Microsoft expands AI team with 10 researchers from Allen Institute
Microsoft has hired at least 10 researchers and staff members from the Allen Institute for AI, including the leadership and core team behind the institute's flagship open-source model project. The departures represent a significant shift in the Seattle-based nonprofit's trajectory as it faces mounting costs to compete in frontier AI research.
The hires include former Ai2 CEO Ali Farhadi, former COO Sophie Lebrecht, and research leaders Hanna Hajishirzi and Ranjay Krishna. Also joining Microsoft are researchers Luca Soldaini, Kyle Lo, Dirk Groeneveld, Pete Walsh, Matt Jordan, and Jake Poznanski-the team that developed OLMo, Ai2's fully open-source language model.
All have joined Microsoft's Superintelligence team, led by AI CEO Mustafa Suleyman. The group is focused on core mission work and AI model post-training, with applications in health care, energy, and AI companions.
Why the exodus matters
The departures reflect a structural challenge facing nonprofit AI research. When Farhadi's departure was announced in March, Ai2 board chair Bill Hilf acknowledged that the cost of competing at the frontier had become unsustainable for a nonprofit institution.
"The cost to do extreme-scale open model research is extraordinary," Hilf said. "It's really hard to do extreme-scale model work inside of a nonprofit."
Behind the scenes, changes to Ai2's funding have also contributed to the exits. The institute's primary backer, the Fund for Science and Technology, shifted from providing annual budgets to a proposal-based funding model.
Microsoft's strategy
The hiring reflects Microsoft's broader effort to reduce dependence on OpenAI for its own AI models. Ai2's team brings expertise in efficient, open-source model development-an area where the nonprofit has produced outsized results relative to its resources.
Microsoft has been investing heavily in its own AI infrastructure and model development. The addition of Ai2's researchers, particularly those with experience in transparent, reproducible model training, strengthens that capability.
Ai2's next chapter
Ai2 confirmed the departures but said its mission remains unchanged. The institute is searching for a new permanent CEO after Farhadi's exit.
Interim CEO Peter Clark outlined the path forward in a May 4 statement, emphasizing longer-term research, open models, and applied AI in scientific discovery, embodied AI, and environmental science.
Among senior researchers remaining at Ai2 is Noah Smith, the senior research director and University of Washington professor who leads a $152 million computing initiative backed by the National Science Foundation and Nvidia.
Ai2 was founded in 2014 by the late Microsoft co-founder Paul Allen with a mission to advance AI research for the common good. The institute has distinguished itself by releasing full model weights, training data, code, and evaluation tools-allowing outside researchers to inspect and reproduce its work.
For professionals in research and science, understanding these institutional shifts is relevant. The movement of researchers between organizations, funding constraints in nonprofit research, and the open-source versus proprietary model debate directly affect how AI tools and methods become available to the broader research community. Consider exploring AI Research Courses and Generative AI and LLM Courses to stay current with developments in model architecture and training approaches.
Your membership also unlocks:
