The Greylock-Andreessen Rivalry in AI, 2020-2026

The Greylock-Andreessen Rivalry in AI, 2020-2026

Two of Silicon Valley's oldest and most influential venture firms have spent the past six years locked in a quiet but intense competition over artificial intelligence. Greylock Partners and Andreessen Horowitz (a16z) have taken fundamentally different approaches to the AI boom-different enough that their portfolios tell almost opposite stories about what actually matters in the age of large language models.

This isn't a rivalry of public feuds or Twitter callouts. It's a rivalry of capital deployment, thesis clarity, and ultimately, returns. Both firms have deployed billions into AI, but the composition of their bets, the timing of their entries, and the sectors they've chosen reveal two competing visions of how AI value actually accrues. One fund leaned hard into infrastructure and foundation models. The other bet that the real money would be in vertical applications and enterprise software. Six years in, the data is still being written-but the patterns are clear enough to matter for any founder or investor trying to understand where AI capital is actually flowing.

The Fork in the Road: 2020-2021

In 2020, neither firm could have known that ChatGPT would arrive in November 2022 and reset the entire AI investment landscape. But their pre-ChatGPT positioning already suggested different philosophies.

Greylock, under the leadership of partners like Reid Hoffman and Daniel Ek's board influence, had spent the 2010s building a thesis around applied AI-machine learning embedded in consumer and enterprise products. The firm had early exposure to companies like Figma (which uses ML for design automation) and was generally bullish on AI-as-a-feature rather than AI-as-a-platform. When the firm looked at the AI landscape in 2020, it saw a maturing ecosystem of ML tools, cloud infrastructure, and the early emergence of transformer-based models, but it wasn't yet convinced that foundation models would become the primary value driver.

Andreessen Horowitz took a different view. Marc Andreessen and Ben Horowitz had been vocal about the potential of software to eat the world, and by 2020, they were increasingly convinced that AI-specifically, the shift from supervised learning to self-supervised learning at scale-would be the next major inflection point. The firm began positioning itself as a serious player in AI infrastructure and was already building relationships with researchers and teams working on large-scale language models. When a16z raised its $2.2 billion AI fund in May 2023, it was the culmination of years of thesis work that had started well before GPT-3's release in June 2020.

The timing matters because it reveals something crucial about venture capital: the best returns often go to firms that identify a trend early but don't overcommit until the market confirms it. Both Greylock and a16z made early bets, but they were betting on different outcomes.

Portfolio Composition: Infrastructure vs. Applications

If you look at Greylock's portfolio today, you'll see a mix of AI companies, but the concentration tells a story. The firm has invested heavily in enterprise AI applications-companies building on top of existing models to solve specific vertical problems. This includes security operations tools, developer productivity platforms, and business process automation. The thesis here is straightforward: foundation models are becoming commoditized (or will be), and the real defensible value sits in domain-specific applications that solve concrete problems for paying customers.

Andreessen Horowitz's portfolio, by contrast, reflects a heavier concentration in AI infrastructure, training, and inference optimization. The firm has been a major backer of companies working on model efficiency, infrastructure for training and serving large models, and the broader ecosystem required to make foundation models economically viable. This includes both direct investments in AI companies and a broader bet on the entire stack-from GPUs and cloud infrastructure to fine-tuning platforms and safety tools.

These aren't just different bets; they're different bets about where defensibility lives. Greylock is saying: "The models are becoming a commodity; the moat is in the application." Andreessen is saying: "The infrastructure and efficiency layer is where the real leverage is; whoever controls the stack controls the market."

Both theses have merit. And both have been tested by the market in real time.

The ChatGPT Moment and the Scramble (2022-2023)

When OpenAI released ChatGPT on November 30, 2022, it didn't just change the conversation about AI-it forced a rapid recalibration of portfolio strategy for every major venture firm. Suddenly, the question wasn't "Will foundation models matter?" but "How much of the value will accrue to the model builders vs. the application builders?"

Greylock's pre-existing thesis actually positioned it well for this moment. The firm had already been investing in companies that could layer on top of foundation models. As the market exploded with demand for AI applications, Greylock was in a position to back companies that could move fast and capitalize on the new capabilities. The firm didn't need to pivot as dramatically as some competitors; its thesis about applied AI had just gotten a massive tailwind.

