Daniel Gross and Nat Friedman's AI Grant Approach, Explained

Daniel Gross and Nat Friedman's AI Grant Approach, Explained

In 2017, before anyone was seriously talking about AI startups as a category, two founders started giving away compute credits and modest grants to open-source AI projects. By 2023, that experiment had evolved into a $1.1 billion venture fund. The story of how Nat Friedman and Daniel Gross built the AI Grant-and what it reveals about investing in AI-native companies-matters more now than ever.

This isn't a story about luck or timing. It's about a deliberate thesis: that the constraint for AI founders wasn't ideas or talent, but access to compute. By removing that constraint before most VCs even noticed it existed, Friedman and Gross built what amounts to a selection mechanism that identified the best AI founders years before the category exploded. The AI Grant didn't just fund startups; it became a filter for finding founders who could move fast and think clearly in a domain where both were rare.

Let's break down how they did it, why it worked, and what it means for founders and investors trying to navigate the AI funding landscape today.

The Genesis: Compute as the Binding Constraint

When Friedman and Gross launched AI Grant in 2017, the framing was straightforward but prescient: AI research and product development required expensive hardware. GPUs were scarce, cloud credits were expensive, and most early-stage founders couldn't afford to experiment at scale. The two identified a gap that most investors hadn't yet recognized.

Daniel Gross came to this insight through his work at Y Combinator, where he'd backed AI-focused founders and watched them struggle with infrastructure. Nat Friedman, who had just stepped down as CEO of GitHub after Microsoft's acquisition, had both capital and credibility in the developer community. Together, they could offer something no traditional VC could: free or heavily subsidized compute.

The initial AI Grant program operated as a non-profit vehicle. Rather than writing equity checks, Friedman and Gross distributed cloud credits and small cash grants-often $50,000 to $250,000-to founders working on open-source AI projects and early-stage AI products. The model was founder-friendly by design. You didn't have to give up equity. You didn't have to hit specific milestones. You got access to resources and, critically, validation from two respected figures in tech.

This was radical for 2017. Most venture capital operates on the assumption that scarcity creates value. VCs create scarcity by being selective with capital. Friedman and Gross inverted the model: they created abundance in one dimension (compute) to create scarcity in another (the founders they could attract).

The Selection Mechanism: How They Identified Winners

The real insight wasn't the compute itself. It was what the grant process revealed about founders.

When you remove the constraint of capital, what do founders do? According to research on the AI Grant program, applicants had to articulate a clear vision for how they'd use the compute. They had to think about infrastructure, scaling, and product-market fit in a way that most founders weren't forced to at that stage. The grant application became a lightweight diligence process.

Consider what this filtered for:

Founders who understood their technical constraints. Most early-stage founders are vague about infrastructure. They say things like "we'll scale when we raise Series A." Founders who could articulate exactly what GPU allocation they needed, what training time looked like, and what inference costs would be showed they'd thought deeply about the unit economics of their product.

Founders building on AI, not just using it. The grant had a clear bias: it went to founders building AI-native products or infrastructure, not to companies bolting AI features onto existing businesses. This was important. It meant Friedman and Gross were backing founders who believed AI was core to their company, not an add-on.

Founders with distribution or insight. Because the grant didn't come with board seats or heavy-handed mentorship, it attracted founders who were confident enough to work independently. They also tended to be founders who had existing credibility-either through prior exits, research papers, or a following in developer communities.

By 2021, the AI Grant had evolved from a pure grant program into something closer to a venture accelerator. Friedman and Gross began offering compute-for-equity deals: founders could get access to a supercomputer cluster (the "Andromeda Cluster") in exchange for a small equity stake and the option to invest further. This was a hybrid model-part grant, part venture.

The Andromeda Cluster became crucial. Rather than distributing cloud credits from AWS or Google, they built their own infrastructure. This meant they could offer cheaper compute, more reliable access, and a shared community of AI founders all using the same resources. It also meant they had insight into which founders were actually shipping and scaling versus which were just experimenting.

The Transition to NFDG: From Grant Program to $1.1B Fund

By 2023, the AI Grant had become a funnel. Hundreds of founders had gone through the program. Some had raised institutional funding. Some had been acquired. Some had quietly shut down. But the ones that survived and scaled had proven something important: they could build AI products that users actually wanted, and they could do it lean.

This success led to the launch of NFDG (Nat Friedman Daniel Gross), a $1.1 billion venture fund. The fund represented a formalization of what had been learned. Rather than just offering compute, they now offered capital, mentorship, and-crucially-continued access to their infrastructure.

The fund thesis was explicit: invest in AI-native companies at the earliest stage, when most traditional VCs were still skeptical about AI as a category. The check sizes were larger than the original grants-typically $1 million to $5 million for seed and Series A rounds-but the philosophy remained the same. Friedman and Gross were betting on founders who understood their technical constraints and had a clear vision for an AI-native product.

What made NFDG different from other early-stage AI funds that launched around the same time? Several things:

They had pattern recognition. By 2023, Friedman and Gross had already backed dozens of AI founders through the grant program. They knew what worked and what didn't. They could spot a founder who was thinking clearly about inference costs versus training costs, or who understood the difference between a fine-tuned model and a retrieval-augmented generation system.

They had infrastructure. Most VCs can write checks. NFDG could write checks and give founders access to compute. This was a competitive advantage, especially in 2023 when GPU scarcity was real and expensive.

They had credibility in the AI community. Friedman had built GitHub into a platform that was essential to developers. Gross had spent years at Y Combinator backing technical founders. When they backed an AI startup, other investors paid attention.

How the Selection Process Actually Works

For founders considering applying to AI Grant or NFDG, it's worth understanding what they're actually looking for.

The application process is deliberately lightweight. Unlike some accelerators that require 20-page business plans, NFDG's process is more conversational. They want to understand three things:

What problem are you solving? Not "we're using AI to do X," but "here's why X matters, and here's why AI is the right tool." This matters because it filters for founders who have thought about the problem space, not just the technology. Founders who can articulate why AI is necessary-not just convenient-tend to build more defensible products.

What's your unfair advantage? This could be domain expertise (you spent 10 years in biotech before starting a biotech AI company), technical expertise (you published papers in the field), or distribution (you have an existing user base). The point is to understand why you specifically are the right founder to build this.

What's your path to product-market fit? This is where the selection mechanism really works. Founders who can articulate a clear path from current state to a product that users will pay for are more likely to get funded. And they're more likely to succeed. This is especially important for AI companies, where it's easy to get seduced by the technology and lose sight of the user.

Once a founder is accepted, the process is structured but not rigid. They get access to compute (either cloud credits or the Andromeda Cluster, depending on the program). They get introductions to other founders in the cohort. They get office hours with Friedman and Gross. And if they're progressing well, they get the option to raise a larger round from NFDG.

The Economics: Why $250K Grants Punch Above Their Weight

One of the most interesting aspects of the AI Grant model is how much value it created relative to the capital deployed. A $250,000 grant is modest by venture standards. But in the context of early-stage AI, it was transformative.

Here's why: at the point when most founders would be approaching investors for a seed round, an AI Grant recipient already has:

• Proof of execution. They've built something, shipped it, and gotten feedback. They've done this on the grant, not on investor capital.
• Validation from smart people. Getting the grant is itself a signal. When you approach other investors, you can say "Nat Friedman and Daniel Gross backed this." That opens doors.
• Runway and focus. The $250,000 gives a small team 6-12 months to focus on product, not fundraising. For AI companies, this is crucial. The pace of change is so fast that even 6 months of focused work can move the needle significantly.
• Infrastructure. If they're using the Andromeda Cluster, they have reliable, cheap compute. This is a competitive advantage against bootstrap founders or founders using cloud credits from larger funds.

When it comes time to raise a seed round, grant recipients are typically in a much stronger position than they would have been without the grant. They have traction, they have validation, and they have a story. This is why AI Grant alumni have raised billions in follow-on funding.

The grant also serves as a filter for founders who are serious. Applying for a grant is free and easy. But actually executing on the grant-shipping a product, iterating based on feedback, proving that you can move fast-is hard. Only founders who are genuinely committed make it through. This is why the grant has such a high success rate relative to the number of applicants.

Lessons for Founders Raising Capital

If you're an AI founder raising capital, what can you learn from the AI Grant model?

First, understand your technical constraints. Before you pitch investors, you should be able to articulate exactly what compute you need, what it costs, and how that affects your unit economics. This is one of the things that separates AI founders who think clearly from those who don't. When you're pitching, investors will ask about inference costs, training costs, and how those scale with users. If you can't answer these questions, you're not ready to raise.

For more on this, check out our guide on AI startup valuations and how to pitch AI projects for private funding.

Second, build on real infrastructure, not just open-source models. One of the reasons Friedman and Gross's approach worked is that they weren't betting on founders who just fine-tuned GPT-3. They were betting on founders who understood how to build systems around models. This means thinking about retrieval-augmented generation, prompt engineering, data pipelines, and inference optimization. The founders who do this well are the ones who raise the most capital and build the most valuable companies.

Third, focus on founder-investor fit, not just capital. The AI Grant model works because Friedman and Gross are genuinely interested in helping founders succeed, not just in returning capital. When you're raising, think about whether your investors are people you want to work with for 7-10 years. Are they going to be useful when things get hard? Do they understand your space? Have they backed founders like you before? These questions matter more than the check size.

For perspective on this, explore our capital raising playbooks for founders and common fundraising myths.

Lessons for Investors

For investors and fund managers, the AI Grant model offers several insights that are worth thinking about.

First, scarcity creates opportunity. In 2017, compute was scarce for early-stage founders. By identifying this constraint and removing it, Friedman and Gross created a moat. They could attract founders that traditional VCs couldn't. By 2023, when everyone wanted to invest in AI, NFDG had already backed the best founders. This is a lesson in identifying bottlenecks before they become obvious.

Second, early selection is powerful. The grant program was, in effect, a way to identify founders at an earlier stage than traditional venture. By the time founders were raising Series A, NFDG had already seen them execute. This gave them a significant information advantage. If you're an emerging fund manager, think about how you can get early visibility into founders. This might be through a grant program, through technical mentorship, or through infrastructure access. The point is to find a way to see founders before they're "ready" for traditional venture.

Third, founder-friendly terms compound over time. The AI Grant didn't require equity. It was gift capital, or compute-for-equity at very favorable terms. This meant that founders who went through the program felt grateful and aligned with Friedman and Gross. When they raised Series A, they often came back to NFDG. When they raised Series B, they brought other founders. This network effect is powerful and hard to replicate with traditional venture terms.

For more on how the venture landscape is evolving, check out recent trends in AI funding and how major funds like a16z are approaching AI.

The Andromeda Cluster: Infrastructure as Competitive Advantage

One of the most under-appreciated aspects of NFDG's strategy is the Andromeda Cluster. This is a supercomputer cluster that NFDG makes available to founders, either as part of the grant program or as a benefit of raising from them.

The Andromeda Cluster is important for several reasons. First, it's cheaper than cloud compute. When you're training large models, the difference between $0.50 per GPU-hour and $1.00 per GPU-hour matters. Over the course of a year, that's the difference between $50,000 and $100,000 in compute costs. For a seed-stage company, that's significant.

Second, it's reliable. Cloud providers like AWS and Google have been experiencing GPU shortages. The Andromeda Cluster, because it's dedicated to NFDG-backed founders, is reliably available. This matters for founders who are on tight timelines.

Third, it creates a community. Founders on the Andromeda Cluster are using the same infrastructure, which means they run into each other in Slack channels, at office hours, and at events. This creates a network effect. Founders learn from each other, refer customers to each other, and sometimes co-found companies together.

This is a model that other funds have tried to replicate. Andreessen Horowitz's $20 billion AI fund has made infrastructure access a central part of their strategy. But NFDG had a head start, and the community they've built is hard to replicate.

Notable AI Grant Alumni and Outcomes

The proof of the AI Grant model is in the results. Several companies that went through AI Grant have gone on to raise significant capital or achieve successful exits.

