Gokul Rajaram's Eight Moats: A Framework for What Survives AI

The “SaaS Apocalypse” narrative has a problem: it assumes AI replaces software. Gokul Rajaram, one of Silicon Valley’s most prolific operator-investors, argues the real question isn’t whether SaaS survives, but which companies have the structural defenses to thrive. He’s built a framework for answering that question, and it starts with counting moats.

An Operator’s Education

Rajaram’s investing thesis was forged across four companies that each taught him something distinct:

Google taught him that the best companies have a remarkable product at their core. Google’s philosophy was “build it remarkably and they will come.” GTM was never Google’s specialty; the product did the work. This became his first filter: does this company have a product so good it creates pull?

Facebook taught him the power of ecosystem lock-in. You could copy Facebook’s product, you could try to reverse-engineer the network, but you couldn’t replicate the developer ecosystem, the advertiser ecosystem, the data ecosystem. Facebook was a platform, and platforms create compounding defensibility.

Square (Block) taught him operational rigor in physical-world businesses. When you’re moving money and hardware, execution discipline matters in ways pure software companies never face. It also showed him that fintech serving small merchants can build massive scale by creating non-consumption markets.

DoorDash was the masterclass in local logistics and marketplace dynamics. Tony Xu’s deeper understanding of the same market everyone else was also chasing was the difference between DoorDash and its competitors.

The Eight Moats Framework

Rajaram proposes eight types of moats. Any company with four or more is “pretty damn secure”:

1. Data moat: Proprietary data that improves the product and compounds over time
2. Workflow moat: Deep integration into the customer’s daily workflow, making switching painful
3. Regulatory moat: Compliance complexity that creates barriers to entry
4. Distribution moat: Owned channels or embedded distribution advantages
5. Ecosystem moat: Third-party developers, integrations, or marketplace participants who build on top of you
6. Network moat: Value that increases with each additional user
7. Physical infrastructure moat: Hardware, logistics, or real-world assets
8. Scale moat: Unit economics that improve with volume, making it impossible for smaller players to compete

The framework isn’t theoretical. Rajaram applies it live to the Atlassian vs. Monday comparison:

Atlassian has at least five moats: workflow (deeply embedded in engineering teams), ecosystem (massive marketplace of plugins), data (years of project history), distribution (bottoms-up adoption), and scale. It would be extraordinarily painful for an engineering organization to rip out Jira, even if they hate it.

Monday is more vulnerable. It has workflow integration but fewer ecosystem lock-ins. The switching cost is real but lower. In the AI era, Monday’s horizontal positioning makes it more susceptible to AI tools that can replicate basic project management.

The “SaaS Apocalypse” Is Overblown

Rajaram pushes back hard against the narrative that AI will destroy SaaS:

“I think anything four or more, you’re pretty damn secure.”

Markets are overreacting. The companies with deep moats will absorb AI as a feature, not be displaced by it. The ones at risk are companies with weak moats, thin workflow integration, and no data advantage. But that was true before AI, too. AI just accelerates the sorting.

The critical distinction is bolt-on AI vs. real AI products. Most companies are doing it wrong: they’re adding an AI chatbot or a “magic” button to an existing product without rethinking the core workflow. Real AI products change the fundamental interaction model. They don’t add AI on top; they rebuild around it.

For vertical AI SaaS specifically, Rajaram argues you can still build $10B+ companies, but only with full-stack ownership. “You cannot be a single product company. Vertical products, you’ve got to really own full stack. It’s harder otherwise to be a 10 plus billion dollar company.”

Pricing Power as the Ultimate Test

The conversation takes an interesting turn when Harry Stebbings brings up margins and pricing:

Rajaram doesn’t worry about current margins. He worries about pricing power potential: does this company have the ability to raise prices in years 3, 4, and 5? Durable, defensible companies will figure out margins. The reverse isn’t true.

He uses the Chanel analogy: Chanel raises handbag prices 10% every six months, for the same product. That’s the extreme of pricing power, built on brand, scarcity, and desirability. In software, the equivalent is a product so embedded in the customer’s workflow that switching is unthinkable, giving you perpetual pricing leverage.

The question of who you sell to matters enormously. Palantir has fewer than a thousand customers paying $20-100M+ each. Robinhood needs hundreds of millions of users because ARPU is low. Both can work, but the risk profiles are completely different. Atlas (selling to billionaires) vs. Robinhood (selling free to everyone) represent the two extremes.

The Market Sizing Trap

Rajaram’s biggest miss on market sizing: Shopify.

