Basel-based Atlas Bio announced $12.5 million in total funding on October 27, 2026, according to Tech.eu, to build an AI foundation model that runs in silico clinical trials and predicts patient-level drug efficacy and safety signals before Phase II. The company says roughly 70% of drugs fail at that stage, which is the problem its software is built to attack.
The round splits into a $10 million seed led by Creandum and a previously unannounced $2.5 million pre-seed led by First Momentum Ventures and LEA Partners. It landed in a week crowded with big checks, and our roundup of AI funding rounds led by Tenstorrent’s $1.5B shows how unusual a sub-$15M biotech-AI raise looks next to chip and infrastructure deals. The size is small. The target is not.
This analysis covers what Atlas Bio actually announced, which claims the available coverage supports, which ones it does not, and what the Atlas Bio funding says about where investors want AI to land in drug development. It also treats the product as a software problem, because a foundation model for clinical trials lives or dies on data pipelines, validation, and reproducibility.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
What Atlas Bio Announced on October 27
Atlas Bio operates from Basel, Switzerland, with a listed address at Sandgrubenstrasse 44, 4058 Basel. Its product is an AI foundation model for in silico clinical trials, meaning simulated studies that estimate how a drug candidate performs in patients before a real Phase II trial begins. According to the company’s announcement as reported by Tech.eu, the system aims to predict efficacy and safety signals at the level of individual patients rather than for a population average.
Katharina Wilhelm, a Partner at Creandum, joined Atlas Bio’s board as part of the financing. The company plans to use the money to expand its scientific and commercial teams, deepen validation of the platform, and onboard additional enterprise clients through 2026. That last phrase matters. Atlas Bio already describes itself as selling to enterprise customers, and the stated plan is to add more of them before the year ends, which gives the market a two-month window to see whether new names appear.
The Economic Times and The Guardian Nigeria News both picked up the story, and The Information also covered the raise. Across those outlets the core facts match. Where coverage diverges is in how hard it leans on the animal-testing angle, a point we separate out below.
The Numbers Behind the Atlas Bio Funding
Two rounds make up the total. The seed is the headline, but the pre-seed tells you something too: Atlas Bio raised $2.5 million quietly before it went public with the platform, and it only disclosed that money on October 27. Investors backed the company before anyone outside the cap table could judge the model.
| Item | Detail | Reported by |
|---|---|---|
| Total funding | $12.5 million | Tech.eu, Economic Times |
| Seed round | $10 million, led by Creandum | Tech.eu |
| Pre-seed round | $2.5 million, led by First Momentum Ventures and LEA Partners, previously unannounced | Tech.eu |
| Board change | Katharina Wilhelm (Partner, Creandum) joins | Tech.eu |
| Announcement date | October 27, 2026 | Tech.eu, Economic Times |
| Headquarters | Basel, Switzerland | Company listing |
| Stated use of funds | Scientific and commercial hiring, platform validation, more enterprise clients through 2026 | Tech.eu |
For scale, $12.5 million buys a small team and a modest compute budget. It does not buy a clinical program. Atlas Bio is a software company selling predictions, and the capital reflects that. A drug developer burning through a Phase II trial can spend more than $100 million, per the company’s own framing of the problem, so a model that earns its keep by avoiding even one doomed trial would pay for itself many times over. That arithmetic is the entire pitch.
Why the 70% Phase II Failure Rate Is the Target
Atlas Bio says about 70% of drugs fail in Phase II, and that the stage typically costs more than $100 million and takes two to three years, as reported by the Economic Times. Phase II is where a candidate first has to show it works in patients with the target disease, after Phase I has mostly checked basic safety. Plenty of compounds that look clean in early work stall here because the efficacy is not there.
That is why the Phase II failure rate became the anchor for the story. A drug that fails at this stage wastes the money already spent on discovery and Phase I. It also burns two to three years that a sponsor cannot recover, and it delays the next candidate in line. If a model can flag a likely failure earlier, even imperfectly, the savings stack up on both axes, dollars and calendar time.
Note the exact framing. These numbers come from Atlas Bio and the outlets repeating its description of the market. We have not seen an independent 2025 or 2026 dataset in the sources behind this story that pins down a fresh industry-wide Phase II success rate, so treat 70% as the company’s stated figure. It is a widely repeated order of magnitude in drug development, but the precise value shifts by therapeutic area and by how a study defines success.
