Synthetic Biology in 2026: Ginkgo's Collapse, Twist's Resilience, and the AI-Protein Pivot

The synthetic biology industry spent the late 2010s and early 2020s on a story that did not survive contact with the public markets. Foundries — large central facilities that would take customer-supplied designs and produce engineered organisms on a service-bureau model — were going to be the AWS of biology. Ginkgo Bioworks, the canonical example, went public via SPAC in 2021 at a roughly 17-billion-dollar valuation and promised exponential scaling. By 2024 the share price had collapsed by more than 95 percent, and through 2025 the company restructured around a narrower set of programs and partnerships. The foundry-of-everything model did not work.

What did work, increasingly, is the synthesis side (Twist Bioscience), the application-tied platforms (Cradle, Asimov), the diagnostic and therapeutic adjacencies (Adaptive Biotechnologies), and the convergence with AI protein design that Demis Hassabis won the 2024 Nobel Chemistry for. This post walks through where the synthetic biology field actually sits in 2026.

The Ginkgo collapse and what it taught the industry#

Ginkgo’s model was a horizontal foundry — design organisms for any customer in any application, take some equity or royalty in the output, and let the volume of the foundry drive the unit economics down over time. The model had two problems that became evident by 2023. First, the customer base never grew to the scale that would have made the foundry economics work; the long tail of small synthetic-biology programs was much smaller than the addressable market the company had projected. Second, the per-program work was much more bespoke than the AWS analogy implied. Each program required real biology judgement, real wet-lab tuning, and real iteration, which the foundry headcount and infrastructure model did not absorb gracefully.

The 2024 and 2025 restructuring at Ginkgo cut headcount, narrowed the program portfolio, and shifted positioning toward a smaller number of high-value partnerships rather than the long-tail volume model. The honest reading of where Ginkgo sits in 2026 is that it is a real but smaller company doing real biology, no longer the platform-of-everything the SPAC pitched.

The wider industry lesson is that the foundry-as-a-service model does not work as a standalone business, and that the synthetic biology companies that survived and scaled either sit upstream in the supply chain (synthesis) or downstream in specific applications (therapeutics, diagnostics, materials).

Twist Bioscience and the synthesis layer#

Twist Bioscience makes DNA. Synthesised oligos, longer genes, and the synthetic-DNA libraries that essentially every other player in the field needs as raw material. The business is unglamorous compared to the foundry pitch — selling consumables to research labs and biotech companies — but it scaled. Twist’s revenue grew through 2023, 2024, and into 2025 as the broader synthetic biology and protein-engineering fields generated more demand for synthetic DNA, and the company expanded into adjacent markets including antibody libraries and biopharma services.

The Twist story is the boring-but-working version of the synthetic biology thesis. The picks-and-shovels position survives even when the foundry model does not, because the underlying demand for synthetic DNA continues to grow regardless of which application companies are winning.

Codex DNA, Asimov, and the engineering-tools layer#

Codex DNA, now operating as Telesis Bio, sells the BioXp benchtop instrument that automates DNA assembly inside customer labs. The positioning is the opposite of the central foundry — bring the synthesis capability into the customer’s own lab so they do not have to outsource. Asimov, founded by MIT synthetic biologists, sells genetic-circuit design software and a cell-engineering platform aimed at the higher end of the bioproduction stack. Both companies represent the toolchain layer that the foundry model was supposed to subsume but didn’t.

The 2026 picture is that the synthetic-biology toolchain looks more like the chip-design EDA industry than like AWS — a small number of specialist tooling and consumables vendors, plus a larger number of application-specific players who buy the tooling and build their own programs in-house.

