During discussions at HSBC New York Tech Week, one question surfaced repeatedly: as AI becomes easier to build, what actually makes an AI company difficult to replace?

Founders can win early customers with a compelling product. But as more companies gain access to the same models and tools, lasting advantage comes from something deeper. Across conversations with founders, investors, and enterprise leaders, three themes emerged: proprietary data, network effects, and strategic relevance.

The companies that win the next decade of enterprise AI may not be the ones with the best interface. They will be the ones that become hardest to replace.

According to Gartner, “worldwide spending on AI is forecast to total $2.59 trillion in 2026, a 47% increase year-over-year”. With that much capital flowing in, competition for enterprise contracts is fierce. Winning a deal is no longer the hard part. Keeping it is.

A wedge is a small, sharp entry point. It solves one painful task fast and proves value quickly. A wedge gets you in the door, but it won't keep you there.

Seema Amble, partner at Andreessen Horowitz, made clear this isn't new.

"Wedges have always been used in software."

What's changed is how many AI companies now rely on them. Voice agents, automation tools, and transcription software are common wedges today. They automate work in a tangible way and show return on investment quickly. That's exactly why enterprise buyers say yes.

But a wedge alone won't build a lasting company. Seema's point was direct: the goal is to use that wedge to become a system of record over time or build something far more durable. You can't replace an ERP, a loss system, or core infrastructure on day one. Trust has to be earned first.

Founder takeaway: Your first product doesn't need to be your moat. It needs to earn trust. The mistake many AI startups make is assuming their wedge is the business. The real goal is to use that initial win to become embedded in workflows, data, and decision-making over time.

Durable businesses rely on moats that don't depend on the interface. Seema pointed to a few classics that still hold weight:

• Proprietary data: Information no one else can easily access or replicate.
• Network effects: Value that grows as more users join, like the network of Dashers that powers DoorDash.
• Bridging digital and physical worlds: Connecting software to real-world operations that are difficult to copy.

As Seema noted, these advantages matter because they cannot be replicated as easily as a model or interface. A CIO may be able to recreate a basic AI application using various AI tools. Recreating years of proprietary data, operational integrations, and accumulated customer value is far more difficult.

The threat is real. As Seema put it, a CIO could theoretically use an AI tool and try to build the software themselves. Durability is what stops them. It's what makes rebuilding too difficult to bother with.

Vladimir Keil, CEO of Lio, learned this lesson the practical way. His company deploys AI agents to handle procurement end to end. The early wins came easily because, as he said, they started with the principle of really listening to customers' problems and caring about solving them.

But he also made a revealing mistake.

In one case, Lio delivered savings in the low seven figures for an enterprise client. The client’s initial spend was in the low five figures – an eye catching gap between value delivered and value captured.

Vladimir landed those first deals on the wedge alone. He didn't highlight the bigger strategic relevance. So he had to fight his way up to the C-suite himself and make the case in person. The savings got him the meeting. His persistence closed it. That executive later became an investor.

The lesson is one every founder should sit with:

"How can you have this ambiguity of selling a wedge, but also highlighting the strategic relevance?"

You need both at once. Lead with the quick, tangible win. But never let the buyer forget the millions you can save them over time.

This connects directly back to Seema's framework. The wedge opens the door. Strategic relevance, backed by proprietary data and deepening value, builds a business worth multiple millions a year.

As Vladimir noted, the real question isn't whether he can sell into enterprises. It's:

"How much value can I provide?"

When the answer becomes large enough, paying becomes a no-brainer.

Founder takeaway: Winning the first deal is only half the challenge. Enterprise buyers need to understand both the immediate ROI and the long-term strategic value of your solution. Lead with the quick win, but don't stop there.

Durability is about to become the dividing line between AI winners and non-winners. Here's how to prepare.

For founders

Use a wedge to get in the door, but begin building your moat immediately. Ask yourself what becomes more valuable every time a customer uses your product. If the answer is nothing, competitors may eventually catch up.

Capture more of the value you create. Saving a client millions while charging thousands isn't humility - it's a missed opportunity.

For those assessing companies

Look past the first contract and focus on what’s durable. Does this company own proprietary data? Is a network effect building? Can a CIO simply rebuild this in-house?

As Adam Milsom of HSBC Innovation Banking noted, these clients run notoriously long sales cycles, so conviction has to rest on lasting value - not short-term momentum.

For enterprise leaders

The build-versus-buy question is yours to own.

Seema noted that every founder now hears:

"Can I just use AI for this?" 

Sometimes building makes sense. Often it doesn't. The right vendor offers data, scale, workflow integration, and domain expertise that aren't easily recreated internally.

The conversations at HSBC New York Tech Week pointed to a broader shift in enterprise AI. The question is no longer whether AI creates value. It's whether that value compounds over time.

The companies that endure won't simply automate tasks. They'll build proprietary data, deeper integrations, network effects, and strategic relevance that customers can't easily recreate.

In a market where AI capabilities are becoming increasingly accessible, durability may be the most important competitive advantage of all.