Generate Biomedicines: When Biology Becomes Engineering

The Bet

In 2024, the Nobel Prize in Chemistry went to three people for the same idea from two directions. Half went to David Baker, for designing proteins that had never existed in nature. The other half went to Demis Hassabis and John Jumper of Google DeepMind, whose AlphaFold can predict how a protein folds - the step that used to take a graduate student years - in minutes.

One half learned to read the language of proteins. The other half started writing in it.

That second half is the entire thesis behind a company that went public three months ago. As NVIDIA's Jensen Huang likes to put it:

"For the very first time in human history, biology has the opportunity to be engineering, not science. When something becomes engineering, not science, it becomes... exponentially improving - it can compound on the benefits of previous years."

Generate Biomedicines - ticker GENB - is the clearest pure-play bet that he's right. It uses generative AI to design new protein drugs the way a developer writes code: name the target, let the model propose molecules, then build and test the best ones. In February 2026 it priced an IPO at $16 a share, raised about $369 million net, and gave public investors their first look at a company NVIDIA had quietly backed while it was still private.

We used Barebone to pull the prospectus, the first-quarter numbers, the two big pharma contracts, and the trial registry into one place, then checked the story against the filings. Here's what survived: the stock closed at $13.02 on May 20 - roughly 19% below its IPO price. Two of the largest drugmakers on earth have signed deals worth up to $2.9 billion. The company burns about $80 million a quarter and has never sold a drug. And its lead molecule is already in Phase 3.

What Generate Actually Does

Most drug discovery is still a search problem. You screen enormous libraries of existing molecules, or you tweak ones that already work, and you hope something binds. Generate flips it into a design problem. Its models are trained on the physics and sequence patterns of proteins, and they generate brand-new ones - de novo, meaning from scratch rather than copied from nature - aimed at a chosen target. It is the same generative leap that turned language models from search engines into writers. Baker's Nobel was the proof of concept; Generate is trying to make it a production line.

The speed is the tell. Generate says its first clinical antibody, GB-0669 - an AI-designed molecule built to neutralize SARS-CoV-2 - went from computer design to a human trial in 17 months. The industry usually spends five to seven years in the lab before a molecule reaches a person. Its lead program went from design to Phase 3 in about four years, a timeline CEO Mike Nally calls a demonstration of "programmable biology."

That is the bull case in one word: repeatability. If you can shorten the part of drug development that is slow, expensive, and random - the discovery - you change the economics of an industry where a new drug takes a decade, costs billions, and fails roughly 90% of the time.

The Validation: Two Pharma Giants Paid Up

You don't have to take Generate's word that the platform works. Two of the largest drugmakers on earth have rented it.

Partner	Signed	Up-to value	Structure
Amgen	Jan 2022	up to $1.9B+	5 programs (a 6th added later, +$370M); also a Series C equity investor
Novartis	Sep 2024	up to $1.0B+	$65M upfront, including $15M equity; tiered royalties

But read the structure before you read the headline. These are biobucks - industry slang for deal values that are mostly success-contingent milestones, paid only if molecules actually advance through trials and reach the market. The upfronts are real money. The billions are options.

You can see the gap in Generate's own income statement. Across the last twelve months, all of its collaboration revenue - research funding plus the milestones it has actually hit - added up to about $30 million. That is what an "up to $2.9 billion" platform has converted into cash so far.

The deals validate the science. They don't yet pay for it. That distinction is the whole investment.

The Pipeline Rests on One Asset

Generate has four programs in or entering the clinic, but the weight is wildly uneven.

Program	Target / type	Stage	Note
GB-0895	Anti-TSLP antibody	Phase 3 (severe asthma)	SOLAIRIA-1 & -2, ~1,600 patients, 40+ countries
GB-0669	Anti - SARS-CoV-2 antibody	Phase 1 complete	First AI-designed antibody Generate dosed in humans
GB-4362	MMAE neutralizer	Phase 1 (2026)	FDA Fast Track; first patient expected mid-2026
GB-5267	CAR-T (with Roswell Park)	Phase 1 (2026)	Ovarian cancer; first patient expected H2 2026

GB-0895 is the story. It is a long-acting antibody engineered with AI to block TSLP, a protein that drives airway inflammation, and it is dosed once every six months - a Phase 1 study of 96 patients showed a roughly 89-day half-life, which is what makes twice-a-year dosing plausible. In December 2025 Generate launched two global Phase 3 trials, SOLAIRIA-1 and SOLAIRIA-2, enrolling about 1,600 adults and adolescents with severe asthma. The company calls them the first global Phase 3 studies of a long-acting anti-TSLP antibody; it is also one of the first antibodies designed with generative AI to reach Phase 3 at all.

