchtop enzymatic printer and Twist's silicon-chip synthesis attack the 'build' bottleneck directly, shortening the wait between design and test. Faster builds compress the cycle, but the test-and-learn steps and the fundamental failure rate of designs remain the binding constraints downstream.

Foundation models like Arc's Evo 2 and Profluent's protein language models aim to make each design more likely to work, reducing how many expensive wet-lab cycles the loop needs. Stronger priors are leverage on the whole system, but they are trained on the very experimental data the slow loop produces — a chicken-and-egg dependency the sector hasn't fully broken.

Mammoth's NanoCas is roughly one-third the size of Cas9 — small enough to package with a promoter into a single AAV — and hit ~60% liver editing of PCSK9 while being aimed at tissues beyond the liver; Arbor and Scribe engineer similarly compact editors. Shrinking the payload to fit the delivery vehicle is a powerful lever, but AAV's own immunogenicity and tissue tropism limits remain.

Dyno Therapeutics uses AI to redesign AAV capsids for better targeting and immune evasion, licensing improved vectors to major pharma. Purpose-designing the delivery vehicle rather than the cargo attacks the bottleneck head-on, but a capsid that works in mice often disappoints in humans, and validation is slow and expensive.

Beam's risto-cel and Caribou's allogeneic CAR-T edit cells outside the body and reinfuse them, avoiding the in-body delivery problem altogether. It works today for blood and immune indications, but requires harsh conditioning regimens (a patient died in Beam's sickle-cell study), is expensive and bespoke, and simply cannot reach solid tissues.

Arc Institute's Evo 2 — 40B parameters trained on 9T+ nucleotides across 100,000+ species — can flag disease-causing mutations and design bacterial-length genomes, released open-source to seed an 'app store for biology.' Pretraining at genome scale is the most direct bet, but predicting that a sequence exists is far easier than predicting it functions, folds, and is safe in a human.

Profluent established that scaling laws apply to protein design and used its models to generate the first AI-created CRISPR system, monetized in a $2.25B-ceiling Lilly deal. The LLM playbook is real for proteins, but functional labels are scarce relative to internet corpora, and immunogenicity and manufacturability are downstream failure modes models struggle to see.

Basecamp Research is assembling one of the world's largest novel-biodiversity sequence databases, and Ginkgo's Datapoints sells autonomously generated experimental data, on the bet that data — not model architecture — is the constraint. Owning unique data is a durable edge, but generating enough high-quality functional labels at reasonable cost is exactly the wet-lab bottleneck the field hasn't cracked.

IBBIS operates the Common Mechanism — free, open-source screening software (commec) meant as a global baseline for nucleic-acid synthesis screening. A shared open standard lowers the cost of doing the right thing, but adoption depends on trust and resources, and academic and benchtop settings remain the hardest to reach.

Executive Order 14292 (May 2025) directed revision of the federal screening framework, and the Biosecurity Modernization and Innovation Act of 2026 (S. 3741) would require a Sequences-of-Concern list and cross-provider screening to catch split orders. Enforceable rules close the free-rider gap, but rulemaking runs on multi-year clocks, adds cost and liability to every order, and struggles to keep pace with distributed synthesis.

Proposed frameworks push screening down to short oligonucleotides and add cross-provider order matching to detect fragments ordered separately and assembled later. Catching the sub-200-bp and split-order loopholes is where real risk lives, but it demands information-sharing and safe-harbor structures across competing providers that don't yet fully exist.

Ansa Biotechnologies and Elegen push enzymatic and long-construct synthesis toward the lengths and fidelity that genome-scale assembly needs, and Twist extended complex gene synthesis to 7,000 bp. Longer, cleaner writes are prerequisite infrastructure, but assembling and delivering genome-scale constructs into a working cell is a further, unsolved leap.

Academic efforts like Synthetic Yeast (Sc2.0) and genome-recoding projects demonstrate that organisms can be substantially rewritten, and Constructive Bio pursues recoded organisms with expanded genetic codes. These prove the concept at the research frontier, but each is a heroic, bespoke effort far from a repeatable engineering discipline.

Tessera's Gene Writing uses engineered mobile genetic elements to write everything from single bases to whole genes, aiming to insert large sequences precisely. Programmable large-cargo insertion would unlock genome-scale edits, but efficiency, precision, and in vivo delivery of big payloads all remain open.