#OSAIWeek brought a packed room to GitHub HQ for an evening of sharp debates on the future of open models. The night’s three debate topics explored what openness costs, what it buys, and who gets access. M12 Managing Partner Cheryl Cheng took on the access question, arguing for the motion, “The democratization of AI requires affordable access to intelligence, not public access to model weights,” opposite Ben Brooks, Head of Public Policy at Black Forest Labs. Thank you to Microsoft’s Michael Schwarz for moderating, to the Open Intelligence Foundation and GitHub for hosting, and to everyone who came out for a lively night of debate.
M12, Microsoft's Venture Fund
Venture Capital and Private Equity Principals
San Francisco, California 52,796 followers
We accelerate the future of technology through investments, insights, and meaningful partnership with Microsoft.
About us
M12 is Microsoft’s corporate venture capital fund. We invest across key focus areas in the most promising early-stage companies that are disrupting markets in the business of today and tomorrow.
- Website
-
http://m12.vc
External link for M12, Microsoft's Venture Fund
- Industry
- Venture Capital and Private Equity Principals
- Company size
- 11-50 employees
- Headquarters
- San Francisco, California
- Type
- Public Company
- Founded
- 2016
- Specialties
- venture capital, VC, investing, technology, corporate venture capital, cloud infrastructure, AI, ML, Business SaaS, communication, productivity, big data, analytics, and security
Locations
-
Primary
Get directions
1355 Market St
San Francisco, California, US
Employees at M12, Microsoft's Venture Fund
Updates
-
In agent loops, latency compounds with every model call, which makes speed as critical as capability. In TinyFish’s production web-agent tests, Inception’s Mercury 3 completed significantly more live navigation tasks than GPT-6.1-Sol, at a fraction of the per-token cost. This kind of real-time reasoning aligns directly with our thesis on the systems that will enable agentic workloads at scale. The launch follows Mercury 2’s recognition as one of TIME’s Best Inventions of 2026 last week. We’re proud to support Inception, and congratulations to Stefano Ermon, Aditya Grover, Volodymyr Kuleshov, and the entire team.
When we launched Mercury 1, it was the first proof point that diffusion LLMs could be built for commercial applications. The open question in the AI community was whether diffusion LLMs could scale to compete with autoregressive models on complex use cases. Mercury 3 answers that question: they can. Mercury 3 is by far the most capable diffusion LLM ever released. It outperforms GPT-6 Luna and Haiku 5.5 on quality while running at 1000+ tok/sec. We’ve scaled Mercury more than 40x in total parameters, and intelligence across a range of standard benchmarks has risen 3x from Mercury 1 to Mercury 3. The scaling laws that drove progress in autoregressive models hold for diffusion too. Mercury 3 is available today on the Inception API Platform and through OpenRouter, Baseten, and Vercel. For the launch, we’re offering it 50% off, at just $0.10 per 1M input and $0.35 per 1M output. Thank you to the Inception team for making this real, our investors for their continued belief, and to all our customers for putting Mercury models to work.
-
Enterprises want the power of frontier AI without giving up control of their data. That’s pushing “sovereign AI,” once mostly a government concern, into the enterprise conversation, and open models are at the center of it. Emily Forlini at Fortune explored this shift, drawing on insights from two M12 portfolio companies. Cinder CEO Glen Wise is hearing from a growing number of businesses figuring out how to deploy open models, and Armada CEO Dan Wright says sovereign AI is where the market is heading. https://lnkd.in/eGbqVFCZ Both companies signed Microsoft’s open-weight letter this summer. Open models only matter if teams can run them in production, and that’s the layer these founders are building.
-
For startups, the road to government customers runs through FedRAMP, and for most early-stage teams it’s the longest, least predictable part of the sales cycle. During #SFTechWeek, M12 and Okta Ventures brought together a full room of founders and sales leaders at Okta’s SF HQ to talk through how to get there. Amy Johanek and Greg Hall CISSP, CEH who lead federal advisory at Okta, shared what agencies look for when evaluating and buying from startups. On the startup panel, M12’s Alan Du moderated a conversation with Katherine Hyde of Knox Systems, Inc. and Akshay Verma on navigating FedRAMP to reach government buyers. Thank you to our panelists and Stephen Lee, and to Austin Arensberg and the Okta Ventures team for co-hosting.