While specific names and valuations change over time, the pattern is clear: AI Grant alumni tend to raise more capital, at higher valuations, than founders who didn't go through the program. This is partly because the grant program selects for high-quality founders. But it's also because the grant gives founders a runway to prove their concept before approaching institutional investors.

The grant program has also been a source of deal flow for NFDG. Founders who go through the grant program and show strong execution often become NFDG's best seed and Series A investments. This is a virtuous cycle: the grant program generates deal flow, the deal flow generates returns, and the returns fund more grants.

How to Apply and What to Expect

If you're an AI founder interested in applying to AI Grant, here's what to expect:

The application is straightforward. You'll need to describe your project, your team, and what you're building. You don't need a full business plan or a pitch deck. A clear, concise description of your idea is enough.

The timeline is relatively fast. Applications are typically reviewed within a few weeks. If you're selected, you'll have a call with someone from the team to discuss the details of the grant and the compute you'll need.

The commitment is modest. You're not giving up equity (unless you opt into the compute-for-equity model). You're not committing to regular board meetings or investor updates. You're just committing to building something and keeping the team informed about your progress.

The follow-on is optional but valuable. If you're progressing well and interested in raising capital, NFDG can help facilitate introductions to other investors or lead a seed round themselves. But this isn't required. Some founders go through the grant program and bootstrap to profitability. Others raise from other sources. The point is that you have optionality.

For more on the mechanics of raising capital, check out our step-by-step guide to creating a capital raising plan and our resource on raising capital without warm introductions.

Why the Model Works: The Deeper Insight

At its core, the AI Grant model works because it's aligned with how AI companies actually develop. AI companies don't need capital first. They need compute, clarity, and feedback. Capital comes later, once you've proven that your product works.

Traditional venture capital is built on the assumption that capital is the constraint. You raise capital, you hire people, you build a product, you find product-market fit. But for AI companies, especially in the early days, compute is the constraint. You need to experiment with models, data, and infrastructure to figure out what works. Once you've figured that out, capital becomes useful for scaling.

By removing the compute constraint, Friedman and Gross enabled founders to focus on the actual hard problem: building a product that users want. This is why the model is so effective. It's not just about the capital or the compute. It's about understanding what founders actually need at each stage and providing it.

The Future of AI Funding and the Grant Model

As the AI market matures, the dynamics are shifting. Compute is becoming less scarce. More funds are offering infrastructure access. The grant model itself is becoming more competitive.

But the core insight-that early-stage selection and founder support matter more than capital-is becoming more important, not less. As AI becomes a crowded category, the funds that can identify the best founders early and help them succeed will outperform. This is why NFDG's model is likely to remain competitive even as the market evolves.

For founders raising in 2025, the lesson is clear: seek out investors who understand your space, who can provide value beyond capital, and who are genuinely interested in your success. The best capital is the capital that comes with support, infrastructure, and network. This is the AI Grant model in a nutshell, and it's a model that's worth studying regardless of whether you're raising from NFDG or from another source.

Explore 10 game-changing AI startup ideas to understand the kinds of companies that are attracting capital in 2025. And if you're building in AI, consider whether a grant program might be a good fit for your stage and your needs.

Conclusion: A Model Worth Replicating

The Friedman-Gross approach to AI funding is notable not because it's unique-many funds now offer infrastructure and mentorship-but because it was prescient. In 2017, when AI was still a niche category, they identified the constraint that would matter most: compute. By removing that constraint, they attracted the best founders and positioned themselves to dominate the AI funding landscape when it exploded.

For founders, the lesson is to seek out investors who understand your constraints and can help you overcome them. For investors, the lesson is to think deeply about what founders actually need at each stage, and to find ways to provide it before it becomes obvious. And for everyone watching the AI market, the lesson is that the funds that win aren't the ones with the most capital-they're the ones with the best founders, and the best way to attract founders is to solve their actual problems.

The AI Grant started as an experiment. It became a model. And it's now a playbook that other funds are trying to replicate. That's the highest compliment you can pay to an investment thesis.