MONTY by March Capital

Tokens, Compute and the Open AI Stack | Vipul Ved Prakash, Co-Founder & CEO at Together AI

Guest: Vipul Ved Prakash, CEO & Co-founder, Together AI — together.ai
Host: Sumant Mandal, Founder & Managing Partner, March Capital

Vipul Ved Prakash has spent two decades building systems that run at internet scale. One was a spam filter that still screens about a fifth of the world's email. Another was a real-time social search engine that Apple bought in 2014. When he co-founded Together AI in 2022, just before ChatGPT launched, he bet that AI would need an open ecosystem as much as it needed frontier labs. That bet now looks like the consensus view.

In this episode, March Capital's Sumant Mandal sits down with Vipul to trace his path from Cloudmark and Topsy to Apple's Siri group, and then to Together. Together is a platform that makes open-source models easy, cheap, and safe to run in production.

Tune in as Vipul and Sumant get into what he calls this year's "quiet revolution" in open models, why infrastructure must keep pace as tokens become more valuable, and what he expects the future to look like as open models match frontier intelligence.

***

Together AI is the AI Native Cloud for inference, compute and model shaping. AI natives and enterprise use Together AI's full stack AI platform for every step in their AI development journey.

The Montgomery Summit returns to Santa Monica on March 9-10, 2027.

***

00:00 Introduction to Together.ai and Vipul Ved Prakash
02:58 The Journey of AI: From Spam Filters to Foundational Models
05:52 Building an Open Ecosystem for AI
08:48 The Economics of Open Source vs. Closed Models
12:11 Regulatory Challenges and Industry Pushback
14:58 The Future of LLMs and AI Models
18:01 Infrastructure Challenges in AI Development
20:56 Global Perspectives on AI Infrastructure
23:53 Governance and Regulation in AI
27:03 The Vision for Together.ai
30:12 Final Thoughts and Future Directions

Guest: Alex Ratner, CEO & Co-founder, Snorkel AI — snorkel.ai
Host: Solomon Hailu, Partner, March Capital

For years, AI researchers treated data as someone else’s problem — janitorial work, beneath the real science of models and architectures. Alex Ratner bet his career against that assumption. It’s now a $3.5 billion company.

In this episode, March Capital partner Sol Hailu sits down with Alex Ratner, CEO and co-founder of Snorkel AI, fresh off a $350M Series E fundraise and a year in which Snorkel grew nearly 20x.

Alex traces the arc from the Snorkel open-source project at the Stanford AI Lab to the shift he calls Data 1.0 to Data 2.0: from throwing volume at the wall to designing expert curricula and RL environments for models already operating at “PhD+” level. Along the way — why the data spec is increasingly the product spec, why no enterprise will hand its most differentiated knowledge to an off-the-shelf model, and what it means that models have become eval-aware, able to tell when they’re being tested.

He and Sol get into the jagged frontier — why a model can crack a Millennium Prize problem and still fumble an expense report — and land on what Alex calls the defining problem of the moment: for the first time in AI history, our ability to measure what models can do is lagging the capabilities themselves. His answer isn’t fewer benchmarks. It’s more of them, built in the open and harder to game.

**

Snorkel is hiring and expanding its Open Benchmark Grants program — researchers working on open benchmarks are invited to get in touch.

The Montgomery Summit returns to Santa Monica on March 9-10, 2027.

**

00:00 Introduction to AI and Snorkel
02:56 The Evolution of AI Data and Snorkel's Journey
06:04 Data-Centric AI: The New Paradigm
08:56 Data 1.0 vs Data 2.0: Shifting Focus
11:52 Collaborative Data Design and Customer Engagement
15:08 The Importance of Data in AI Applications
18:08 Navigating AI Safety and Model Development
21:12 Trends in Open vs Closed AI Models
24:06 Defining Scope for AI Implementation
26:58 The Evolution of AI Deployment
27:28 AI Safety and Its Challenges
29:02 Understanding AGI and Its Implications
30:08 The Jagged Frontier of AI Capabilities
32:34 Evaluating AI: The Need for Robust Benchmarks
36:13 Pacing AI Development for Safety
39:13 Funding Growth: Series E and Future Prospects
43:00 Exciting Applications of AI in Healthcare
46:18 Shifting Beliefs in AI Development

[Monty2026: The Infrastructure Race]

Pervinder Johar (CEO, Avathon) and Abhishek Shukla (Managing Director & Head of North America, Prosperity7 Ventures, Saudi Aramco) discuss sovereign AI with moderator Keith Strier (SVP, Global AI Markets, AMD).

The panel frames sovereign AI not as an industry vertical, but as a question of national control: whether a country truly owns the value chain when intelligence is extracted from its own data and geography.

Shukla describes how the Gulf region is converting its energy advantage into an AI moat. Johar, whose company builds physical AI for oil rigs, mining, and defense logistics, explains why these zero-hallucination-tolerance environments differ fundamentally from enterprise LLM deployments.

[Monty2026: The Infrastructure Race]

Steve Westly (Founder & Managing Partner, The Westly Group) in conversation with Daniel Weiss (Co-Founder & Managing Partner, Angeleno Group).

