Philippe Laffont, founder of Coatue Management and one of the most consequential technology investors of the last 25 years, sat down at Global Alts New York 2026 for a rare public conversation. Coatue manages nearly $100 billion. Laffont has been a technology investor since the firm launched with $50 million in 1999. His five major investment ideas over that period have compounded into one of the highest-returning long-duration technology funds in institutional investing

The AI capital stack is where Philippe Laffont is spending most of his time, and he has seen this movie before. Laffont framed his investment career in terms of five major technology waves: the PC era, PC networking, the internet, mobile internet and cloud computing, and now AI. “In technology you have to think about the big trends,” he told the room. “When you latch on to one, you peel the onion and there are 50 layers.” The implication for the current AI capital stack is that the visible layer — the model companies and the hyperscalers — is not necessarily where the compounding happens for the next decade.

How Laffont maps the AI capital stack

Coatue’s framework for the AI capital stack separates three layers. The infrastructure layer — compute, power, data centers — is where the current CapEx boom is most visible and where the near-term returns are concentrated in a small number of hyperscalers. The model layer — the foundation models and the companies building on top of them — carries more uncertainty about which players will hold margin. The application layer is where Laffont sees the most interesting forward opportunity, because AI-native applications can reach scale with far less capital than prior technology waves required.

“I’ve only had five good ideas in 27 years,” he said. “And today we have AI, which seems to be such a big trend. But last time I felt at the center of everything was 1999. So I’m a little bit worried.”

Where value accrues in the AI capital stack

Laffont pushed back on the consensus that the model companies will dominate the value chain. The cost to train and maintain models is enormous. The firms that fall behind the leading model cannot recover easily. And the application companies building on top of the models can switch providers if one model falls behind. The value in prior technology waves did not always accrue to the infrastructure layer. Often it accrued to the application and distribution layer that was most directly in front of the end user.

The Apple analogy came up explicitly. Laffont has a museum of every Apple product in his office. He noted that Apple’s most important layer in the mobile internet wave was not the device itself. It was the App Store, which captured a structural toll on every transaction across the wave.

What allocators should underwrite in the AI capital stack

The practical takeaway for allocators is about duration and layer selection. The infrastructure layer is investable now but is also the most crowded and most sensitive to CapEx digestion timing. The model layer carries binary risk around who remains at the frontier. The application layer has the longest duration and the widest dispersion, which means manager selection matters most there. Allocators mapping AI exposure across venture, growth, and public managers can use Allocator Intelligence on iConnections to surface specialists across all three layers of the AI capital stack.