A grant can do more than keep an organization alive. It can determine which teams exist, how quickly they grow and what resources they can use. This part of the network now contains enough detail to follow that process beyond the headline amounts.
The conditions behind the money
Resolution’s announcement divides its $160m Coefficient award into a $108m base and $52m conditional on hiring success and compute needs. The base also supports some onward grants. Those conditions connect institutional growth to the release of resources; the announcement does not disclose the detailed milestones or payment schedule.
OpenAI then supplied $10m in API credits, intended for work including automated formal proofs. That support gives the laboratory a practical place in the research operation alongside its philanthropic funder. The credits are resources tied to a particular provider, rather than unrestricted cash.
Shared funding does not settle the strategy
Resolution’s new Agent Foundations team openly carries forward a MIRI-adjacent research tradition. Jeremy Gillen’s joining statement also questions whether some research automation could accelerate dangerous capabilities. He explicitly withholds blanket endorsement of the organization’s other work. A well-financed institution can contain a real dispute over what progress should mean.
From a funded project to something people use
Epoch reports a $2.74m Coefficient grant for its data-center tracker. Its September expansion describes a public explorer covering 86 facilities and an estimated 44% of global AI compute. Here the output can actually be inspected: satellite images, capacity estimates and construction histories.
Epoch also supplies technical consultations and benchmark work to government and industry. Redwood, another Coefficient grantee, describes advisory relationships with DeepMind and Anthropic and a joint safety-case paper with UK AISI. Funding sustains the people who produce the measurements and frameworks institutions consult when making decisions.
Who makes the choices?
Coefficient’s structure separates its research and operating employers from the entities awarding grants. Its Tailwind programme selects prospective founders addressing catastrophic AI risk, offering support from early exploration through major expansion. Selection begins before an organization even exists.
Moskovitz’s September 11 statement connects his support for METR and Redwood to investigations, public communication and policy advice. He separately says Anthropic shares sit in his foundation, and identifies Good Ventures as a beneficiary of Anthropic’s growth. The legal shareholding vehicle remains unnamed in those statements.
The picture is increasingly concrete: capital builds research capacity; compute suppliers furnish its tools; researchers develop measurements and advice; public and private decision-makers use that work. The next question is whose judgments survive disagreement—and what a funder, supplier or client can require through the agreement.