Pathos AI raises $237 million led by Bristol Myers Squibb to bring its AI to Alzheimer's

What Pathos announced

Pathos AI, a New York drug developer that runs its programmes through its own AI platform, said on 23 October 2026 that it has closed a $237 million Series E round led by Bristol Myers Squibb, at a post-money valuation of about $3 billion. On the same day it announced multi-year collaborations with Bristol Myers Squibb (BMS) and Tempus AI to build a curated dataset and agentic AI tools for Alzheimer's disease and other neurological conditions.

The deal has three parts. BMS invests and gets access to the Pathos platform, called Foundry. Tempus builds a large de-identified neurology dataset and licenses it to Pathos. And Pathos applies Foundry to that data to study how neurological diseases progress and to design clinical trials. Reuters summarised the same terms on Friday morning.

For Pathos, which until now has described itself as an oncology company, it is the first public move into neuroscience. Its pipeline today is entirely in cancer.

The $237 million Series E

According to the Pathos announcement, the Series E was led by Bristol Myers Squibb with participation from existing investors. Pathos did not name the other investors, the size of the BMS cheque or the price per share. It said the raise reflects operational momentum after it used Foundry to expand its clinical-stage cancer portfolio through a series of rapid deals, bringing the pipeline to four active clinical assets.

Chief executive Iker Huerga said Pathos is building a platform to systematically improve the probability of success in drug development, and that the money lets it scale its AI engine while steering current programmes toward key clinical milestones. The company has no approved medicine yet, so the valuation rests on its clinical pipeline and on the platform it says chooses and steers those programmes.

Why Bristol Myers Squibb is involved

BMS gets two things beyond its equity stake. It receives non-exclusive strategic access to Pathos Foundry across its broader research and development portfolio. And it secures three-year exclusive rights to use the platform for its neuroscience development programmes, the area where it already has research, clinical expertise and a growing portfolio of Alzheimer's and other neurological programmes.

Simon Davies, senior vice president and head of data and quantitative sciences at BMS, said AI has the greatest potential when it is grounded in high-quality data and applied by scientists who understand diseases and drug development. He said BMS hopes to deepen its understanding of disease progression, improve clinical-trial design and identify the patients most likely to benefit from potential therapies.

BMS is also buying more data directly. A Tempus 8-K filing shows that on 14 October BMS signed a separate oncology statement of work with Tempus worth $137.5 million in fixed payments through 31 December 2031. It covers de-identified multimodal cancer records, access to Tempus's Lens analysis software and research services, and explicitly allows BMS to use that data with Foundry, which the filing says BMS has separately licensed from Pathos.

The Tempus neurology dataset

Tempus says it will apply the infrastructure it built in oncology to create a comprehensive, longitudinal neurology dataset from hundreds of thousands of patients, designed to track outcomes over time. Pathos will license it under a five-year agreement worth $55 million.

The filing is more specific than the press release. Tempus will use commercially reasonable efforts to deliver 500,000 de-identified records of patients diagnosed with Alzheimer's disease or other specified neurological conditions, at a rate of 100,000 records a year. Pathos pays $100 per delivered record, up to $50 million, plus $5 million for record updates. Because the fee depends on delivery, Tempus only earns the full amount if it actually assembles all 500,000 records.

Tempus also licensed Pathos an agentic tool for querying and analysing the data, which the filing calls the Tempus Agent. A side letter lets Pathos sublicense the data and the tool to BMS, and bars Tempus from licensing the same neurology records to anyone else until 18 months after the later of each record's delivery or 22 October 2027. Together the oncology and neurology agreements are worth $192.5 million to Tempus over about five years.

What Foundry is supposed to do

Pathos describes Foundry as an agentic system made of thousands of specialised AI agents that aggregate everything known about a drug programme in real time, including clinical data, genomics, biological specimen data and the competitive landscape, and then recommend the next decision for that programme. The company says it uses Foundry to pick which clinical assets to license, to keep probability-of-success estimates updated, to find patient sub-groups and to change trial designs while studies run.

In neurology, Pathos says it will use Foundry to interrogate the Tempus dataset at scale, generate and test hypotheses on disease biology, and inform the design of more targeted, adaptive clinical trials. Huerga described the goal as a self-learning drug development system. He also said that intelligence alone does not bring a therapy to market, which is why the deal combines data, AI and a large drugmaker's development expertise.

It is a different bet from AI drug-discovery companies that design new molecules. Pathos mostly licenses assets that already exist and uses AI to decide how to develop them. Other recent deals have taken the molecule or lab route, such as Lilly's AI drug-delivery agreement with Metis and OpenAI's purchase of Orchestra Bio to add drug-programme agents to its Rosalind science effort.

The Tempus connection

Pathos and Tempus are not strangers. Pathos says on its about page that it was founded in 2022 by Eric Lefkofsky and Ryan Fukushima. Lefkofsky is also the founder and chief executive of Tempus, and Huerga, who became Pathos chief executive in 2025, previously worked at Tempus.

The Tempus filing addresses this directly. It says transactions with Pathos fall under the company's related person transaction policy because of overlap in executive leadership and equity ownership, and that a special committee of disinterested Tempus directors reviewed and approved the neurology order form and side letter. The Pathos press release does not mention the relationship.

Lefkofsky said that progress against a disease as devastating as Alzheimer's needs unprecedented data scale and computing power, and that the collaborations unite BMS expertise, Pathos agentic AI and Tempus data infrastructure.

Where the money goes in oncology

Most of the new money still goes to cancer. Pathos says a key priority is advancing JSKN016, a TROP2/HER3 bispecific antibody-drug conjugate it licensed from Alphamab Oncology in August, toward registrational studies. In that August deal Pathos agreed to pay Alphamab $125 million upfront and up to about $2.09 billion in milestones for rights outside mainland China, Hong Kong, Macau and Taiwan.

The funding will also support pocenbrodib in metastatic castration-resistant prostate cancer and relapsed multiple myeloma, AZD4241, an ER-targeting PROTAC partnered with AstraZeneca for breast cancer, and DO-2, a MET kinase inhibitor for lung cancer. Pathos also said it will keep scaling Foundry itself.

What is still unknown

Several important numbers are missing. Pathos did not disclose how much BMS invested, what BMS pays for Foundry access, or any milestone or royalty terms on drugs that might come out of the neuroscience work. The Tempus filing confirms that BMS has separately licensed Foundry from Pathos but gives no terms.

There is also no evidence yet that Foundry improves success rates. Pathos says it built and validated its approach in oncology, but it has not published trial results showing that AI-guided decisions produced better outcomes than conventional development. Alzheimer's is one of the hardest areas in medicine, with a long record of late-stage failures, and the Tempus dataset does not exist yet: records will arrive over five years, at 100,000 a year, if Tempus can source them.

Finally, the dataset is real-world clinical data, not trial data. Records like these can show how patients progress, but they carry gaps and biases that make cause and effect hard to prove.

Why it matters

The deal shows how large drugmakers are buying AI for development, not only for discovery. BMS is paying for a platform that chooses trial designs and patient groups, plus exclusive access to data it cannot easily assemble itself, and it is locking up both for neuroscience for several years.

It also shows the value now placed on patient data. Tempus is charging $100 per de-identified neurology record, and the exclusivity window shows that the records themselves are treated as the scarce asset. Large health-record collections are increasingly the raw material for AI research, as with Meta's Muse study across five million patient records, and money is flowing to AI-run science more broadly, including Warp Sciences and its robot chemistry labs.