AI Philanthropy Should Buy the Breakthroughs Markets Won’t

The AI boom could see anywhere from $37 billion to $100 billion in annual philanthropic spending over the next few years. The question is, how can this money be spent effectively? A new blog series from CGD makes the case for highly cost-effective areas that philanthropists may want to consider funding. This sixth and final blog in the series makes the case for investing in innovations that markets wont supply on their own. 

In the first blog in this series, Rachel Glennerster and I suggested that new philanthropists could spend tens of billions of dollars on proven cost-effective interventions. This blog is about buying the solutions that don’t yet exist. A legally binding promise to pay for a breakthrough can change what companies work on years before a dollar is paid out. And if no one succeeds, there is no payout. This blog proposes that the new AI philanthropy should do two things: (i) fund innovation through pay-for-success “pull” mechanisms and (ii) help build a permanent institution that can administer a series of these.

Why markets leave breakthroughs on the table

Innovation is the main driver of economic growth and well-being. Yet markets under-supply innovation, particularly in areas where the commercial return lags the social value, such as pandemic countermeasures, vaccines for diseases that affect poor people, and technologies to mitigate climate change. COVID-19 vaccines offer a striking example: the ability to produce three billion vaccine courses annually was worth an estimated $5,800 per course to the world, but buyers paid just $6–$40. Four common market failures drive this wedge:

1. Much of the benefits spillover to people who never buy the product, so no buyer will pay for it and no firm can charge for it.
2. Innovations that are easy to copy see their returns competed away by imitators.
3. A few large buyers (usually governments) can bargain prices down to the point they never repay the cost of invention.
4. Political and social pressure lead to price constraints on what innovators can charge, particularly in a crisis.

Most innovations worth funding run into more than one of these, resulting in less innovation than society would want.

Pull funding can help fix market failures for innovation

Innovation needs both “push” and “pull” funding. Push funding pays for inputs, such as research grants, whether or not firms produce a successful innovation. Pull mechanisms, such as prizes and advance market commitments (AMCs), pay for verified outcomes. By rewarding success rather than effort, funders are able to specify what they want and then the market delivers the team and technology. Pull works well when we know what we need but not who is best placed to solve it. With the Institute for Progress, we built the Atlas of Innovation to help funders decide what tool is right for a particular problem.

Pull funding has demonstrated its value, but AMCs—guarantees to subsidize or purchase a future product that meets prespecified requirements from any qualifying supplier—remain rare. The first “pilot” AMC, the $1.5 billion Pneumococcal AMC, launched in 2009, backed by five governments and the Gates Foundation, to guarantee a market for pneumococcal vaccines suitable for low-income countries. By 2020, vaccines supported by the AMC had reached more than 150 million children and saved an estimated 700,000 lives. Frontier, a now $1.8 billion commitment to purchase permanent carbon removal led by Stripe, showed that a similar demand-pull model can work beyond global health. Yet 17 years after the Pneumococcal AMC, only a handful of AMCs have followed because price tags are large, mechanisms take time to design, and few institutions can administer them.

The menu is ready and looking for funders

A realized AI windfall changes what is feasible. At the Market Shaping Accelerator, which I direct, we have designed a $10 billion to $12 billion slate of pull mechanisms, ranging from $12 million for safe-drinking-water sensors to multibillion dollar commitments for pandemic countermeasures (see Table 1). There are many more innovations that could be accelerated using these mechanisms, but expectations of available funding and the need for bespoke institutions have held us back from pursuing more opportunities.

The size of the price tag has itself shaped what we pursue. We never fundraised for broad-spectrum antivirals, even though it’s estimated to deliver roughly a thousand dollars of benefit for every dollar spent, because we thought the $3.5 billion price tag was too big to fill. If the new AI philanthropy is looking for worthwhile ways to deploy the windfall, here is a menu.

Table 1. Pull pipeline with price tags

What philanthropy could commit today, paid only if it works 

All figures are one-time contingent commitments, paid only on verified success — not annual spending. If no qualifying innovation is delivered, funders pay nothing.
BCR = benefit–cost ratio. DALY = disability-adjusted life year, one year of healthy life lost to illness, disability, or early death.
Returns are not comparable across rows: the underlying analyses monetize different things over different horizons and populations. Status as of August 2026.

Philanthropy can get pull mechanisms off the ground, even if governments ultimately provide most of the funding. It can often more easily back innovations whose benefits cross borders, make commitments that extend beyond annual budgets and election cycles, and act before a crisis makes the need obvious. Governments should also be involved, a recent White House science report describes pull mechanisms as “powerful and underutilized.” Philanthropic funding can be catalytic: the Gates Foundation contributed $50 million to the pneumococcal AMC, alongside $1.45 billion from five governments.

Funding the mechanisms in Table 1 is the immediate opportunity. Some are already live and fundraising; others are designed and ready for an anchor funder. Each can move forward through an existing or bespoke institutional arrangement. But if pull funding is to become a routine tool rather than a series of one-off efforts, we need to build shared infrastructure that makes future mechanisms easier to launch.

Build a home for AMCs

For pull to become commonplace, we need institutions able to house it. The Pneumococcal AMC used the World Bank to hold donor funds in trust and provide financial management, and multilateral development banks (MDBs) could host more development-focused AMCs. Donors should ask them to. But plenty of pull mechanisms are either too small to justify their own trust fund or fall outside an MDB’s mandate, and those have nowhere to live.

A standing institution—call it “the AMC bank”—could standardize contracts, make long-term commitments credible, arrange independent verification, and administer payments. Building it will take time, and who hosts it is an open question. (A 2022 CGD blog suggested one way to structure such an institution.) It needs to be credible enough for firms to trust its promises and for funders, both philanthropic and public, to feel secure holding their money there.

It is not a prerequisite for the current pipeline, but it would allow us and others to pursue many more opportunities.

Two ways to act now

New philanthropists can act now in two ways: fund one of the mechanisms already in the pipeline or help build the standing institution that makes launching future AMCs easier. The first buys breakthroughs. The second makes paying for them routine.

We are looking for partners ready to do either. If that could be you, please get in touch at [email protected].

Thanks to the entire Market Shaping Accelerator team for helpful feedback and comments.