Andreessen Horowitz, meanwhile, had to accelerate its plans. The firm's conviction in foundation models had been validated, but the market was moving faster than anyone had anticipated. In May 2023, a16z announced its $2.2 billion AI fund, a signal that the firm was doubling down on infrastructure and model-adjacent plays. This was a massive capital deployment, and it signaled a clear bet that the infrastructure layer would be where the returns would be concentrated.

But here's where the story gets interesting: both firms faced a timing problem. Greylock could move quickly on application investments, but the best application opportunities were moving at venture speed-meaning you had to get in early and accept the risk that the underlying model capabilities might not develop as fast as expected. Andreessen could deploy capital into infrastructure, but infrastructure investments often have longer time horizons and depend on the broader AI ecosystem maturing in specific ways.

Valuation Dynamics and the Reality Check

One of the most important dynamics of the 2020-2026 period has been the radical shift in AI startup valuations. In 2021, an AI startup with a good thesis and a strong founder could raise at a 10-15x revenue multiple. By 2023, that multiple had compressed significantly, and by 2024-2025, the market was demanding actual unit economics and proof of product-market fit.

This compression hit application companies harder than infrastructure companies in some ways, but it also created opportunities. Greylock's focus on applications meant the firm was investing in companies that needed to demonstrate revenue traction relatively quickly. This forced discipline-founders couldn't just raise on a vision; they had to show that customers would actually pay for their AI tools. For Andreessen, infrastructure investments had different timelines and metrics, but they were also subject to the same scrutiny: Can you actually build a defensible moat? Can you achieve unit economics that make sense?

The valuation reality also revealed something about market structure. By 2024, it was becoming clear that the AI market was bifurcating: a small number of very large, well-capitalized companies (mostly working on foundation models and infrastructure) and a much larger long tail of smaller companies (mostly working on applications). This bifurcation favored the firms with the most capital and the clearest thesis. Greylock and Andreessen both had capital, but they had different theses about where the value would concentrate.

Specific Bets and Performance Signals

Let's look at some concrete examples. Greylock has been an investor in companies like 7AI, a security operations platform that uses AI to detect and respond to threats. This is a classic application play-taking foundation model capabilities and applying them to a specific, high-value vertical where customers will pay for accuracy and speed. The company has raised multiple rounds and is tracking toward significant revenue.

Andreessen, meanwhile, has been backing companies working on model efficiency and infrastructure. The firm's $20 billion AI fund, announced in 2025, signals a major bet on the infrastructure layer and on the idea that the U.S. needs to build out its AI capabilities in response to geopolitical competition. This is a different kind of bet-less about individual company performance and more about the trajectory of the entire AI ecosystem.

Both bets have some validation. Greylock's application companies are generating revenue and building defensible positions in their verticals. Andreessen's infrastructure investments are benefiting from the massive buildout of AI capabilities and the increasing recognition that the U.S. needs to invest in AI infrastructure for both economic and national security reasons.

But the performance metrics are different. For Greylock's application companies, you're looking at metrics like revenue growth, customer acquisition cost, and lifetime value. For Andreessen's infrastructure companies, you're looking at metrics like adoption, developer mindshare, and the scale of the ecosystem they're supporting.

The Founder Perspective: Where Should You Raise?

If you're a founder trying to decide between Greylock and Andreessen for your AI startup, the choice depends heavily on what you're building. Here's the framework:

Raise from Greylock if you're building:

• A vertical application (industry-specific AI software)
• An enterprise SaaS product with AI as a core feature
• A consumer product that uses AI to solve a specific problem
• Something that needs to demonstrate revenue and unit economics relatively quickly

Raise from Andreessen if you're building:

• AI infrastructure (training, inference, optimization tools)
• A horizontal platform that other AI companies will build on top of
• Something that requires massive capital and has a multi-year path to profitability
• Something that aligns with the firm's thesis about geopolitical competition and U.S. AI leadership

Of course, this is a simplification. Both firms invest across categories. But the thesis clarity matters because it affects how the firm will support you, what metrics they'll care about, and how patient they'll be with your path to profitability.

When you're pitching an AI project, it's crucial to understand the investor's thesis and align your narrative accordingly. If you're pitching Greylock, emphasize the specific vertical, the customer pain point, and your path to revenue. If you're pitching Andreessen, emphasize the scale of the opportunity, the infrastructure you're building, and how it enables the broader AI ecosystem.