Westly and Weiss, both longtime clean energy investors, argue that the AI and data center buildout represents the largest energy investment opportunity in history. They note that Anthropic and OpenAI, two of the fastest-growing companies ever by user adoption, already consume more energy than any manufacturing firm.

Weiss's key comparison is that hyperscalers will spend more on CapEx in 2026 alone than all U.S. investor-owned utilities are projected to spend over the next five years combined. Both frame U.S. AI leadership as the strategic counterweight to China's dominance in solar, batteries, and EVs.

[Monty2026: The Infrastructure Race]

DJ Sampath (SVP, AI Software & Platform, Cisco), Jon Jones (Chief Revenue Officer, CoreWeave), and George Mathew (Managing Director, Insight Partners) discuss where durable AI value resides with moderator Rohit Choudhary (Founder & CEO, Acceldata)

Jones describes CoreWeave's trajectory from GPU rental to near-Fortune-100 scale in under five years. Sampath outlines a "trifecta" of enterprise AI challenges: infrastructure constraints, a trust deficit around agents, and a data gap from insufficient machine-data training.

Mathew argues that compound AI systems built around domain-specific data, not generic models, will capture most enterprise value.

[Monty2026: The Infrastructure Race]

Charles Zedlewski (Chief Product Officer, Together.ai), Robert Brennan (Co-Founder & CEO, OpenHands), and Nader Khalil (Director, Developer Technology, Nvidia) discuss open-source AI with moderator Yvonne Wassenaar (Senior Advisor & Board Director).

The panel argues that open AI systems, including accessible weights, open training methods, and community-driven inference stacks, are reaching performance parity with proprietary frontier models for most enterprise use cases at a fraction of the cost. Khalil frames Nvidia's role as enabling the full open-source stack.

Brennan positions OpenHands as proof that production-grade agentic AI can be built entirely on open foundations, while Zedlewski notes that nearly every major coding agent already runs some meaningful share of inference on open-weight models.

[Monty2026: Leadership Insights]

Rory O'Driscoll (Partner, Scale Venture Partners), Charly Kevers (CFO, Carta), Kris O'Toole (Principal, Semler Brossy), and Sean Feller (Partner, Gibson Dunn) discuss equity compensation strategy with moderator Deanna Kempinski (Senior Manager, HR Consulting, Baker Tilly)

The panel traces how extended time-to-liquidity, now averaging 13--14 years versus 4--5 historically, has reshaped the employee value proposition for equity. O'Driscoll notes that 75--80% of secondary transactions concentrate in a handful of marquee names, leaving most startup employees without meaningful liquidity for nearly a decade.

The group highlights a widening divide: SaaS companies are now heavily dilution-constrained while high-growth AI companies routinely offer engineers $10 million or more in equity for the same role

[Monty2026: Leadership Insights]

Micheline Nijmeh (CMO, ThoughtSpot), Ozge Tuncel Ozcan (CRO, Forter), and Meghan Gendelman (CMO, B2B, Canva) discuss AI-driven go-to-market strategy with moderator Ishan Mukherjee (Co-Founder & CEO, Rox)

Nijmeh describes ThoughtSpot's CEO mandate that every employee learn agentic AI, with no new headcount or tools approved until teams have first tried building their own agents. Ozcan says Forter's board no longer gives credit for adopting AI because it is now a baseline expectation, with revenue per employee as the primary metric boards track.

Gendelman offers a cautionary tale from a previous company where a mass AI content rewrite tanked SEO traffic because it lacked a point of view, underscoring that AI output still needs human editorial judgment.

[Monty2026: Leadership Insights]

Micheline Nijmeh (CMO, ThoughtSpot), Ozge Tuncel Ozcan (CRO, Forter), and Meghan Gendelman (CMO, B2B, Canva) discuss AI-driven go-to-market strategy with moderator Ishan Mukherjee (Co-Founder & CEO, Rox)

Nijmeh describes ThoughtSpot's CEO mandate that every employee learn agentic AI, with no new headcount or tools approved until teams have first tried building their own agents. Ozcan says Forter's board no longer gives credit for adopting AI because it is now a baseline expectation, with revenue per employee as the primary metric boards track.

Gendelman offers a cautionary tale from a previous company where a mass AI content rewrite tanked SEO traffic because it lacked a point of view, underscoring that AI output still needs human editorial judgment.

[Monty2026: Leadership Insights]

Part I features Ali Riley (4-Time Olympian, New Zealand Soccer), Ezra Frech (2-Time Paralympian, Team USA), and Ryan Murphy (9-Time Olympic Medalist; VP, Norwest Venture Partners) with moderator Johnny Hooper (2-Time Olympian, Water Polo).

Part II features David Michael (Chief Information Officer, LA28), Gene Sykes (Board of Directors, United States Olympic & Paralympic Committee, LA28), and Michelle McGuire (Partner & Chief Data and AI Officer, Deloitte) with moderator Jamie Montgomery (Co-Founder & Managing Partner, March Capital).

Athletes share first-person accounts of competing at the highest level and preparing for a home Olympics. Frech makes the case that the Paralympics, representing 15% of the global population, is a once-in-a-generation opportunity to shift how America views disability and elite performance. LA28 leadership then details the logistics of hosting: 351 medal events across more than 50 sports using 75% existing venues, a $2.5 billion sponsorship target already exceeded, and a ticket lottery offering 72 million tickets.