Cradle and the AI-protein-design convergence#

Cradle, founded in 2021 in Amsterdam and Zurich, sits at the intersection of synthetic biology and AI. The company sells protein-engineering software that uses generative ML models — descendants of the AlphaFold and RoseTTAFold work that won Hassabis and Baker the 2024 Nobel Chemistry Prize — to design candidate protein sequences for customer-specified properties. The wet-lab side, where the candidates get expressed and tested, is the part Cradle does itself or partners on. The business model is software-and-services rather than foundry, and the customer base spans biopharma, industrial enzymes, and food-tech.

Cradle is the cleanest example of the application-tied recovery of synthetic biology. The model is not “we will run a foundry for anyone”; it is “we will use AI to design proteins for specific properties, and we will partner with customers to take those proteins through the wet lab.” The unit economics work because the per-program value is high and the AI lift on the design phase is real.

Adaptive Biotechnologies and the diagnostic adjacency#

Adaptive Biotechnologies, the Seattle-based immunology platform, sits next to the synthetic biology field rather than in it. The company’s immune-sequencing platform — clonoSEQ for minimal-residual-disease testing in haematological cancers, and the broader immune-receptor sequencing portfolio — uses high-throughput sequencing of T-cell and B-cell receptors to characterise immune state. The business is largely a clinical-diagnostics one with reimbursement-driven revenue, and the synthetic-biology adjacency comes through the partnership work with Microsoft and others on the immune-system data infrastructure.

Adaptive is the example of the field where the underlying sequencing-and-synthesis capabilities of synthetic biology meet a real clinical use case with established reimbursement and a clear go-to-market.

The mRNA and AI-protein convergence#

The biggest structural shift in 2024 and 2025 is the convergence between the synthetic biology toolchain and the AI-protein-design work coming out of DeepMind (AlphaFold 3), the Baker lab at Washington (RoseTTAFold, RFdiffusion), Generate Biomedicines, Isomorphic Labs, and the wider academic community. The Nobel committee’s decision to give the 2024 Chemistry Prize jointly to David Baker, Demis Hassabis, and John Jumper was the formal recognition of how durable this shift is.

The practical effect on synthetic biology is that protein design — historically an enormously labour-intensive part of any new biology program — is increasingly an AI-driven step rather than a wet-lab iteration step. You feed a desired functional property and a set of constraints into a model, you get back a handful of candidate sequences ranked by predicted function, and the wet lab work focuses on validating the top candidates rather than searching the sequence space by hand. This compresses program timelines, reduces wet-lab cost per candidate, and shifts the competitive frontier toward whoever has the best models and the best wet-lab validation loop.

The mRNA side, downstream of the COVID-era vaccines, has continued to mature. Moderna, BioNTech, and a long tail of smaller players have moved mRNA into therapeutic categories beyond infectious disease — cancer vaccines, rare-disease replacement therapies, and immunomodulation. The synthetic-biology connection is the toolchain underneath, where Twist-style synthesis and AI-designed sequences feed into the mRNA-construct pipeline.

What this means for the broader tech ecosystem#

For AI and data teams not in the biotech industry, the synthetic biology story has two practical implications. First, the compute and data infrastructure for protein-design work — large multi-modal models, structure prediction, generative diffusion-style sampling — has converged on the same accelerator stacks and the same MLOps patterns that power frontier language models. The talent pool overlaps, the tooling overlaps, and the GPU demand overlaps. Second, the biology-data infrastructure problem — managing terabyte-per-program sequencing data, lineage tracking across hundreds of wet-lab variants, regulatory-compliant data residency for clinical work — is a real opportunity for the data-platform and data-engineering vendors that have built equivalent capability in other regulated industries.

Where pdpspectra fits#

Our AI and LLM integration practice and data-engineering practice build the model infrastructure and pipeline platforms that synthetic biology and AI-protein-design teams need for high-throughput candidate generation, wet-lab data ingestion, and lineage tracking.

Related reading: the AI clinical trials post, the multimodal AI 2026 post, and the open-source LLMs in production post.

---

Synthetic biology is not the platform business the SPAC era promised; it is something more useful and more grounded. Talk to our team about your AI-for-biology project.