How In Silico Clinical Trials Are Supposed to Work
An in silico trial replaces some or all of a real study with a computer simulation. Instead of enrolling patients and dosing them over months, a model estimates how a population of virtual patients would respond. Atlas Bio’s version is a foundation model, which means one large model trained on broad biological data and then pointed at specific drugs and diseases, in the same spirit as a language model that gets prompted for different tasks.
The Training Data: About Two Million Human RNA Readouts
The Economic Times reports that the model trains on about two million human RNA readouts. RNA readouts measure which genes are active in a sample, so they act as a snapshot of cell state. Training on human measurements, rather than on animal tissue, is the technical basis for the claim that the model captures human biology more directly than a mouse study can.
Two million sounds large, but a data engineer will ask the next questions immediately. How many distinct donors and diseases does that cover? How were batch effects between labs handled? Did the training set include treated samples, or only baseline expression? None of that appears in the available coverage, and it decides how far the model generalizes to a new compound.
The 12-Week Simulation Claim
According to a company statement reported by the Economic Times, a simulation takes 12 weeks. That is far shorter than the two to three years a Phase II trial typically takes, though the comparison needs care. Twelve weeks is the time to produce a prediction, and a prediction is not an outcome. A regulator will still want real patient evidence before approving a drug, so the speedup applies to the decision about whether to run the trial, not to the trial itself.
What Is Confirmed and What Is Not
Headlines about this story use strong verbs, and not all of them hold up against the sources. The funding facts are solid because multiple outlets report them from the announcement. The claims about replacing trials and animal studies are softer. Available coverage describes replacement as an aim or a potential application. Nobody has reported a completed replacement of a Phase II trial by Atlas Bio’s model.
| Claim | Status | What the coverage shows |
|---|---|---|
| $12.5M total raised, $10M seed plus $2.5M pre-seed | Confirmed | Reported from the announcement by Tech.eu and others |
| Creandum led the seed | Confirmed | Named in the announcement coverage |
| Model trained on about two million human RNA readouts | Reported | Economic Times |
| Simulation takes 12 weeks | Company statement | Economic Times, attributed to the company |
| About 70% of drugs fail in Phase II | Company figure | Stated by Atlas Bio, not independently tested here |
| Model replaces Phase II trials | Unconfirmed | Described as an aim or potential use |
| Model replaces animal studies | Unconfirmed | Described as an aim or potential use |
| Basel biotech client skipped animal studies and went straight to human trials | Unconfirmed | Appears in secondary reporting only, not clearly confirmed by the company site or the funding announcement |
The last row deserves a flag. Some secondary coverage says a Basel-based biotech client already skipped animal studies and moved directly into human trials using Atlas Bio’s output. We could not find that confirmed on the company’s official site or in the cited funding announcement, so we report it as unverified. If it holds up, it becomes the first real proof point. Until then it is a story, not a data point.
One more gap: no named speaker gave an on-record quotation in the sources we reviewed, so this article paraphrases the announcement and does not quote any executive or investor. Readers should expect fuller statements from Atlas Bio and Creandum in follow-up interviews.
The Animal Testing Question and the FDA Backdrop
The headline framing says Atlas Bio wants to move beyond animal testing, and the timing helps it. The U.S. Food and Drug Administration published a roadmap in April 2025 aimed at reducing animal testing in preclinical safety studies, according to the agency’s own materials. The strategy leans on what the FDA calls New Approach Methodologies, a bucket that includes AI-based toxicity models, human cell lines, organoids, and other computational or laboratory systems.
The agency also announced a plan to phase out animal-testing requirements for monoclonal antibodies and other drugs, with implementation starting for investigational new drug applications and NAMs data encouraged in submissions. In April 2026, the FDA reported that it had met its first-year goals for reducing animal testing and modernizing preclinical development. Congress had already opened the door in December 2022 with the FDA Modernization Act 2.0, which opened the door to alternatives to animal testing. Further legislation to expand non-animal methods has been proposed, though the sources behind this story did not confirm its enactment.
That policy climate is a tailwind, but it is not an endorsement of Atlas Bio. The FDA’s NAMs push focuses on preclinical safety work, while Atlas Bio’s pitch centers on predicting efficacy and safety signals ahead of Phase II. The overlap is real but partial. A regulator that welcomes computational evidence in an application has not said it will accept a simulation in place of a patient trial, and nothing in the coverage suggests Atlas Bio has asked it to.