The Market Shift: From Foundation Models to Applications (2024-2025)

One of the most significant developments in the AI venture market over the past two years has been a shift in capital allocation. In 2023, there was a rush to fund foundation model companies and infrastructure. But by 2024, the market had started to recognize that the foundation model space was consolidating around a small number of very well-capitalized players (OpenAI, Google, Anthropic, xAI, etc.), and the real opportunity for venture-backed companies was in applications.

This shift validated Greylock's thesis. The firm's focus on applications suddenly looked prescient-not because Greylock had predicted the exact timing, but because the firm had been consistent about where the real defensible value would be. As AI funding data from Q2 and Q3 2024 shows, venture capital is increasingly flowing toward AI applications, not foundation models or raw infrastructure.

But this shift doesn't mean Andreessen's thesis is wrong. The firm's infrastructure investments are still valuable, and the $20 billion AI fund signals that Andreessen believes infrastructure will remain a critical layer. The difference is that Andreessen is now being more selective about which infrastructure plays to back, recognizing that not every infrastructure company will have venture-scale returns.

Risk Factors and the Downside Case

Both theses have risks. Let's be clear about them.

For Greylock's application thesis, the risk is that the moat in applications turns out to be thinner than expected. If foundation models continue to improve rapidly and become even more commoditized, then the advantage of building a domain-specific application might erode quickly. A well-funded competitor with a similar idea could outpace you. Or a larger company could build a competing product in-house and leverage its existing distribution to win the market. These are real risks, and they've killed venture-backed companies before.

For Andreessen's infrastructure thesis, the risk is that the infrastructure layer consolidates around a small number of very large players (cloud providers, chip makers, etc.), and venture-backed infrastructure companies get squeezed out. There's also the risk that the geopolitical situation changes in ways that make U.S. AI infrastructure less critical, or that international competitors build equally capable infrastructure at lower cost.

Both risks are real. And both firms are aware of them. That's why the best venture investors don't bet on a single outcome; they build portfolios that can win across multiple scenarios.

The Competitive Dynamics: Speed vs. Patience

One of the most interesting dynamics between these two firms is their different approach to speed and patience. Greylock, with its focus on applications, tends to move quickly. The firm can identify a vertical opportunity, back a founder, and start seeing revenue within 12-24 months. This is the classic venture capital playbook-move fast, iterate, and scale.

Andreessen, with its infrastructure focus, is more willing to play the long game. Infrastructure investments often take longer to mature, but they can have much larger outcomes if they succeed. The firm's $20 billion AI fund is explicitly structured to be patient capital, willing to support companies through multi-year development cycles.

Both approaches have merit. Speed allows you to capitalize on market opportunities quickly and build defensible positions before competitors enter. Patience allows you to build more defensible, harder-to-replicate technology. The question is which approach will generate better returns in the AI era.

Historically, in venture capital, patience has often been rewarded. The companies that became the largest and most valuable were often those that took time to build defensible technology and network effects. But the AI era might be different. The pace of change is so fast that speed might matter more than ever. A company that can get to market quickly with a working AI application might be able to build defensible customer relationships before competitors catch up.

Looking at the Numbers: Portfolio Size and Capital Deployment

Let's talk about the actual capital deployment. Greylock's total assets under management are approximately $3-4 billion (across all funds). Andreessen Horowitz's AUM is significantly larger, estimated at $30+ billion across all funds. When Andreessen raised its $2.2 billion AI fund in 2023, it was deploying roughly 7% of its total AUM into a single thesis. When it raised its $20 billion AI fund in 2025, it was signaling an even more aggressive bet on AI.

Greylock, by contrast, has been more measured in its AI fund deployment. The firm has AI investments across its various funds, but it hasn't raised a dedicated mega-fund for AI in the same way Andreessen has. This reflects different philosophies about capital concentration. Andreessen is willing to put massive amounts of capital behind a single thesis. Greylock is more diversified.

For founders, this matters. Andreessen can write larger checks and can support companies that need significant capital to build infrastructure. Greylock is better suited for companies that need more moderate amounts of capital and want a partner that's diversified across multiple sectors.