Everything behind it is just entering humans. So the platform thesis, for now, has a single late-stage proof point - and it is in a crowded respiratory market where AstraZeneca and Amgen's Tezspire already sells against the same target.

The Money

A clinical-stage biotech is a balance sheet with a science project attached. Here is Generate's, as of March 31, 2026:

Metric	Figure
Cash & marketable securities	$516.6M
Net IPO proceeds (Feb 2026)	$369.3M
Q1 2026 net loss	$61.7M (vs. $44.3M a year earlier)
Q1 2026 R&D	$57.8M
Operating cash burn (Q1)	~$80M
Stated runway	into the first half of 2028

The IPO did its job: it refilled the tank from $221.5 million at year-end to more than half a billion. At $13.02 a share and about 128 million shares out, the market caps Generate near $1.7 billion. Strip out the cash and the market is paying roughly $1.1 billion for the platform and the pipeline - a price that exists entirely on the promise of trials that have not read out.

Revenue went down year over year, from $8.8 million to $7.2 million, while the loss grew. That is normal for a company moving a drug into Phase 3 - late-stage trials are the most expensive thing in biotech - but it means the clock is real. Roughly $80 million a quarter against $516 million buys about six quarters before the next raise.

The NVIDIA Question

The detail that travels fastest is NVIDIA's involvement, so it's worth being precise about what is actually known. NVentures, NVIDIA's venture arm, joined Generate's $273 million Series C in 2023, alongside Amgen, in a round led by Flagship Pioneering - the firm that also founded Moderna. That is the bet: a private, strategic stake taken three years ago, made visible only because the company is now public.

It is partly self-interested. NVIDIA sells the picks and shovels of digital biology - its BioNeMo platform runs the kind of models Generate builds - so equity in a marquee customer is strategy as much as conviction. Read it as a vote, not a verdict.

Because NVIDIA has placed this bet before. In July 2023 it put $50 million into Recursion (RXRX), another AI-drug-discovery company; the stock doubled on the announcement, then gave most of it back. Recursion booked about $75 million of revenue in 2025 against a net loss north of $640 million, trades in the low single digits today - far below its 2021 IPO - and the analysts who cover it sit at Hold. NVIDIA's logo got investors paid for a day. It did not make the underlying biology work faster.

The Uncomfortable Truth

Here is the part the pitch skips. Generate is a clinical-stage biotech with no approved product, and that is the most binary asset class in public markets. The company is valued on a future, not a business.

No drug designed by generative AI has ever been approved by the FDA. GB-0895 would be among the first even to ask. Phase 3 is where most respiratory antibodies that looked clean in Phase 1 quietly fail, and the trial is a binary event: if SOLAIRIA misses its endpoint, the platform story does not cushion the fall - the stock reprices toward its cash and its contracts, and a $1.1 billion platform premium can evaporate in a single press release.

The partnerships are validation, but they are also a reminder. "Up to $2.9 billion" has produced about $30 million of realized revenue. The real signal of platform value isn't signing a deal; it's a partner paying a milestone - and those are still mostly ahead.

And the category itself has overpromised before. AI drug discovery has been "five years away" for a decade; Recursion is the cautionary tale, not the outlier. The field is full of platforms that produced elegant molecules and no approvals. Generate may be better. It has not yet proven it where it counts, which is in a pivotal trial.

What This Means

You are not buying an asthma drug at $13. You're buying a wager that protein design has become repeatable - that the Nobel-winning ability to write a new protein can be turned into a pipeline, and that pharma's willingness to pay for it eventually shows up as cash rather than press releases.

The signals that decide it are concrete. The SOLAIRIA Phase 3 readouts are the binary. Whether Amgen or Novartis exercise more options - and pay milestones - is the validation that matters. The cash runway and the timing of the next raise set the clock. And above all the category question: whether any AI-designed drug, Generate's or anyone's, actually reaches approval.

At $13.02, below where it came public, Morgan Stanley - one of the few banks covering a three-month-old stock - rates GENB Overweight with a $22 target. That is a real call, but it rests on the same trials everyone else is waiting on.

The Nobel committee already settled whether AI can design a protein. The open question is whether designing one and selling one are the same skill. Generate is the first company big enough, and public enough, to find out in front of everyone.

---

Data: Barebone | Sources: Generate Biomedicines IPO pricing release and Q1 2026 results, GB-0895 Phase 3 announcement (Dec 1, 2025), Amgen (2022) and Novartis (2024) collaboration releases, Generate $273M Series C (Sept 2023), Recursion - NVIDIA collaboration (July 2023), 2024 Nobel Prize in Chemistry | Prices as of May 20, 2026 close