Where Atlas Bio Sits Among Approaches
Drug developers have several ways to predict human response before a large trial. Each trades off cost, speed, and fidelity. The table below compares the broad categories in qualitative terms, because the sources behind this story provide no head-to-head benchmark data between named competitors.
| Approach | Biology used | Speed | Main weakness |
|---|---|---|---|
| Animal studies | Non-human species | Months | Species differences can mislead on human efficacy |
| Human cell lines | Human cells in a dish | Weeks | Misses whole-body effects |
| Organoids and organ-on-chip | Human-derived tissue models | Weeks to months | Hard to scale and standardize |
| Atlas Bio-style foundation model | Human RNA data in software | 12 weeks per simulation, per the company | Accuracy not yet shown in public, independent validation |
| Real Phase II trial | Actual patients | Two to three years | Costs more than $100 million, per Atlas Bio |
The table makes one point plainly. Atlas Bio is not trying to beat the real trial on accuracy. It is trying to reduce how often a sponsor commits to a doomed one. That positions the product as a screening layer, which is a more defensible claim than outright replacement and one that regulators can tolerate without changing a single rule.
Market Impact: Where Early-Stage AI Money Is Flowing
Creandum is a European venture firm, and its decision to lead a $10 million seed in a Basel startup signals appetite for vertical AI companies that sit outside the chatbot and infrastructure trade. Investor attention this fall has gone to compute and models. Our look at Meta’s $95B capital hunt and the squeeze on neoclouds shows how much capital the big builders now demand. Atlas Bio sits at the opposite end of that spectrum, with a small raise tied to a narrow, expensive problem.
Europe’s seed market has also grown more selective about size. When we tracked Europe’s biggest Q3 seed rounds, led by Callosum’s $100M, the headline checks dwarfed anything in biotech-AI. Against that backdrop a $12.5 million total looks conservative, and conservative may be right for a company that still has to prove its model against real outcomes.
The commercial model, by Atlas Bio’s own description, targets enterprise clients, which in this field means drug developers. Selling to them is slow. Procurement cycles run long, pilots need data-sharing agreements, and a biotech’s chief scientist will want to compare predictions against results the company already holds. Expect the sales story to rest on retrospective tests, where the model predicts outcomes of trials that have already finished.
The Software Engineering Problem Inside the Pitch
Strip away the biology and Atlas Bio is a software product with unusual testing requirements. A foundation model that outputs patient-level predictions needs the same discipline engineers apply to any system where a wrong answer is expensive. That means versioned training data, reproducible runs, held-out evaluation sets, and a clear record of which model version produced which prediction.
Evaluation is the hard part. Teams building language-model applications have learned to calibrate automated graders against human labels before trusting them, as we showed in our 13-step guide to calibrating an LLM judge with kappa. A clinical-trial simulator faces a stricter version of the same test. Its predictions must agree with real trial outcomes, and the agreement has to hold on drugs and diseases the model never saw in training. Developers working across the software and developer tooling category will recognize the pattern, since leakage between training and test data is the classic way a model looks better than it is.
For engineering teams in pharma, the practical lesson is to demand an audit trail. Ask any vendor in this space for the data lineage, the evaluation protocol, and the failure cases, not just the headline accuracy. Atlas Bio says it will deepen platform validation with this round, and that line item is where the company will either build or lose credibility.
Historical Context: From Animal Models to Human-Relevant Data
Drug development has leaned on animal studies for decades because regulators asked for them and because no good substitute existed. The rationale was reasonable. A compound that harms a mouse or a dog should not go into a person. The limitation is that a compound which looks safe and effective in an animal can still do nothing, or something harmful, in humans, and that gap feeds the attrition rate Atlas Bio is targeting.
The last few years shifted the ground. The FDA Modernization Act 2.0 passed in late 2022. The agency’s April 2025 roadmap went further by putting a phase-out of certain animal requirements on a timeline. The April 2026 progress update gave the policy a track record. At the same time, cheaper sequencing produced large human RNA datasets, which is the raw material for a model like Atlas Bio’s. Policy and data arrived together, and companies like this one sit where they meet.
