Reid Hoffman Leaves Microsoft’s Board to Bet on AI Drug Discovery

Reid Hoffman is leaving Microsoft’s board of directors to focus full-time on Manas AI, the artificial intelligence (AI)-powered cancer drug discovery company he co-founded with oncologist and Pulitzer Prize-winning author Dr. Siddhartha Mukherjee. The LinkedIn co-founder joined Microsoft’s board in 2017, the year after the tech giant’s $26.2 billion acquisition of LinkedIn, and is stepping away after telling CEO Satya Nadella on a recent episode of his “Possible” podcast that he was ready to “get back to founder mode.”

Hoffman notified Microsoft on June 2 that he would not seek re-election at the company’s 2026 annual shareholder meeting, per a Form 8-K filed with the U.S. Securities and Exchange Commission (SEC). Manas AI has raised over $50 million across two seed rounds, targets triple-negative breast cancer, prostate cancer, and lymphoma, and has been building toward what Hoffman calls a “Move 37” moment in computational chemistry. No AI-discovered drug has yet received regulatory approval anywhere in the world.

A Tenure Built on AI Bets

Hoffman’s nine years on Microsoft’s board ran through the company’s most consequential period of technological change. He arrived as the founder who had just sold his company and spent nearly a decade inside one of the world’s most important technology boardrooms as it reshaped itself around AI. When he joined, Microsoft was primarily a cloud and productivity business. By the time he announced his exit, it had built AI into nearly every product it sold and committed tens of billions to the partnerships that made that possible.

The major moves during his tenure, in sequence:

1. 2019: Microsoft committed its first $1 billion to OpenAI. Hoffman was already an early investor in OpenAI and sat on its board when the investment closed.
2. 2022: Hoffman co-founded Inflection AI with Mustafa Suleyman, who had previously helped build DeepMind.
3. March 2023: Hoffman resigned from OpenAI’s board, citing potential conflicts of interest with his Greylock Partners portfolio companies building products on OpenAI’s platform and his personal stakes in AI startups.
4. 2024: Microsoft completed a $650 million acqui-hire of Inflection AI, structured as a licensing deal that brought Suleyman to the company as CEO of a new Microsoft AI division, with several Inflection colleagues joining him.

Those bets, particularly the OpenAI investment, changed Microsoft’s trajectory over the years that followed. Nadella integrated AI models into Microsoft 365, Azure, and Windows through a product suite called Copilot, shifting the company’s competitive position across cloud services and enterprise software. Hoffman’s presence on the board through that transformation gave him a view of how the AI investment strategy unfolded from the inside.

Filed with the SEC on June 5, the 8-K states that the board departure “is not as a result of any disagreement with management on any matter relating to the Company’s operations, policies, or practices.” Microsoft thanked him for his contributions.

What Manas AI Is Building

Manas AI launched in January 2025 with an operating model its founders describe as full-stack: building its own drug candidates from discovery through clinical trials rather than licensing AI discovery capabilities to pharmaceutical partners. Most AI drug companies operate as platform businesses, collecting milestone payments as candidates advance through someone else’s pipeline. Manas retains its programs, which captures more value if a drug succeeds but requires bearing all development costs before any revenue arrives.

The pipeline combines generative computational chemistry, advanced molecular docking, and wet-lab validation under one roof, with a separate research initiative called Project Cosmos running in parallel. Cosmos aims to build a foundation model for predicting how drug molecules interact with biological targets, designed to improve in accuracy with each program the company runs, so later drug programs run against a progressively more refined map.

Mukherjee serves as CEO. A practicing oncologist at Columbia University and the author of “The Emperor of All Maladies: A Biography of Cancer,” which won the Pulitzer Prize in 2011, he brings domain expertise in oncology to the company’s computational programs. The company’s initial disease targets are triple-negative breast cancer, prostate cancer, and lymphoma, three areas with significant unmet clinical need. Hoffman has described his own role as the company’s “AI guy” to Mukherjee’s “bio guy,” a division meant to keep AI-driven discovery grounded in clinical biology.

Funding and key backers at launch:

• $24.6 million first seed round, January 2025, led by General Catalyst and Greylock, with Hoffman investing personally
• Total raised: over $50 million across two seed rounds
• Uses Microsoft Azure as its core computing platform under a founding partnership announced at the company’s launch

Manas AI has the potential to compress the timeline to discovery of effective drug candidates while increasing the likelihood of success in clinical trials. This is a unique opportunity to dramatically change the drug discovery landscape and make a positive impact on billions of people around the world.

Ken Frazier, chairman of health assurance initiatives at General Catalyst and former chairman and CEO of Merck, speaking at the company’s January 2025 launch. General Catalyst led the seed round.

AlphaGo in the Chemistry Lab

In game two of AlphaGo’s March 2016 match against world Go champion Lee Sedol, DeepMind’s AlphaGo placed a stone at the board’s fifth row, a position professional commentators initially assumed was a mistake. The move had a one-in-10,000 probability of being played by any human master, per DeepMind’s own analysis. It was decisive. AlphaGo won that game and took the match 4-1, and the move entered AI history as a demonstration that systems trained through self-simulation could discover strategies no human teacher had devised.