He saw Shopify at around a billion-dollar valuation and thought: how many e-commerce merchants are there, really? The TAM felt constrained. What he missed was that Shopify wasn’t selling e-commerce; it was enabling anyone on the planet to sell anything. It created a non-consumption market.

“That’s what you want platforms to do. They literally make it possible for every person to think of the possibility of selling.”

This pattern recurs in every iconic company: Google, Facebook, Uber, Airbnb. They all look like they’re in small markets until you realize they’re creating behaviors that didn’t exist before. Venture capital, at its core, is about betting on new behaviors.

The corollary: non-consumption is the biggest challenge in market sizing. Bottoms-up analysis with customer segmentation is necessary, but it will systematically underestimate platforms that create new categories.

Navigating $300M+ Series A Valuations

The current fundraising environment has Series A rounds at $300-500M valuations for companies with $3M in revenue. Rajaram’s advice to Series A investors like Stebbings:

You can’t build a Series A fund doing deals at $300-400M. Ownership won’t be sufficient. Instead:

• Mix seed/incubation bets (backing exceptional founders at inception) with Series A bets
• Use concentration as a friend: 15 companies per fund instead of 30-40, allowing more time per company
• Reserve 35% for follow-on, because the fund will likely be named after one company (the way Founders Fund has “the Anduril fund,” “the SpaceX fund”)

On the doubling-down philosophy, he contrasts two approaches through Trade Desk’s seed investors: Founder Collective (never does pro-rata, only first checks) vs. Roger Lee at I Ventures (doubles down again and again). Both generated massive returns. Founder Collective got an enormous multiple; Roger got a huge dollar return. Different philosophies, both valid.

The unsung hero example: Napoleon Ta at Founders Fund, who leads the growth practice and decides which portfolio companies to double down on. That function, knowing when to exploit vs. explore, may be the most important job at any venture firm.

When to Sell: IRR Over MOIC

Most early-stage firms obsess over MOIC (multiple on invested capital) and ignore IRR. Rajaram shares the cautionary tale: a firm returned 7x MOIC over 20 years, which works out to a teens IRR. “That is a venture venture firm,” he says, implying it’s not a great result despite the headline multiple.

His framework: at every liquidity opportunity, calculate the go-forward IRR. If it’s lower than what you’ve promised your LPs, sell at least a portion.

He endorses Fred Wilson’s strategy: sell a third, hold a third, trade a third when an asset is fully liquid.

The secondary market development has been one of the most important changes in venture over the past few years. Now is a “hyperliquid” period, and firms should be thoughtful about whether holding serves their LPs or just their narrative.

The Mega Fund Orphan Problem

Mid-level partners at mega funds are spinning out in waves: Nikico Bonatos, Max Gazour, Arif Yan Muhammad. Rajaram expects more.

The core problem: mega funds deploy $15M Series A checks as optionality and lead generation for future rounds. It works as a strategy, but when the partner who wrote the check leaves, the company is orphaned. No advocate on the board, no relationship with the replacement. Repeat founders have started seeing through this, having lived through the experience of losing their champion at a mega fund.

“Being a mid-level partner at a mega fund is not what it’s cracked out to be.”

The industry is cycling back toward small partnerships with deep founder relationships, the way venture worked 20-30 years ago.

Afterthoughts

A few threads worth pulling on:

• Pattern matching is the investor’s worst enemy. Rajaram dismissed Quince (now valued at $10B) at $100M because “D2C companies are on the downswing.” The company had a 35-40% repeat purchase rate, higher retention than most consumer apps, and it was right there in the blurb. Within every “bad” category, there are great companies. The antidote: examine each company’s remarkability on its own terms.

• The courage of re-betting. Rajaram considers Mike Moritz betting on Instacart his favorite venture capital bet, not because it was obvious, but because Moritz had just lost $370M on Webvan in the same space. “Think about the first-principles thinking needed and the courage needed to make that bet.”

• Figma: 500-1,000x. Rajaram held Figma for 13 years from angel to IPO. His angel portfolio’s highest multiple. A reminder that the truly great outcomes require patience that borders on unreasonable.

• Young founders are AI-maxing. Rajaram has invested in more dropouts in the past few months than in the previous 15 years combined. He doesn’t think dropping out is the right move for most, but the best of this generation are adopting AI tools in ways that make them genuinely more productive than experienced operators.

• What excites him most about the next decade: “The most ambitious entrepreneurs are finally tackling the hardest problems.” The answer to Peter Thiel’s “we wanted flying cars and got 140 characters” is finally coming into focus.