Earlier waves of computational drug design promised similar gains and delivered more slowly than advertised. That history argues for patience. The difference now is the use of foundation models trained on human data and the explicit regulatory encouragement for non-animal methods, but the standard of proof has not changed. A model earns trust by being right on held-back cases, repeatedly, in front of skeptics.
Risks and Open Questions
Several questions stay open after the announcement. First, accuracy: the coverage gives no performance figures, so there is nothing to compare against a coin flip or against an expert panel. Second, scope: the sources do not say which diseases or drug classes the model handles well. Third, validation: the company says it will deepen it, which implies the current state is not final.
There is also a messaging risk. If the market hears that AI replaces animal tests and Phase II trials, and the product turns out to be a screening tool, disappointment follows. The company’s own wording, as relayed by outlets, frames replacement as a goal. Investors who read it as a present capability will misjudge the timeline. Buyers should hold Atlas Bio to the narrower claim it can defend now, that its model helps rank and prioritize candidates before an expensive study.
Finally, there is dependence on regulators. A model can be excellent and still sit unused if drug sponsors fear that an agency will discount simulation-based evidence. The FDA’s openness to NAMs helps, but each application is judged on its own record.
What Investors and Biotech Buyers Should Watch
Four signals will show whether the round was well placed. The first is named customers: Atlas Bio plans to onboard more enterprise clients through 2026, so any public logo would count as evidence. The second is published validation, ideally a retrospective study that predicts completed trials and shares the method. The third is hiring, since the company says it will grow its scientific and commercial teams. The fourth is regulatory contact, meaning any sign that a sponsor has submitted simulation output as part of a filing.
Bigger-picture economics matter too. Consultants have been arguing that AI’s value depends heavily on cost assumptions, and our summary of McKinsey’s $2.52T AI value curve and its $850B cost risk makes the point that returns shrink if inputs get pricier. Drug discovery sits at the high-value end of that curve. A single avoided Phase II failure saves more than the whole of Atlas Bio’s $12.5 million raise, which is why even a modest hit rate could be commercially meaningful.
Five Predictions for In Silico Trials
These are our analysis, not statements from Atlas Bio or its investors.
- Atlas Bio names at least one enterprise customer before the end of 2026. The company said it would onboard more clients this year, and a funded startup needs a public reference to support its next round.
- The first published validation will be retrospective. Predicting trials that already finished is the cheapest way to show accuracy, and it needs no regulator sign-off.
- The animal-testing claim stays a goal through 2027. Until a sponsor publicly confirms it skipped animal work on the strength of a simulation, the replacement language will remain aspirational.
- More European seed investors back vertical biology-AI companies. A Creandum-led round gives peers cover to write similar checks outside the large-language-model trade.
- Regulators ask for validation standards before they accept simulation evidence. Expect pressure for shared benchmarks, which would favor companies that publish their evaluation methods early.
Frequently Asked Questions
How much did Atlas Bio raise?
Atlas Bio announced $12.5 million in total on October 27, 2026, made up of a $10 million seed round led by Creandum and a $2.5 million pre-seed round led by First Momentum Ventures and LEA Partners.
What does Atlas Bio build?
It builds an AI foundation model for running in silico clinical trials. The model aims to predict patient-level drug efficacy and safety signals before Phase II, based on training data of about two million human RNA readouts, according to the Economic Times.
Why does the 70% Phase II failure rate matter?
Atlas Bio says about 70% of drugs fail at Phase II, a stage that costs more than $100 million and takes two to three years. Catching likely failures earlier would save both money and time.
Does Atlas Bio replace animal testing today?
Not according to the confirmed record. Coverage describes replacing animal studies and Phase II trials as an aim or potential application. A report that a Basel biotech client skipped animal studies appears in secondary coverage only and is not clearly confirmed by the company.
How long does an Atlas Bio simulation take?
The Economic Times reports 12 weeks, citing a company statement. That is the time to produce a prediction and does not replace the evidence a regulator expects from a real trial.
Who joined Atlas Bio’s board?
Katharina Wilhelm, a Partner at Creandum, joined the board as part of the financing.
What will Atlas Bio do with the money?
The company plans to expand its scientific and commercial teams, deepen platform validation, and onboard additional enterprise clients through 2026.
Is the FDA backing in silico trials?
The FDA has encouraged New Approach Methodologies, including AI-based toxicity models, and reduced some animal-testing requirements. That policy supports non-animal methods in preclinical work but does not amount to approval of simulation in place of patient trials.