Hoffman borrows “Move 37” for what he believes AI can eventually do in chemistry: discover molecular structures that researchers, conditioned by decades of pharmacological convention, would never propose. The search space for possible drug molecules runs to roughly 10^60 compounds, a figure so large that systematic human exploration of it is physically impossible. Human-directed drug discovery has historically been constrained to chemical families and structural motifs that chemists already understand. AI models trained on molecular data can explore beyond those constraints, finding candidates at positions no medicinal chemist would reach first.

Go and drug discovery share a structural challenge: both involve search spaces too large for exhaustive human exploration, and the payoff of any given move only reveals itself long after it is made. In Go, the strategic value of a fifth-row stone might not become visible for dozens of subsequent moves. In drug discovery, a molecule’s clinical value takes years to confirm. Hoffman’s argument is that AI can navigate that delayed search more thoroughly than human experts, finding candidates no trained chemist would prioritize.

Mukherjee’s clinical practice provides the biological judgment to evaluate what the models produce. Project Cosmos is the longer-range infrastructure play: a foundation model for molecular interactions that accumulates accuracy over time, so each new drug program runs against a more refined map than the one before it. No AI drug company has demonstrated that compounding capability at scale.

No AI Drug Has Reached FDA Approval

As of December 2025, no AI-designed drug had received U.S. Food and Drug Administration (FDA) approval or equivalent clearance from any major regulator anywhere in the world. Drug Target Review’s year-end assessment of the field concluded the sector remains in “a proof-of-concept phase rather than a proven paradigm shift.” One drug company CEO told the publication the industry had “really let us all down” with “failure after failure.” Over 150 programs from AI-native companies were in clinical development by early 2026, with Phase I success rates for AI-designed candidates running 80 to 90 percent, well above the historical 40 to 65 percent average for traditionally discovered drugs. The pharmaceutical industry’s overall attrition rate runs close to 90 percent at Phase III, where complex patient populations and extended timelines expose efficacy gaps that earlier tests don’t catch, and AI-designed compounds have not yet generated enough late-stage clinical data to show whether AI changes those odds.

Manas AI entered the sector at the modest end of the funding range. Its two seed rounds compare with larger raises by contemporaries:

At the company’s launch, Mukherjee and Hoffman argued that model quality and biological judgment, not the size of the initial funding round, determine outcomes in AI drug discovery. Xaira Therapeutics launched in 2024 with $1 billion in funding, more than 20 times Manas AI’s total seed haul, and has not publicly disclosed clinical program milestones. Recursion Pharmaceuticals, which acquired UK-based AI drug company Exscientia and operates one of the larger AI drug platforms in the sector, has no FDA-approved molecule. The field has attracted billions in capital without producing a single approved drug, and Manas AI will need to clear the same bar everyone else is still working toward.

Nine Years of Competing Interests

By the time Hoffman made his exit public, the web of relationships he was navigating included Microsoft, OpenAI, Greylock Partners, Inflection AI, and Manas AI simultaneously. His resignation from OpenAI’s board in March 2023 was the first structural acknowledgment of the problem: the Greylock portfolio included companies building on OpenAI’s platform, his personal holdings included stakes in AI startups using those same models, and continuing to sit on OpenAI’s board while representing those interests was a governance position he could not sustain.

Inflection AI introduced a different kind of complexity. Hoffman co-founded the company with Suleyman in 2022 while already on Microsoft’s board. When Microsoft completed the $650 million acqui-hire in 2024, Hoffman was a director of the buying company and co-founder of the company being absorbed. The transaction was structured as a licensing arrangement; Suleyman and several Inflection employees arrived at Microsoft as executives while the original Inflection entity continued to exist separately.

Manas AI’s founding partnership with Microsoft adds another layer. The startup uses Azure as its core computational infrastructure, meaning the company Hoffman chairs has a commercial relationship with the company whose board he sits on. That arrangement was in place from Manas AI’s founding in January 2025, when Hoffman was already a Microsoft director. The governance overlap ends when he steps down. The Azure partnership doesn’t.

The Board Seat Microsoft Needs to Fill

Suleyman runs Microsoft AI as CEO, so the company retains an operational connection to part of the Inflection founding team. The board directorship is harder to replace. No current Microsoft director has Hoffman’s combination of hands-on AI investing experience and direct founder credibility in the AI community. He was the one board member who had sat on the other side of the table from Microsoft when it was deciding whether to back OpenAI, and who co-founded the company Microsoft later paid $650 million to bring inside. Recruiting a replacement with comparable influence in the AI startup ecosystem will not be straightforward. The company has not named a successor.

Hoffman will remain a director until the 2026 annual shareholder meeting, which Microsoft has not yet scheduled. The company typically holds the meeting in late November or December. By then, Manas AI will have been running cancer research programs for nearly two years.