-
-
Forbes’ first Agentic 20 looked for AI agents companies are actually putting to work. Two M12 portfolio companies made the list. Nous Research’s open-source agent Hermes has been downloaded more than 22 million times since February, and Dillon Rolnick’s team is now taking it into the enterprise so companies can run agents and models on their own servers. Stuut, led by Tarek A., runs order-to-cash for more than 150 enterprises, with agents that work through billing disputes, deductions and payment matching. Stuut says more than $3 billion has moved through its platform. Read the full list from Forbes: https://lnkd.in/erh6--m5
-
Hermes has been downloaded 22 million times since February. For most open-source AI assistants, viral adoption among developers ends the story. For Nous Research, it's the beginning. The real test: can adoption translate to enterprise scale? M12 Managing Partner Peter Lenke, CFA frames what most investors miss — it's not volume. It's security, governance, scale, and trust. Nous raised $90 million at a $1.5 billion valuation to close that gap. Peter spoke with Yuliya Chernova at The Wall Street Journal about what separates developer enthusiasm from enterprise readiness. Read here: https://lnkd.in/gNUPMXNN Congratulations to Hervé Bizira, Dillon Rolnick and the Nous Research team.
-
-
Join us Tuesday at Okta HQ for a conversation on selling to the public sector, with operators and buyers who have worked through those deals firsthand.
Most startups don't see selling to the public sector as feasible early on. That's changing as the path becomes more accessible, with startups like Knox Systems, Inc. significantly accelerating time to FedRAMP authorization. As part of #SFTechWeek, M12, Okta and Knox are hosting Unlocking the Public Sector at Okta HQ on Oct. 6 at 4pm. Join to hear from operators who have scaled government sales and buyers who have purchased from startups directly, including M12 Managing Partner Alan Du and Steven Tamm, Greg Hall CISSP, CEH, Derek Strausbaugh, Katherine Hyde, Austin A. and Amy Johanek. RSVP here: https://lnkd.in/gBEKWyhv
-
-
Next week at #SFTechWeek, Microsoft and Ask a Chief of Staff are hosting two sessions for operators leading AI adoption, including an interactive session where Chiefs of Staff bring business problems to Microsoft data scientists.
📣 Ask a Chief of Staff is hosting not 1️⃣, but 2️⃣ panels at SF TECH WEEK by a16z! This year, we're so proud to be partnering with the Microsoft ISD Data Science practice to bring you 2️⃣ panels: 1️⃣ Bringing a Chief of Staff Mindset to AI Transformation https://lnkd.in/g7zwXRSK 🗓️ Monday, October 5th, 3:00PM - 5:00PM PST How is AI changing the way we work and what can Chiefs of Staff do to stay ahead of that transformation? 🤖 In this panel, we'll explore how leaders and operators are approaching AI implementation in practice: from identifying opportunities and driving adoption to managing change and helping teams adapt. Excited to be moderating panelists that come from our Ask a Chief of Staff community: → Jordan Ramsay, COO & former Chief of Staff at LatchBio → Tzu Kit C.tchBio → Tzu Kit C., Chief of Staff to the CEO, Founder at Atlas Computing → Danielle Lovell, Chief of Staff to the CEO at Standard Cybernetics 2️⃣ Applied AI, Live Asks & Offers: Chiefs of Staff and Microsoft Data Scientists https://lnkd.in/gKGA7p56 🗓️ Wednesday, October 7th, 3:00PM - 5:00PM PST 🙋🏻♀️ In this interactive session, Chiefs of Staff from organizations of varying sizes and stages will each present a real operational challenge they're currently facing: from process inefficiencies and knowledge management to reporting, decision-making, and cross-functional coordination. 🎉 Space is limited for each panel so make sure you sign up early to guarantee your spot!
-
-
As AI workloads scale, so do the demands on data centers, and a growing question is whether facilities can become intelligent enough to dynamically manage power, cooling, capacity, and operational complexity. Startups like etalytics are applying AI to that challenge. As part of #SFTechWeek, M12, NVIDIA, and etalytics are hosting The AI Power Crunch: Energy, Compute and the Race to Scale AI. M12 Managing Partner Michael Stewart joins Scott Wallace of NVIDIA and Niklas Panten of etalytics for a panel on how much useful compute can be delivered from every megawatt. October 6, 6:00 PM. RSVP here: https://lnkd.in/e9WGcNFT
-