TL;DR
- Key Hires: Microsoft has recruited former Ai2 CEO Ali Farhadi and three other top researchers to join Mustafa Suleyman’s Superintelligence team.
- Strategic Shift: Suleyman is now focused on building frontier AI models in-house after stepping back from Copilot product oversight.
- Nonprofit Impact: Ai2 faces one of its largest talent losses as the cost of frontier AI research pushes top scientists toward Big Tech.
- Broader Pattern: Microsoft has been aggressively hiring AI talent from Google DeepMind, Meta, and OpenAI while developing its own MAI models.
Microsoft has hired former Ai2 CEO Ali Farhadi and three other top AI researchers from the Allen Institute for AI and the University of Washington to join Mustafa Suleyman’s Superintelligence team. Hanna Hajishirzi, Ranjay Krishna, and former Ai2 chief operating officer Sophie Lebrecht are also joining the effort. Suleyman took charge of the team after a Copilot leadership change freed him from overseeing consumer products.
Microsoft’s Superintelligence team, formed in November 2025, has already recruited researchers from Google DeepMind, Meta, OpenAI, and Anthropic. Suleyman called Hajishirzi “one of the most cited researchers of natural language processing in the world, full stop” in a LinkedIn post announcing the hires. With four new additions from a single institution, Ai2 now faces one of its largest talent losses since its founding.
Talent Microsoft Is Acquiring
Hajishirzi, a University of Washington professor, co-leads the OLMo open-source language model project at Ai2. She also serves as co-principal investigator on a $152 million initiative, backed by the NSF and Nvidia, to build open AI models for scientific research. Days before her Microsoft move was revealed, she appeared at Nvidia’s GTC conference on a panel about the future of open models alongside Nvidia CEO Jensen Huang.
Her dual expertise in open-source model training and large-scale NLP research makes her a natural fit for Suleyman’s ambition to build frontier models in-house.
Meanwhile, Krishna, also a UW professor, led development of Ai2’s Molmo multimodal models, a family of open vision-language systems that earned recognition for competing with proprietary alternatives at a fraction of the cost. Multimodal AI, which combines visual and language understanding, is a key frontier in model development where Microsoft has lagged behind competitors like Google and OpenAI.
As a result, Krishna’s work directly complements Hajishirzi’s language model expertise, giving Microsoft a materially broader base of open-research knowledge to draw from as it builds its own frontier systems.
Furthermore, Lebrecht, Ai2’s former COO, brings operational experience running a research nonprofit at scale. She co-founded the AI company Neon Labs and holds a PhD in cognitive neuroscience from Brown University. Her background offers Microsoft practical insight into building and managing AI research organizations from the ground up, including hiring pipelines, publication strategies, and cross-team coordination.
Notably, Farhadi, Hajishirzi, and Krishna will retain their faculty positions at the UW Allen School of Computer Science and Engineering, maintaining a bridge between academia and corporate research.
Building on this talent base, Suleyman has framed his team’s mission as building “humanist superintelligence” through safer, controllable, and more capable AI systems. Under Farhadi’s leadership, Ai2 released more than 100 models in a single year, a pace of open-source output that now benefits Microsoft’s internal ambitions. By acquiring an entire leadership layer from a single research institution rather than picking off individual contributors, Microsoft gains not just talent but an established collaborative dynamic that typically takes years to build.
Microsoft’s Broader Superintelligence Strategy
Microsoft’s Ai2 hires fit a broader pattern of assembling frontier AI talent at speed. On March 17, Satya Nadella announced a Copilot leadership update that freed Suleyman from Copilot oversight, with Jacob Andreou, formerly SVP at Snap, leading the unified Copilot experience as EVP.
“Progress at the AI model layer is more critical than ever to our success as a company over the next decade and is foundational to everything we build above it. We are doubling down on our superintelligence mission with the talent and compute to build models that have real product impact.”
Satya Nadella, Microsoft CEO (via Microsoft blog post)
Nadella’s emphasis on the model layer reflects a strategic shift. Microsoft spent the past two years integrating OpenAI’s models into products across its stack, from Microsoft 365 Copilot to Azure AI services. However, the company is now building parallel capacity to develop its own frontier models.
Since July 2025, Microsoft has poached 20 engineers from Google’s DeepMind and unveiled in-house MAI-1 and MAI-Voice-1 models, signaling a sustained push toward internal AI capabilities that complements rather than replaces its OpenAI partnership.
In addition, Suleyman’s repositioning brings him closer to the technical frontier. Before joining Microsoft in March 2024 through the Inflection AI acquisition, he co-founded Google’s DeepMind. Leading a dedicated superintelligence effort with hand-picked researchers aligns more closely with that background than managing consumer Copilot products did.
In his internal memo, Suleyman committed to building world-class models for Microsoft over the next five years, with a long-term frontier-scale compute roadmap already locked in place.
Ai2 Faces an Uncertain Future
For Ai2, the departures raise questions about the viability of nonprofit AI research at frontier scale. Paul Allen, the late Microsoft co-founder, founded the organization in 2014 with a vision of a nonprofit lab that could advance AI for the common good. Farhadi left his role leading Ai2 on March 12 after leading the organization for more than two and a half years.
Before joining Ai2, he had co-founded Xnor.ai, which Apple acquired in 2020 for an estimated $200 million, according to GeekWire. He later led machine learning efforts at Apple before returning to Ai2 in July 2023.
“The cost to do extreme-scale open model research is extraordinary. It’s really hard to do extreme-scale model work inside of a nonprofit.”
Bill Hilf, Ai2 board chair (via GeekWire)
Hilf’s assessment captures a structural challenge facing all nonprofit AI labs. According to GeekWire, Ai2’s primary backer is now the Fund for Science and Technology, a $3.1 billion foundation created under Paul Allen’s instructions. FFST is shifting to a proposal-based funding process that favors applied uses of AI over the costly work of frontier models, though all Ai2 programs for 2026 remain fully funded.
Consequently, with FFST prioritizing real-world AI applications over open-source foundation models, Ai2’s signature research program, building fully open alternatives to proprietary AI systems, may face a resource squeeze even before accounting for the departures. Farhadi and board chair Hilf had been discussing the leadership transition for about six months before his departure, suggesting the move was carefully planned rather than abrupt.
Farhadi himself has said he wants to pursue research ambitions at the frontier of large-scale AI, work that requires the billions of dollars in compute that only for-profit companies can deploy.
In response, Ai2 interim CEO Peter Clark said the organization’s initiatives are backed by an experienced team with the continuity needed to carry the work forward, expressing confidence in expanding the impact of current efforts. Clark, a founding member of Ai2, served the same interim role after founding CEO Oren Etzioni departed in 2022. His return to the interim position underscores both institutional stability and a recurring pattern of leadership turnover at the nonprofit.
More broadly, a widening structural gap between nonprofit and corporate AI research underlies the talent drain. Organizations like Ai2 can incubate open-source breakthroughs and train top researchers, but they cannot offer the compute budgets, compensation packages, or deployment scale that Big Tech commands. Each departure reinforces the cycle: as top talent leaves, the remaining team faces greater pressure to deliver with fewer resources, making the next round of departures more likely.
Farhadi, Hajishirzi, and Krishna will retain their University of Washington faculty positions, preserving a partial bridge between the academic and corporate worlds. But for Ai2, an organization built on Paul Allen’s vision of open AI research now watching its leadership depart to Microsoft, the company Allen co-founded, the path forward depends on whether FFST’s pivot toward applied AI can sustain the nonprofit through a period when frontier model development demands resources only for-profit companies can muster.
