Biology's Fourth Act: Inside AI BioDesign, the $95 Million Bet to Engineer Life Beyond Evolution

Charles Darwin called it “endless forms most beautiful” — the spectacular diversity of life that evolution has produced over billions of years. But on September 3, 2026, three of Seattle’s scientific powerhouses made a bet that everything nature has ever built is only a rounding error of what biology could become. Their new joint venture, AI BioDesign, is a $95 million attempt to map and then exploit that uncharted design space — not with test tubes and guesswork, but with AI models proposing designs and robotic-scale experiments grading them.

What was announced

AI BioDesign is a collaborative research accelerator between the Allen Institute, University of Washington Medicine, and Fred Hutch Cancer Center, supported by $95 million from the Fund for Science and Technology (FFST), the private foundation in the Paul G. Allen philanthropic ecosystem that also bankrolls the Allen Institute itself.

The mission statement is deceptively simple: build open AI models, datasets, assays, and tools that let scientists create biological solutions for human health, environmental challenges, and advanced technologies. The targets named in the announcement range from the medical to the almost science-fictional — new drugs for cancer and neurodegeneration, enzymes that break down plastics in the ocean, and biological computers that consume a fraction of the power of silicon chips.

What makes this different from yet another well-funded academic center is its architecture. AI BioDesign is explicitly built as a continuous learning platform: AI models propose new biological designs, scientists build and test those designs at scale, and the results feed back into the models so that each cycle becomes more accurate. It is the classic design-build-measure-learn loop of engineering, applied to living systems — with machine learning sitting permanently in the loop.

The people behind it

The scientific leadership reads like a shortlist of modern computational biology. David Baker, lead scientific director of AI BioDesign, won the 2024 Nobel Prize in Chemistry for computational protein design and directs the UW Medicine Institute for Protein Design. Jay Shendure, co-lead scientific director, is the force behind the UW Medicine Brotman Baty Institute for Precision Medicine and the Seattle Hub for Synthetic Biology, and a Howard Hughes Medical Institute investigator. Principal investigators include Sanjay Srivatsan (Fred Hutch) and Sud Pinglay, with Rui Costa serving as president and CEO of the Allen Institute.

Baker framed the launch as an inflection point: “For the first time, the speed of AI is beginning to match the experimental power of synthetic biology. That changes the question from ‘What has nature already made?’ to ‘What else is possible, and how can we test it?’ AI BioDesign can help turn that vast unknown into models that can help us solve some of humanity’s hardest problems.”

Shendure went further, casting the project as the next chapter in the history of engineering: “Over the last two centuries, engineering has transformed the world at least three times: the Industrial Revolution, electrification and mechanization, and the digital revolution. We believe engineering’s fourth act lies at the intersection of AI and biology. Biology is code that builds: DNA carries digital instructions, and cells turn those instructions into physical systems with extraordinary precision.”

Why the premise matters

The core insight is about search space. Natural biodiversity represents a vanishingly small sample of the trillions of possible DNA sequences — evolution explored only what random mutation and selection happened to reach. Modern AI protein-design tools (the lineage Baker helped create with RoseTTAFold and related systems) have already shown that models trained on natural proteins can generalize beyond nature, proposing stable, functional molecules that have never existed.

But there is a bottleneck: those models are still trained overwhelmingly on data from what nature has made. The counterfactual universe of designed-but-untested sequences produces almost no experimental feedback. AI BioDesign’s answer is to become its own data factory — generating perturbations and designed sequences specifically chosen to teach the models where their assumptions break, then testing thousands of designs simultaneously with multiplex assays.

That last point is the quiet differentiator. The announcement is careful to position the effort as “distinct yet complementary” to the global wave of foundation models, virtual-cell systems, and lab-in-the-loop platforms. Its distinction: multiple modular models built from tractable biological problems, new data from designed sequences and perturbations, multiplex experiments that test many designs at once — and, unusually for a field drifting toward proprietary moats, open sharing of everything: models, datasets, assays, reagents, and benchmarks.

The Seattle stack

The collaboration maps cleanly onto each institution’s strengths. The Allen Institute brings two decades of experience running big-science, open-science platforms — the Allen Cell Explorer, Allen Brain Observatory, and immunology references that entire fields build on. UW Medicine contributes the Institute for Protein Design and the Brotman Baty Institute’s genome-engineering depth. Fred Hutch adds cellular systems, genomics, and the translational muscle of a top cancer center — Srivatsan’s stated ambition is genomic datasets scaled up enough that AI can find patterns “to inspire solutions to biological problems, such as designing cells that can remove cancer from the body.”

Rui Costa emphasized the model-in-the-loop philosophy: “AI BioDesign combines this approach with AI models to guide which data we generate next, so experiments and models improve together in a continuous cycle of learning and testing.”

The honest caveats

Three challenges are worth watching. First, the design-build-measure-learn loop only compounds if the experimental arm keeps pace with the modeling arm — multiplex assays are powerful but not free, and $95 million over a multi-institution program does not go as far as it would inside a single company. Second, “open everything” is a genuine competitive strategy against proprietary labs like EvolutionaryScale or Big Pharma’s internal engines, but it also means the value capture — and the safety questions around increasingly capable biological design tools — land on the whole field rather than one balance sheet. Expect dual-use screening to become a live topic as the shared reagents and design models mature. Third, biological computers and ocean-cleaning enzymes are horizon goals; the nearer-term deliverables will be less glamorous datasets, benchmarks, and assay protocols.

Still, the structure is right. Rather than betting on one monolithic foundation model, AI BioDesign is betting that biology rewards many grounded models, relentless experimental feedback, and open iteration — the same formula that made the protein-design field leap forward in the first place.

For a field where the last two years brought AlphaFold-era structure prediction and AI-designed drugs entering clinics, AI BioDesign is the next logical escalation: not predicting what biology made, but systematically learning what biology could make. Engineering’s fourth act, as Shendure would have it, now has a $95 million down payment.