OpenAI Foundation Commits $25 Billion to Science, First Grants Target Cancer Vaccines and Alzheimer's

The OpenAI Foundation recently announced a batch of grants totaling more than $125 million to build open health and life sciences data, under a project called "Public Data for Health." This is the latest concrete expenditure to land from the $25 billion first-phase commitment the Foundation has disclosed.

To understand the scale of this money, one must return to the structure at the completion of the restructuring: the OpenAI Group equity held by the Foundation was valued at approximately $130 billion, of which $25 billion will be distributed as a first-phase commitment, focused on two areas—health and curing disease, and AI resilience (technical solutions to reduce the risks of advanced AI). Jacob Trefethen, the Foundation's life sciences lead, said the Foundation plans to distribute more than $1 billion in grants across all areas in its first year; it has already approved more than $250 million in the life sciences direction, focused on three challenge tasks.

There is a structural feature here that is easily overlooked: the Foundation's resources come almost entirely from OpenAI Group equity, and it also holds a warrant—if the Group's stock price rises more than 10-fold after 15 years, the Foundation will receive a large amount of additional equity. OpenAI says the Foundation will be the largest single long-term beneficiary of its success. This means that the Foundation's grant-making capacity and the valuation of the company it oversees are two ends of the same chain.

This is not a moral accusation, but a governance fact: a foundation that makes decisions on the basis of public interest has its funding sources, appreciation expectations, and grant areas highly bound to a commercial company. Whether it can remain independent does not depend on statements, but on institutional arrangements.

In its early stage, the Foundation puts more money into producing and preserving data, and less directly into specific R&D pipelines. Trefethen's explanation is that intelligence plus more observation of the world is what can bring breakthroughs against disease; on problems like Alzheimer's, human knowledge is not yet sufficient to develop truly good therapies, and more data of various forms is needed.

In its announcement, the Foundation gave the reasons why charitable funding needs to step in: certain datasets with high public value are difficult for any single institution to fund and build, either because they lack the capacity, or because commercial incentives are insufficient to support them in carrying out such projects. On this basis, the Foundation proposed three categories of priority data for funding: "linked data" spanning multiple scales such as molecules, cells, and tissues; "scarce data" that cannot be reacquired once missed; and "direct data" that directly measures biological and clinical states.

This orientation is clearly reflected in specific projects. The "Public Data for Health" project includes: funding the University of North Carolina at Chapel Hill's "Generative Immunotherapy Initiative" to advance personalized cancer vaccine development; funding the open pharmacokinetic and toxicology data project (OpenADMET) led by a UC San Francisco team to predict the ability of drugs to cross the blood-brain barrier and other biological membranes, and organizing public prediction competitions; and establishing the Common Technical Document Commons (CTD Commons) to collect and make public legacy data from bankrupt biotechnology companies and FDA filing documents from failed drug programs.

The last item is especially noteworthy: it collects filing documents from failed programs. Under commercial logic, there is almost no incentive to make such data public, yet it is precisely the key material for preventing later entrants from repeating the same mistakes. The Foundation's choice to enter from here shows that it has a clear judgment about "what the market is unwilling to do."