The Geopolitical Factor: AI as National Priority

One theme that's become increasingly important in the Greylock-Andreessen dynamic is geopolitics. The U.S.-China AI competition has become a major factor in venture capital decision-making. Andreessen has been vocal about the importance of U.S. AI leadership, and the firm's $20 billion AI fund is explicitly framed as a bet on American AI capabilities.

Greylock hasn't been as vocal about the geopolitical dimension, but the firm is clearly aware of it. Many of Greylock's application investments have national security implications-security operations tools, defense tech, etc. In fact, Capitaly's analysis of defense tech positioning shows how venture investors are increasingly thinking about AI in the context of geopolitical competition.

This geopolitical dimension could become increasingly important in the coming years. If the U.S. government decides to invest more heavily in AI infrastructure or to restrict certain types of AI research, it could shift the entire venture landscape. Andreessen's positioning as a cheerleader for U.S. AI leadership might give the firm an advantage in accessing government contracts and support. Greylock's focus on applications might be less directly affected by geopolitical shifts, but it's still relevant.

Sector-Specific Deep Dives

Let's look at a few specific sectors where Greylock and Andreessen have taken different approaches.

Enterprise Security: Greylock has been a major investor in AI-powered security tools. The thesis is straightforward: security is a vertical where customers will pay premium prices for AI-powered solutions that can detect and respond to threats faster than humans. Companies like 7AI are examples of this thesis in action. Andreessen has also invested in security, but more often in the infrastructure layer-tools that make it easier to build AI-powered security applications.

Developer Tools: Both firms have invested in AI-powered developer tools, but with different emphases. Greylock has backed companies building specific tools for specific use cases (e.g., code generation for particular programming languages). Andreessen has invested in broader platforms that developers can use to build AI applications. This reflects their different theses about where the value accrues.

Enterprise SaaS: This is a category where Greylock has been particularly active. The firm has backed companies that are adding AI features to existing SaaS products or building new SaaS products with AI as a core feature. These companies can often achieve profitability relatively quickly because they're selling into existing markets with known customer acquisition channels. Andreessen has been less active in this space, preferring to focus on infrastructure and platform plays.

The Role of Founder Pedigree and Network

Both Greylock and Andreessen have strong networks and strong track records of backing successful founders. But they have different networks and different types of founders they tend to back.

Greylock has historically backed founders with deep domain expertise-people who understand a specific vertical intimately and have a clear vision for how AI can solve problems in that vertical. The firm's partnership with founders like Reid Hoffman and others reflects this focus on domain expertise and founder-investor fit.

Andreessen has a broader network that spans academia, industry, and government. The firm is comfortable backing founders who might not have deep domain expertise but who have strong technical chops and a clear vision for building infrastructure. This reflects Andreessen's willingness to back platform plays and infrastructure companies.

For founders, understanding these differences matters. If you're a domain expert with a clear idea for an AI application, Greylock might be a better fit. If you're a strong technologist with a vision for building infrastructure, Andreessen might be more aligned.

Market Trends and Momentum

As of 2025-2026, the market momentum seems to be shifting toward applications. According to 2026 AI market trend analysis, venture capital is increasingly flowing toward companies that are building applications and solving specific problems, rather than toward foundation models or raw infrastructure.

This momentum validates Greylock's thesis. But it doesn't invalidate Andreessen's thesis. Infrastructure is still critical, and the firms that build defensible infrastructure can still achieve significant returns. The difference is that the market is becoming more selective about which infrastructure plays to back.

The 2026 AI investor's playbook emphasizes the shift from foundation models to application builders and the importance of focusing on real value creation, not just hype. This is a framework that both Greylock and Andreessen should be able to operate within, but it might favor Greylock's thesis more directly.

Lessons for Founders and Investors

What can we learn from the Greylock-Andreessen rivalry? A few key takeaways:

1. Thesis clarity matters. Both firms have clear theses about where AI value will accrue. They might be different theses, but the clarity allows them to make consistent investment decisions and to support their portfolio companies effectively. If you're a founder, you want to raise from investors who have a clear thesis that aligns with your business.

2. Timing and patience are different things. Greylock's focus on applications suggests the firm is willing to move quickly to capitalize on market opportunities. Andreessen's focus on infrastructure suggests the firm is willing to be patient and wait for the market to mature. Both approaches can work, but they have different implications for portfolio companies.

3. Capital concentration vs. diversification. Andreessen is willing to concentrate capital behind a single thesis (AI infrastructure). Greylock is more diversified. Both approaches have merit, but they have different implications for the types of companies each firm will back.

4. Geopolitics is becoming increasingly important. The U.S.-China AI competition is reshaping venture capital. Founders who can align their business with geopolitical priorities (e.g., building defensible U.S. AI capabilities) might find it easier to raise capital.

5. The market is moving toward applications. As foundation models become more commoditized, the real value is accruing to companies that can build applications on top of them. This validates Greylock's thesis and suggests that founders building applications might have an easier time raising capital and achieving returns.

What's Next: 2026 and Beyond

Looking forward, the Greylock-Andreessen rivalry will likely continue to evolve. A few scenarios to watch:

Scenario 1: Applications dominate. If the market continues to shift toward applications, Greylock's thesis will be validated, and the firm's portfolio companies should see strong returns. Andreessen's infrastructure investments might underperform, but the firm's size and capital give it the ability to weather a few underperforming bets.

Scenario 2: Infrastructure consolidates. If the infrastructure layer consolidates around a small number of very large players, Andreessen's infrastructure bets might not generate venture-scale returns. But the firm's $20 billion AI fund gives it the capital to make a few big bets that could pay off if the consolidation doesn't happen as expected.

Scenario 3: The market bifurcates. The most likely scenario is that the market bifurcates, with some value accruing to infrastructure companies and some value accruing to application companies. In this scenario, both Greylock and Andreessen should be able to generate strong returns, but their returns will come from different parts of the market.

Regardless of which scenario plays out, the Greylock-Andreessen rivalry is instructive for anyone trying to understand AI venture capital. It shows that there's no single "right" thesis-there are multiple ways to win in AI. The key is to have a clear thesis, to commit capital behind it, and to be willing to support your portfolio companies through the inevitable ups and downs of the market.

Practical Implications for Your Fundraising

If you're raising capital for an AI startup, here's how to think about the Greylock-Andreessen dynamic:

First, understand your own thesis. Are you building an application or infrastructure? Are you solving a specific vertical problem or building a horizontal platform? Your answer to these questions should inform which investors you approach.

Second, understand the investor's thesis. Before you pitch, research the investor's recent investments and public statements. Do they have a clear thesis? Does your business align with it? If not, you might be wasting your time.

Third, speak their language. If you're pitching Greylock, talk about the specific vertical you're addressing, your path to revenue, and your unit economics. If you're pitching Andreessen, talk about the scale of the opportunity, the infrastructure you're building, and how it enables the broader AI ecosystem.

Fourth, be honest about your risks. Both Greylock and Andreessen are sophisticated investors who understand the risks in AI. Don't oversell your opportunity or pretend you know things you don't. The best pitch is one that clearly articulates both the opportunity and the risks.

For more detailed guidance on pitching AI projects and raising private money, check out our comprehensive guide. And if you're looking for AI startup valuation benchmarks, we have detailed analysis on what realistic valuations look like in different AI verticals.

Conclusion: Two Paths, Both Valid

The Greylock-Andreessen rivalry in AI is ultimately a rivalry between two different visions of how AI value accrues. Greylock believes the value is in applications-in domain-specific solutions that solve real problems for paying customers. Andreessen believes the value is in infrastructure-in the platforms and tools that enable others to build AI applications.

Neither vision is wrong. Both are plausible, and both have been validated to some degree by the market. The question isn't which thesis is "right," but rather which thesis will generate better returns over the next 5-10 years. And that's a question that only time can answer.

For founders, the key takeaway is this: understand both theses, understand which one your business aligns with, and raise from investors who have conviction in that thesis. The best venture capital partnerships happen when the founder and investor share a clear vision of how value will be created. The Greylock-Andreessen rivalry shows what happens when two of the world's best venture firms have different visions-they both win, but in different ways.

As you navigate the AI fundraising landscape, remember that AI funding data shows the market is increasingly focused on applications. But that doesn't mean infrastructure isn't important. Both layers are critical. The question is where you're building and which layer you're addressing. Answer that question clearly, and you'll be in a much better position to raise capital from the right investors.

The rivalry between Greylock and Andreessen will continue to shape the AI venture landscape. But the real winner will be founders who understand both perspectives and can build businesses that create defensible value regardless of which thesis turns out to be more predictive of future returns.